Re: hibernate vs hibernate-ram ?
On Sat, Apr 12, 2008 at 03:40:39PM +0200, Hans-J. Ullrich wrote: > Second question: Is there a way on a multi processor system to prevent the > system to use both cores (yes, I know, can start as a single core machine) > and after start up using the second (and now empty cpu) for one process ? My guess would be to modify /etc/init.d/rcS from exec /etc/init.d/rc S to exec numactl --physcpubind=0 /etc/init.d/rc S (or better yet to modify the initrd) The affinity of all subprocesses is inherrited, thus should be bound to cpu0. This offcourse will only function on a numa arch. But I guess smp has similar controls. -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: hibernate vs hibernate-ram ?
Hans-J. Ullrich: > > May this might not the right list to ask, but can someone tell me the reason, > why the command "hibernate" and the command "hibernate-ram" behave > different ? From 'man hibernate': If the hibernate script is invoked with a name of the form hibernate-foo then it will use the configuration file /etc/hibernate/foo.conf instead of the default. > This is strange, because "hibernate-ram" is just a symlink to "hibernate". Every UNIX program knows the command line that invoked itself. That way, a program may behave differently depending on the name of the executable that has been launched. > Second question: Is there a way on a multi processor system to prevent the > system to use both cores (yes, I know, can start as a single core machine) > and after start up using the second (and now empty cpu) for one process ? As far as I know, you can only bind specific processed to a specific CPU. That doesn't prevent other processes from using the same CPU, but it really shouldn't matter performance-wise. If the process needs 100% of its CPU's time, it will get it. J. -- If I could have anything in the world it would have to be more money. [Agree] [Disagree] <http://www.slowlydownward.com/NODATA/data_enter2.html> signature.asc Description: Digital signature
Re: hibernate vs hibernate-ram ?
Hans-J. Ullrich 15:40 Sat 12 Apr Hello ! May this might not the right list to ask, but can someone tell me the reason, why the command "hibernate" and the command "hibernate-ram" behave different ? This is strange, because "hibernate-ram" is just a symlink to "hibernate". That app probably makes decisions based on argv[0]. Second question: Is there a way on a multi processor system to prevent the system to use both cores (yes, I know, can start as a single core machine) and after start up using the second (and now empty cpu) for one process ? My idea was, to start the host system on one cpu, and when it is up, to start a virtual machine (in my case "virtualbox") on the other cpu. Beats me. Check out schedtool for setting CPU affinity. Or wait for a more knowledgable replier. -- It has been said [by Anatole France], "it is not by amusing oneself that one learns," and, in reply: "it is *only* by amusing oneself that one can learn." -- Edward Kasner and James R. Newman
hibernate vs hibernate-ram ?
Hello ! May this might not the right list to ask, but can someone tell me the reason, why the command "hibernate" and the command "hibernate-ram" behave different ? This is strange, because "hibernate-ram" is just a symlink to "hibernate". Second question: Is there a way on a multi processor system to prevent the system to use both cores (yes, I know, can start as a single core machine) and after start up using the second (and now empty cpu) for one process ? My idea was, to start the host system on one cpu, and when it is up, to start a virtual machine (in my case "virtualbox") on the other cpu. Somebody told me, this should work. Can I believe him ??? Regards Hans -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: googleearth RAM Cache
Hi all, Concerning the google 'flight control' helicopter-steering mode which suddenly got 'disruptive'. After some tests, i'm still not clear how it works - but it's ok, now. What finally made a difference, was trimming the RAM cache down to 600M. Google default is about 400, for 2G Ram, and i pushed it up to 1024 in the beginning because i though that would enhance things. I think i may have to accept that 2G of slow 333Mhz DDR2memory (though running in dual mode) just is not fast enough for the google aircraft. At least for high speed low-level flight through the Grand Canyon, and investigatin glacial gorges and volvanos in the Sierra Nevada ;) The same applies for ADSL2000 with slow uplink -- and then the googleearth server may have some load at times, too. I always find it not easy to determine the exact RAM type just by the stickers on the bars. They say 'Corsair Value Select VS1GB667D2' which seem to be the same. On each 64M chip there are cryptic numbers like '64M8CFEG PS0900635'. So usually i trust BIOS wich in this case says 'DDRII667 (333MHz)'. I would be interested can i tell by dmidecode what kind of RAM is installed ? It looks like Handle 0x0007, DMI type 5, 24 bytes Memory Controller Information Error Detecting Method: 64-bit ECC Error Correcting Capabilities: None Supported Interleave: One-way Interleave Current Interleave: One-way Interleave Maximum Memory Module Size: 2048 MB Maximum Total Memory Size: 8192 MB Supported Speeds: 70 ns 60 ns Supported Memory Types: DIMM SDRAM Memory Module Voltage: 3.3 V Associated Memory Slots: 4 0x0008 0x0009 0x000A 0x000B Enabled Error Correcting Capabilities: None Handle 0x0008, DMI type 6, 12 bytes Memory Module Information Socket Designation: DIMM0 Bank Connections: 0 2 Current Speed: 5 ns Type: ECC DIMM Installed Size: 1024 MB (Double-bank Connection) Enabled Size: 1024 MB (Double-bank Connection) Error Status: OK Handle 0x0009, DMI type 6, 12 bytes Memory Module Information Socket Designation: DIMM1 Bank Connections: 0 2 Current Speed: 5 ns Type: ECC DIMM Installed Size: 1024 MB (Double-bank Connection) Enabled Size: 1024 MB (Double-bank Connection) Error Status: OK -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
ram/amd64
Is any experience for 2GB ram modules (ecc, registered) on the motherboard Tyan S2895 Thunder K8WE (where installed two dual opterons) ? The motherboard' manual (both revision 1.00, my own, and subsequent revision 1.01) is said to support up to 16 GB, though I notice now that there is a small arterisk for 2GB modules not validated at the time of printing. No surprise that they could not validate 16GB as they officially support Micrisoft only). At any event, should the experience be negative, any suggestion for another mainboard to support 2 dual opterons and 16GB ram? Preferably from where support to Linux is given (thus, no Tyan Europe) Thanks francesco pietra Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. http://farechase.yahoo.com/promo-generic-14795097 -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Tyam m-board/ram/bios/amd64
On Wed, Mar 14, 2007 at 01:16:25AM -0700, Francesco Pietra wrote: > ---1st question: is any comparison for amd64 between > Kingston KVR400D4R3A/2GB > and > Corsair CM75SDS2048RPL-3200/2GB ? > For the latter, easily available here and said by > Corsair compatible with the above mainboard, I was > unable to find volatage and W, which are 2.6V/11W for > the Kingston modules. I have a close approximation of the above in 1 machine (I think the above Kingston and the 1Gb 2700 version from Corsair), 4 corsair on 1 node and 2 kingston on the other. The kernel doesn't really care, the only difference I can see is that according to numastat the interleave_hit on the 2nd node is a little lower (3647 vs 3624). > ---2nd question: I had no troubles with the above > system and therefore never upgraded the bios. Should I > do that, where to get the appropriate bios upgrade, > from Phoenix or Tyan? Tyan, the only problem I had with the last update for my board was that it didn't fit on my usb thumbdrive in floppy emulation mode. But if it ain't broke, dont' fix it (IMHO). -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Tyam m-board/ram/bios/amd64
On a system running amd64 etch, based on Tyan Thunder K8WE S2895 # D 1629-100, revision 1.00, PhoenixBIOS version 1.03.2895 date 03/14/06 I am planning to replace the eight ram modules Kingston KVR400D4R3A/1GB with 2GB modules, same speed. Since I replaced in raid1 Maxtor HD with WD Raptor HD, I never had troubles with the above ram. However, 8GB are not enough for certain QM/MD calculations. I did a planning mistake - I need 4GB per node - have now to pay for, leaving unused the 8 1GB modules. ---1st question: is any comparison for amd64 between Kingston KVR400D4R3A/2GB and Corsair CM75SDS2048RPL-3200/2GB ? For the latter, easily available here and said by Corsair compatible with the above mainboard, I was unable to find volatage and W, which are 2.6V/11W for the Kingston modules. ---2nd question: I had no troubles with the above system and therefore never upgraded the bios. Should I do that, where to get the appropriate bios upgrade, from Phoenix or Tyan? Thanks francesco pietra Expecting? Get great news right away with email Auto-Check. Try the Yahoo! Mail Beta. http://advision.webevents.yahoo.com/mailbeta/newmail_tools.html -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Ernest jw ter Kuile <[EMAIL PROTECTED]> writes: > On Tuesday 07 November 2006 07:10, vitko wrote: >> > I only know of two (in my eyes) valid reasons to build your own kernel: >>> ... >> Two more reasons: >> ... >> > > One more: The debian stock kernel hangs on my system, while self built > kernels > don't. Since there are way too many drivers for not existing devices > complaining > just around that time, I never managed to find the reason. > > Ernest. Compare the 2 .config files and try one that is roughly halfway between the two. repeat till you find the option that causes the hang. MfG Goswin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tuesday 07 November 2006 07:10, vitko wrote: > > I only know of two (in my eyes) valid reasons to build your own kernel: >> ... > Two more reasons: > ... > One more: The debian stock kernel hangs on my system, while self built kernels don't. Since there are way too many drivers for not existing devices complaining just around that time, I never managed to find the reason. Ernest. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Mon, Nov 06, 2006 at 08:26:35PM +0100, Goswin von Brederlow wrote: > > There is almost never a reason to not run one of debian's prebuilt > > kernels. They work perfectly and optimally for probably 99% of users. > > I only know of two (in my eyes) valid reasons to build your own kernel: > > 1. you hate initrd since it too often doesn't work > 2. you need/want some extra patch There is at least a couple of other reasons ~:^) One is the Debian removes some things they find objectionable from the stock kernel sources. A really good example is USB serial dongle support. The Debian kernels have almost all of them removed, probably because they object to the firmware license agreement, or lack thereof. But they are in the stock kernel and so if you need to use one of them you have to build your own kernel, not from Debian kernel sources. Lately, I cannot get USB storage or hubs to work on the stock kernel, unless I do an install and select 'desktop environment' in tasksel. Yuck. > The "stock" kernels aren't slower and the disk space wasted for > unneeded (for you) modules is irelevant on any modern harddisk. The > times when you rebuild your kernel to get a slimmer one and save > memory are long gone. And on amd64 there is no change in optimizations Except maybe on embedded systems. Yes, there are opteron based embedded systems. OK, not a good example, sorry. This thread originated with a problem getting all the physical mem on a Tyan S2877. Now I have one of these, and I experienced the same problem. The problem is that the Bios doesn't say "enable IOMMU", it says "map memory hole?" and your options are 'DISABLED|software|hardware' and no explanation. Lame Bios help text, as always. And it shouldn't be disabled by default either. Sheez. My question is, what is the difference between the 'hardware' and 'software' choices? Is there a performance implication? Please don't reply with speculations, I can speculate myself just fine ~:^) Cheers, a -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, Nov 08, 2006 at 10:43:43AM -0500, Lennart Sorensen wrote:
> I also find initramfs handles LVM and software raid nicely, both of
> which I use, so I will stick with the modular kernels.
Debian initramfs tools used to use mdrun (maybe they still do) that is
horribly broken. There was a time when none of the initrd/initramfs
generators in etch could reliably boot a moderately complex RAID setup
that I've set up. ("Reliable" means continue to boot even if the disks
are physically reordered and/or moved to different controllers etc.)
I hope they are fixed now but I have no desiree to try it.
Gabor
--
-
MTA SZTAKI Computer and Automation Research Institute
Hungarian Academy of Sciences
-
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Freddie Cash sd73.bc.ca> writes: > > On Wednesday 08 November 2006 01:08 am, Gabor Gombas wrote: > > On Tue, Nov 07, 2006 at 12:27:46PM +, Jo Shields wrote: > > > 32-bit OSen have a theoretical limit of 4GiB addressable memory. > > > > PAE gives you 64 GiB physical memory on 32-bit processors that support > > it. > > Yes, but each process is still limited to 4 GB each. PAE just lets you > run multiple processes each with their own 4 GB of memory space. PAE doesn't have any baring on this. Each process has up to 4 GiB (really 3 on Linux by default, but theoretically 4), regardless of physical RAM. You could give and allocate pages for 4GiB for every process with only a few megs of physical memory, provided you had enough backing store (hard drive) to hold it all. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Jo Shields wrote: 32-bit OSen have a theoretical limit of 4GiB addressable memory. Wrong. IA-32 processors have had 36 address lines since PAE was added. That's what CONFIG_HIGHMEM64G does for IA-32 kernels, BTW. It turns on PAE support. However, there's an additional limitation as all expansion cards (e.g. graphics cards) need to have their memory regions mapped within that 4GiB area, lowering the overall amount of "system" memory you can address. This is also wrong. PCI devices with 64-bit BARs can obviously use a 64-bit address. However, most devices have a 32-bit BAR, requiring that the address be below the 4GiB boundary. Obviously then, physically memory has to remapped to accommodate[1]. 64-bit OSen don't have that limitation, No, they still do. The width of the virtual address space has nothing to do with physical addressing limitations. If the PCI device has a 32-bit BAR, then the mapping for its I/O space must occur below 4GiB. If RAM wants to be there, then it must be remapped, regardless of whether your OS is 32-bits or 64-bits or 1024-bits virtual address space. It's worth nothing that no AMD64 processor on the market has 64-bit physical lines or 64-bits of v.a.s. Current processors have 40 physical lines (some early EM64T have 36). The maximum limit is 52-bits. Virtual address space is fixed at 48-bits presently but can be extended to the 64-bit limit. Remember that limitations on physical addressing usually have physical reasons. So if a remapping option exists, it's likely for a damn good physical reason that the OS has nothing to do with whatsoever. Nor has any control over. Thanks, Adam -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Christian Hammers wrote: The is a kernel version called "linux-image-2.6-amd64-k8-smp" or so, try that. The "-generic" has maybe some options set that are not optimal for your system. Impossible, as there aren't any options on x86_64 that control how much physical memory you'll see (unlike IA-32). The glory of 48-bits of physical space. Thanks, Adam -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Emmanuel Fleury wrote: Couldn't it be kernel-space that would not be visible from user-space ? No, because that is virtual and this is physical. The limitations on the two are totally independent. Thanks, Adam -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wednesday 08 November 2006 01:08 am, Gabor Gombas wrote: > On Tue, Nov 07, 2006 at 12:27:46PM +, Jo Shields wrote: > > 32-bit OSen have a theoretical limit of 4GiB addressable memory. > > PAE gives you 64 GiB physical memory on 32-bit processors that support > it. Yes, but each process is still limited to 4 GB each. PAE just lets you run multiple processes each with their own 4 GB of memory space. If you have a process that needs more than 4 GB, PAE will not help you. -- Freddie Cash, LPIC-2 CCNT CCLP Network Support Technician School District 73 (250) 377-HELP [377-4357] [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, Nov 08, 2006 at 11:25:30AM -0500, Matthias Julius wrote: > Yes, but a 32-bit OS still can't use it. Or can it? A 32bit OS with PAE support can use the ram, but you are still limited to 32bit memory space per application. -- Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, 8 Nov 2006, Matthias Julius wrote: Yes, but a 32-bit OS still can't use it. Or can it? Yes the OS can, but single apps will usually still see <= 4Gbytes, unless they do a lot of (slow) black magic. Therefore, if you need a lot of memory overall because you need to run a huge number of processes, 32-bit+PAE may be ok; if you need a single big app to address a lot of memory transparently, you need a 64-bit OS. I do (big quantum chemistry calculations), your mileage may vary. Giacomo -- _ Giacomo Mulas <[EMAIL PROTECTED]> _ OSSERVATORIO ASTRONOMICO DI CAGLIARI Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 Tel. (UNICA): +39 070 675 4916 _ "When the storms are raging around you, stay right where you are" (Freddy Mercury) _ -- Il messaggio e' stato analizzato alla ricerca di virus o contenuti pericolosi da MailScanner, ed e' risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Gabor Gombas <[EMAIL PROTECTED]> writes: > On Tue, Nov 07, 2006 at 12:27:46PM +, Jo Shields wrote: > >> 32-bit OSen have a theoretical limit of 4GiB addressable memory. > > PAE gives you 64 GiB physical memory on 32-bit processors that support > it. Yes, but a 32-bit OS still can't use it. Or can it? Matthias -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, Nov 08, 2006 at 12:02:36AM +0100, Daniel Tryba wrote: > Thanks for the link, but: > -the machine having problems is not running a 64bit kernel > -I already installed kernel headers and tried pointing the vmware-config > to many places without success. /usr/src/linux-headers-x.y.z-w/include where x.y.z-w matches uname -r is the corract place to point vmware-config. I have yet to have a failure in many years of using vmware this way. -- Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 11:07:56PM +0100, Daniel Tryba wrote: > Well, I personally have been trying debian kernels for a couple of > weeks. A machine that has been running Debian/unstable with custom > kernels for the last couple of years and always has been very stable, > now is not detecting the soundcard every couple of reboots. And since I > can't (mainly to lazy to figure out how to) get VMWare Server to work > anymore (I can't seem to find the correct headers for the installed > kernel) I'm making the same conclusion I made a couple of years ago: > custom kernels are much more reliable and easier. If your kernel is linux-image-x.y.z-w then the header package is linux-headers-x.y.z-w. If your kernel is out of date and no longer in the archives, then you should upgrade to one that is if you want headers. > (also rebuilding the initrd a couple of times in a single update on a slow > machine is no fun to watch) I am not sure why that has started happening. It is a fairly recent change. I used to build my own kernels, but with 2.6 I just don't bother anymore. The debian provided ones are perfect and just work. With 2.6.0 early on I always had trouble making a kernel that would actually boot, while the premade kernels just worked, so I stopped bothering, and found out I wasn't missing any features. I also find initramfs handles LVM and software raid nicely, both of which I use, so I will stick with the modular kernels. -- Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 07:10:13PM -0500, Jim Crilly wrote: > > -I already installed kernel headers and tried pointing the vmware-config > > to many places without success. > > Well I don't know what to say about that, I've installed VMWare on numerous > machines with custom and Debian kernels and the only time I've run into > problems with the VMWare kernel modules was with custom kernels mostly > because of some 3rd party patches. It's a vmware issue in this case, 1.0.1-29996 will compile againt the latest deb/unstable kernel (2.6.18.something). -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, 8 Nov 2006, Hamish Moffatt wrote: Maybe there is nowhere to map the card's buffers into the bottom 4Gb of RAM when you have 4Gb of real RAM available. (This would seem to be a problem with all PCI cards plus AGP cards etc but maybe most systems or cards have a workaround?) Google has hits for this error from this driver eg http://lkml.org/lkml/2006/7/23/5 And there's a kernel patch in the thread which may be in recent kernels; boot with "iommu=soft swiotlb=force" to try the workaround. Thanks, tried, same as before (no go). Bye Giacomo -- _ Giacomo Mulas <[EMAIL PROTECTED]> _ OSSERVATORIO ASTRONOMICO DI CAGLIARI Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 Tel. (UNICA): +39 070 675 4916 _ "When the storms are raging around you, stay right where you are" (Freddy Mercury) _ -- Il messaggio e' stato analizzato alla ricerca di virus o contenuti pericolosi da MailScanner, ed e' risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, Nov 08, 2006 at 12:15:41PM +0100, Giacomo Mulas wrote: > On Wed, 8 Nov 2006, Gabor Gombas wrote: > > >Try using the skge driver instead of sk98lin. > > Thanks for the hint, I tried and indeed obtained a little improvement: it > does not hang as with sk98lin. However it still fails to initialise > properly, any trial to connect to anything via that NIC yields a "no route > to host" message. When booting, it issues this cryptic (for me, I do not > speak hexadecimal) message: > > skge :00:0a.0 PCI error cmd=0x117 status=0x22b0 > skge unable to clear error (so ignoring them) > > Nice try, but for the time being I still have no choice but to stick with > the memory hole and thus not having all 4 Gbytes of RAM available... Maybe there is nowhere to map the card's buffers into the bottom 4Gb of RAM when you have 4Gb of real RAM available. (This would seem to be a problem with all PCI cards plus AGP cards etc but maybe most systems or cards have a workaround?) Google has hits for this error from this driver eg http://lkml.org/lkml/2006/7/23/5 And there's a kernel patch in the thread which may be in recent kernels; boot with "iommu=soft swiotlb=force" to try the workaround. Hamish -- Hamish Moffatt VK3SB <[EMAIL PROTECTED]> <[EMAIL PROTECTED]> -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Wed, 8 Nov 2006, Gabor Gombas wrote: Try using the skge driver instead of sk98lin. Thanks for the hint, I tried and indeed obtained a little improvement: it does not hang as with sk98lin. However it still fails to initialise properly, any trial to connect to anything via that NIC yields a "no route to host" message. When booting, it issues this cryptic (for me, I do not speak hexadecimal) message: skge :00:0a.0 PCI error cmd=0x117 status=0x22b0 skge unable to clear error (so ignoring them) Nice try, but for the time being I still have no choice but to stick with the memory hole and thus not having all 4 Gbytes of RAM available... Bye Giacomo -- _ Giacomo Mulas <[EMAIL PROTECTED]> _ OSSERVATORIO ASTRONOMICO DI CAGLIARI Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 Tel. (UNICA): +39 070 675 4916 _ "When the storms are raging around you, stay right where you are" (Freddy Mercury) _ -- Il messaggio e' stato analizzato alla ricerca di virus o contenuti pericolosi da MailScanner, ed e' risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 03:18:36PM +0100, Giacomo Mulas wrote: > I wish this were always true. I tried to get rid of the memory hole, but > then my nice 64 bit Linux kernel would crash (or better get in an infinite > loop) upon initialising the Yukon Gigabit Ethernet NIC (i.e. immediately > after loading the sk98lin module). Try using the skge driver instead of sk98lin. Gabor -- - MTA SZTAKI Computer and Automation Research Institute Hungarian Academy of Sciences - -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 12:27:46PM +, Jo Shields wrote: > 32-bit OSen have a theoretical limit of 4GiB addressable memory. PAE gives you 64 GiB physical memory on 32-bit processors that support it. Gabor -- - MTA SZTAKI Computer and Automation Research Institute Hungarian Academy of Sciences - -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On 11/08/06 12:02:36AM +0100, Daniel Tryba wrote: > On Tue, Nov 07, 2006 at 05:38:50PM -0500, Jim Crilly wrote: > > I've been using the pre-built kernels since I got this AMD64 box without > > any hardware detection issues. If your soundcard disappears every couple of > > reboots I would be more suspicious of the hardware than the kernel. > > This is contradicted by the custom kernels not having this problem. > True, so if you're using the same version of the kernel there's either a patch upstream fixing something that's not in the Debian kernel or it could be an odd timing issue if you're compiling the sound driver in statically but that seems unlikely. > > As for VMWare, the Ubuntu wiki listed the prereqs that I needed and after > > they're installed you can install VMWare just like normal. > > > > https://help.ubuntu.com/community/VMware_Guide%3a_Installing_VMware_Server_on_Ubuntu_6%2e06_LTS_amd64 > > Thanks for the link, but: > -the machine having problems is not running a 64bit kernel > -I already installed kernel headers and tried pointing the vmware-config > to many places without success. > > Again these troubles aren't present when running custom kernels. 2 out of > the 3 machines I'm tried to run deb. kernels on failed. The third on is > running just fine. > Well I don't know what to say about that, I've installed VMWare on numerous machines with custom and Debian kernels and the only time I've run into problems with the VMWare kernel modules was with custom kernels mostly because of some 3rd party patches. Jim. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 05:38:50PM -0500, Jim Crilly wrote: > I've been using the pre-built kernels since I got this AMD64 box without > any hardware detection issues. If your soundcard disappears every couple of > reboots I would be more suspicious of the hardware than the kernel. This is contradicted by the custom kernels not having this problem. > As for VMWare, the Ubuntu wiki listed the prereqs that I needed and after > they're installed you can install VMWare just like normal. > > https://help.ubuntu.com/community/VMware_Guide%3a_Installing_VMware_Server_on_Ubuntu_6%2e06_LTS_amd64 Thanks for the link, but: -the machine having problems is not running a 64bit kernel -I already installed kernel headers and tried pointing the vmware-config to many places without success. Again these troubles aren't present when running custom kernels. 2 out of the 3 machines I'm tried to run deb. kernels on failed. The third on is running just fine. -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On 11/07/06 11:07:56PM +0100, Daniel Tryba wrote: > Well, I personally have been trying debian kernels for a couple of > weeks. A machine that has been running Debian/unstable with custom > kernels for the last couple of years and always has been very stable, > now is not detecting the soundcard every couple of reboots. And since I > can't (mainly to lazy to figure out how to) get VMWare Server to work > anymore (I can't seem to find the correct headers for the installed > kernel) I'm making the same conclusion I made a couple of years ago: > custom kernels are much more reliable and easier. I've been using the pre-built kernels since I got this AMD64 box without any hardware detection issues. If your soundcard disappears every couple of reboots I would be more suspicious of the hardware than the kernel. As for VMWare, the Ubuntu wiki listed the prereqs that I needed and after they're installed you can install VMWare just like normal. https://help.ubuntu.com/community/VMware_Guide%3a_Installing_VMware_Server_on_Ubuntu_6%2e06_LTS_amd64 Jim. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 04:39:00PM +0100, Goswin von Brederlow wrote: > My system locks up when I run "ifconfig eth1 up" on the sk98lin module > but I only have 1Gb of ram. Maybe I should play with the bios > settings. I'd guess that it could be related to IOMMU settings, but that does only make sense with 4Gb or more it seems. But I have seen lockups of the sk98 driver even on my 1Gb 32bit kernel running EMT64 machine (only with <2.6.18 kernels, .18 has been running stable so far). -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, Nov 07, 2006 at 04:34:18PM +0100, Goswin von Brederlow wrote: > > 1) I don't like seeing errors / misdetected hardware in kernel messages. > > Far too many times I've seen loading drivers for hardware that was never > > there. As for kernel, I want it to be clean and do its work for the > > exact hardware I use, > > Haven't seen that with the modularized kernels and discover. Matching > the pci IDs to hardware works well. Well, I personally have been trying debian kernels for a couple of weeks. A machine that has been running Debian/unstable with custom kernels for the last couple of years and always has been very stable, now is not detecting the soundcard every couple of reboots. And since I can't (mainly to lazy to figure out how to) get VMWare Server to work anymore (I can't seem to find the correct headers for the installed kernel) I'm making the same conclusion I made a couple of years ago: custom kernels are much more reliable and easier. (also rebuilding the initrd a couple of times in a single update on a slow machine is no fun to watch) -- When you do things right, people won't be sure you've done anything at all. Daniel Tryba -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tuesday 07 November 2006 10:32 am, John Hannfield wrote: > Actually, Joost or others, do you know what difference the memCLK BIOS > setting has? > > With "MTRR" set to discreet, and "Memhole Mapping" set to software > control my system boots fine, and I can see 4GB RAM. However, I then > install xen then it won't boot. It fails activating the second core of > the CPU. If I change memCLK > from the default (200Mhz) to 100Mhz it boots fine, but then it only > shows 3GB RAM. > Is this memCLK specific to the Opteron or my RAM or what? > > Is there a guide to all these Hammer configurcation paramemters > anywhere? Use the 2.6.18-2-xen-amd64 kernel from Unstable (along with xen 3.0.3), and everything will work correctly with those BIOS settings. Haven't tracked down exactly what the issue is, but 2.6.16 and 2.6.17 Xen kernels on my AMD64 systems won't boot. -- Freddie Cash, LPIC-2 CCNT CCLP Network Support Technician School District 73 (250) 377-HELP [377-4357] [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Actually, Joost or others, do you know what difference the memCLK BIOS setting has?With "MTRR" set to discreet, and "Memhole Mapping" set to software controlmy system boots fine, and I can see 4GB RAM. However, I then install xen then it won't boot. It fails activating the second core of the CPU. If I change memCLKfrom the default (200Mhz) to 100Mhz it boots fine, but then it only shows 3GB RAM.Is this memCLK specific to the Opteron or my RAM or what? Is there a guide to all these Hammer configurcation paramemters anywhere?John
Re: AMD64-generic doesn't see all 4GB RAM?
Giacomo Mulas <[EMAIL PROTECTED]> writes: > On Tue, 7 Nov 2006, Jo Shields wrote: > >> address. 64-bit OSen don't have that limitation, but many BIOSes assume >> that there's only one OS on the market (32-bit XP) so keep the memory >> configuration with a hole at 4GiB. > > I wish this were always true. I tried to get rid of the memory hole, but > then my nice 64 bit Linux kernel would crash (or better get in an infinite > loop) upon initialising the Yukon Gigabit Ethernet NIC (i.e. immediately > after loading the sk98lin module). Actually, to be more specific, it locks > up not when loading the driver, but the first time anything attempts to use > the NIC, sending any traffic through it. > The same kernel boots and works ok with the memory hole, therefore I decided > I preferred to have a working computer with ~300Mb less than a locked up > computer with 4Gb... > > Bye > Giacomo My system locks up when I run "ifconfig eth1 up" on the sk98lin module but I only have 1Gb of ram. Maybe I should play with the bios settings. MfG Goswin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
vitko <[EMAIL PROTECTED]> writes: >> I only know of two (in my eyes) valid reasons to build your own kernel: >> >> 1. you hate initrd since it too often doesn't work >> 2. you need/want some extra patch > > Two more reasons: > > 1) I don't like seeing errors / misdetected hardware in kernel messages. > Far too many times I've seen loading drivers for hardware that was never > there. As for kernel, I want it to be clean and do its work for the > exact hardware I use, Haven't seen that with the modularized kernels and discover. Matching the pci IDs to hardware works well. > 2) The may be some hardware I don't want / need supported, typically > multimedia devices on servers. The less code runs, the smaller > possibility something can go wrong in unexpected places. You can blacklist modules. > Just my USB 0.02. > > Vit MfG Goswin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Tue, 7 Nov 2006, Jo Shields wrote: address. 64-bit OSen don't have that limitation, but many BIOSes assume that there's only one OS on the market (32-bit XP) so keep the memory configuration with a hole at 4GiB. I wish this were always true. I tried to get rid of the memory hole, but then my nice 64 bit Linux kernel would crash (or better get in an infinite loop) upon initialising the Yukon Gigabit Ethernet NIC (i.e. immediately after loading the sk98lin module). Actually, to be more specific, it locks up not when loading the driver, but the first time anything attempts to use the NIC, sending any traffic through it. The same kernel boots and works ok with the memory hole, therefore I decided I preferred to have a working computer with ~300Mb less than a locked up computer with 4Gb... Bye Giacomo -- _ Giacomo Mulas <[EMAIL PROTECTED]> _ OSSERVATORIO ASTRONOMICO DI CAGLIARI Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 Tel. (UNICA): +39 070 675 4916 _ "When the storms are raging around you, stay right where you are" (Freddy Mercury) _ -- Il messaggio e' stato analizzato alla ricerca di virus o contenuti pericolosi da MailScanner, ed e' risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Mon, 2006-11-06 at 17:54 +0100, A J Stiles wrote: > On Monday 06 November 2006 12:08, John Hannfield wrote: > > I have a dual core Opteron 265 running in a Tyan K8WE (S2877) motherboard > > with > > 4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but using > > a fresh install of testing/unstable from the AMD64 etch net-installer, I > > can still only > > see 3.2 GB out of the 4GB RAM... > > There's your problem; you're still running the installer kernel. The > installer kernel is only supposed to work well enough and for long enough for > you to build yourself a new one. Install kernel-package, libncurses5-dev > (menuconfig needs it); then you can just get sources from kernel.org, and > compile them into a .deb package to install with dpkg -i. > > Note: unless you're *very* lucky, you *will* at some point turn off something > you should have left on and your new kernel won't boot. Save all your config > files, have a bootable CD handy, and learn how to use it to alter your LILO > or GRUB configuration to boot the installer kernel. > > If you're still running a "stock" kernel, you're only using about half the > power of Linux . Bollocks. There hasn't been such a thing as an "installer kernel" since Woody (with "idepci" and "bf" minimal kernels). Debian has a selection of kernels with pretty much every available option compiled as a module - meaning they don't waste system resources if you don't have the hardware. The only benefit to compiling your own kernel based on one of those with bits removed versus just using one of them is the shorter compile time - and since the Debian kernels come precompiled, that's only an issue if you're the DD in charge of compiling the damned things. The error the OP reports is a BIOS issue. You could compile a thousand replacement kernels - it wouldn't help. For those with a compilation fetish, try Gentoo or Linux From Scratch. 32-bit OSen have a theoretical limit of 4GiB addressable memory. However, there's an additional limitation as all expansion cards (e.g. graphics cards) need to have their memory regions mapped within that 4GiB area, lowering the overall amount of "system" memory you can address. 64-bit OSen don't have that limitation, but many BIOSes assume that there's only one OS on the market (32-bit XP) so keep the memory configuration with a hole at 4GiB. Reconfigure the BIOS to get rid of the hole at 4GiB. The setting you want will probably be labelled as "MTRR" or "Memory Hole". -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Lennart Sorensen wrote: > On Mon, Nov 06, 2006 at 05:54:00PM +0100, A J Stiles wrote: >> There's your problem; you're still running the installer kernel. The >> installer kernel is only supposed to work well enough and for long enough >> for >> you to build yourself a new one. Install kernel-package, libncurses5-dev >> (menuconfig needs it); then you can just get sources from kernel.org, and >> compile them into a .deb package to install with dpkg -i. >> >> Note: unless you're *very* lucky, you *will* at some point turn off >> something >> you should have left on and your new kernel won't boot. Save all your >> config >> files, have a bootable CD handy, and learn how to use it to alter your LILO >> or GRUB configuration to boot the installer kernel. >> >> If you're still running a "stock" kernel, you're only using about half the >> power of Linux . > > There is almost never a reason to not run one of debian's prebuilt > kernels. They work perfectly and optimally for probably 99% of users. > > The 3.2GB problem has to do with memory remapping which is a BIOS > problem. > > The etch installer is quite good at installing the optimal kernel for > the system. I'm running a stock kernel on a Sun Fire V40z (4 x Opteron 852) with 16 gigs of RAM - the kernel sees all 16 gigs just fine. Regards, Ozz. signature.asc Description: OpenPGP digital signature
Re: AMD64-generic doesn't see all 4GB RAM?
Thanks Joost + JanIt was the Tyan BIOS memory mapping problem as you said.I had already turned the MTRR to discreet mapping, but it did notoccur to me to change the "Software Memory Hole" to software controlled, rather than disabled. Once I did that, the system happily reports the 4GBof memory.Thankyou for your help, it is much appreciated!-- John
Re: AMD64-generic doesn't see all 4GB RAM?
> I only know of two (in my eyes) valid reasons to build your own kernel: > > 1. you hate initrd since it too often doesn't work > 2. you need/want some extra patch Two more reasons: 1) I don't like seeing errors / misdetected hardware in kernel messages. Far too many times I've seen loading drivers for hardware that was never there. As for kernel, I want it to be clean and do its work for the exact hardware I use, 2) The may be some hardware I don't want / need supported, typically multimedia devices on servers. The less code runs, the smaller possibility something can go wrong in unexpected places. Just my USB 0.02. Vit -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
Lennart Sorensen <[EMAIL PROTECTED]> writes: > On Mon, Nov 06, 2006 at 05:54:00PM +0100, A J Stiles wrote: >> There's your problem; you're still running the installer kernel. The >> installer kernel is only supposed to work well enough and for long enough >> for >> you to build yourself a new one. Install kernel-package, libncurses5-dev >> (menuconfig needs it); then you can just get sources from kernel.org, and >> compile them into a .deb package to install with dpkg -i. You can also just apt-get install linux-tree-2.6.x to get the source. >> Note: unless you're *very* lucky, you *will* at some point turn off >> something >> you should have left on and your new kernel won't boot. Save all your >> config >> files, have a bootable CD handy, and learn how to use it to alter your LILO >> or GRUB configuration to boot the installer kernel. Best to start off with the debian config and then remove stuff you definetly don't need. >> If you're still running a "stock" kernel, you're only using about half the >> power of Linux . > > There is almost never a reason to not run one of debian's prebuilt > kernels. They work perfectly and optimally for probably 99% of users. I only know of two (in my eyes) valid reasons to build your own kernel: 1. you hate initrd since it too often doesn't work 2. you need/want some extra patch The "stock" kernels aren't slower and the disk space wasted for unneeded (for you) modules is irelevant on any modern harddisk. The times when you rebuild your kernel to get a slimmer one and save memory are long gone. And on amd64 there is no change in optimizations like on i386 with the 486, k6, k7, Pentium IV, ... settings. > The 3.2GB problem has to do with memory remapping which is a BIOS > problem. > > The etch installer is quite good at installing the optimal kernel for > the system. Except the netinst iso which only has one kernel on the cd. You have to fetch one from the mirrors there. But you can say it still picks the optimal one, the only one present. :) MfG Goswin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Mon, Nov 06, 2006 at 05:54:00PM +0100, A J Stiles wrote: > There's your problem; you're still running the installer kernel. The > installer kernel is only supposed to work well enough and for long enough for > you to build yourself a new one. Install kernel-package, libncurses5-dev > (menuconfig needs it); then you can just get sources from kernel.org, and > compile them into a .deb package to install with dpkg -i. > > Note: unless you're *very* lucky, you *will* at some point turn off something > you should have left on and your new kernel won't boot. Save all your config > files, have a bootable CD handy, and learn how to use it to alter your LILO > or GRUB configuration to boot the installer kernel. > > If you're still running a "stock" kernel, you're only using about half the > power of Linux . There is almost never a reason to not run one of debian's prebuilt kernels. They work perfectly and optimally for probably 99% of users. The 3.2GB problem has to do with memory remapping which is a BIOS problem. The etch installer is quite good at installing the optimal kernel for the system. -- Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Monday 06 November 2006 12:08, John Hannfield wrote: > I have a dual core Opteron 265 running in a Tyan K8WE (S2877) motherboard > with > 4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but using > a fresh install of testing/unstable from the AMD64 etch net-installer, I > can still only > see 3.2 GB out of the 4GB RAM... There's your problem; you're still running the installer kernel. The installer kernel is only supposed to work well enough and for long enough for you to build yourself a new one. Install kernel-package, libncurses5-dev (menuconfig needs it); then you can just get sources from kernel.org, and compile them into a .deb package to install with dpkg -i. Note: unless you're *very* lucky, you *will* at some point turn off something you should have left on and your new kernel won't boot. Save all your config files, have a bootable CD handy, and learn how to use it to alter your LILO or GRUB configuration to boot the installer kernel. If you're still running a "stock" kernel, you're only using about half the power of Linux . -- AJS delta echo bravo six four at earthshod dot co dot uk -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On Mon, 2006-11-06 at 12:08 +, John Hannfield wrote: > > I have a dual core Opteron 265 running in a Tyan K8WE (S2877) > motherboard with > 4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but > using > a fresh install of testing/unstable from the AMD64 etch net-installer, > I can still only > see 3.2 GB out of the 4GB RAM... > > I am using the amd64-generic kernel that net-install gave me. > Do I need to be using a different kernel to get the full 4GB RAM? No, you have to set the right BIOS options if they are available on your motherboard. On my motherboard (Tyan S2885) I had to set the "MTRR mapping" from continuous to discrete and "Software Memory Hole" from disabled to enabled. -- Groeten, Joost Kraaijeveld Askesis B.V. Molukkenstraat 14 6524NB Nijmegen tel: 024-3888063 / 06-51855277 fax: 024-3608416 web: www.askesis.nl -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
On 2006-11-06 John Hannfield wrote: > I have a dual core Opteron 265 running in a Tyan K8WE (S2877) motherboard > with > 4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but using > a fresh install of testing/unstable from the AMD64 etch net-installer, I can > still only > see 3.2 GB out of the 4GB RAM... The is a kernel version called "linux-image-2.6-amd64-k8-smp" or so, try that. The "-generic" has maybe some options set that are not optimal for your system. bye, -christian- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: AMD64-generic doesn't see all 4GB RAM?
John Hannfield wrote: > > I have a dual core Opteron 265 running in a Tyan K8WE (S2877) > motherboard with > 4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but using > a fresh install of testing/unstable from the AMD64 etch net-installer, I > can still only > see 3.2 GB out of the 4GB RAM... Couldn't it be kernel-space that would not be visible from user-space ? Regards -- Emmanuel Fleury | Office: 261 Associate Professor, | Phone: +33 (0)5 40 00 69 34 LaBRI, Domaine Universitaire | Fax: +33 (0)5 40 00 66 69 351, Cours de la Libération | email: [EMAIL PROTECTED] 33405 Talence Cedex, France | URL: http://www.labri.fr/~fleury -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
AMD64-generic doesn't see all 4GB RAM?
I have a dual core Opteron 265 running in a Tyan K8WE (S2877) motherboard with4 x 1GB RAM DDR modules. The BIOS displays the RAM as 4094 MB, but usinga fresh install of testing/unstable from the AMD64 etch net-installer, I can still only see 3.2 GB out of the 4GB RAM...I am using the amd64-generic kernel that net-install gave me.Do I need to be using a different kernel to get the full 4GB RAM?# uname -a Linux debian 2.6.16-2-amd64-generic #1 Sun Jul 16 01:12:23 CEST 2006 x86_64 GNU/Linux# cat /proc/meminfo MemTotal: 3350660 kBMemFree: 3142864 kBBuffers: 7500 kBCached: 153792 kB SwapCached: 0 kBActive: 144088 kBInactive: 24644 kBHighTotal: 0 kBHighFree: 0 kBLowTotal: 3350660 kBLowFree: 3142864 kBSwapTotal: 23101376 kB SwapFree: 23101376 kBDirty: 228 kBWriteback: 0 kBMapped: 11700 kBSlab: 17896 kBCommitLimit: 24776704 kBCommitted_AS: 19272 kBPageTables: 960 kB VmallocTotal: 34359738367 kBVmallocUsed: 11528 kBVmallocChunk: 34359726839 kB# cat /proc/cpuinfoprocessor : 0vendor_id : AuthenticAMDcpu family : 15model : 33 model name : Dual Core AMD Opteron(tm) Processor 265stepping : 2cpu MHz : 1808.354cache size : 1024 KBfpu : yesfpu_exception : yescpuid level : 1wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow pni lahf_lm cmp_legacybogomips : 3620.65 TLB size : 1024 4K pagesclflush size : 64cache_alignment : 64address sizes : 40 bits physical, 48 bits virtualpower management: ts fid vid ttp-- John
solved - Re: 6GB RAM on a Tyan S2875 Tiger K8W - 500 MB missing
Max A. wrote: hi max, hi folks! > Before playing with Memory Hole and MTRR settings in your BIOS setup, > I strongly suggest to update your BIOS to the latest version. > See http://lists.debian.org/debian-amd64/2006/06/msg00344.html just to let you and google know: it worked out. i upgraded the bios to version 3.03 (took me 10 floppy disks to find a working one) and activated these settings: - MTRR mode discrete - enable software memory hole - disable hardware memory hole and voila, got me 6144MB of RAM: Memory: 6158992k/8388608k available (2669k kernel code, 131712k reserved, 1001k data, 196k init) disabling the hardware memory hole is mandatory, otherwise the kernel wont boot, probably because it cannot access memory-mapped PCI I/O. thanks again, Raimund -- Die Lösung für effizientes Kundenbeziehungsmanagement. Jetzt informieren: http://www.universal-messenger.de Pinuts media+science GmbH http://www.pinuts.de Dipl.-Inform. Raimund Jacob [EMAIL PROTECTED] Krausenstr. 9-10 voice : +49 30 59 00 90 322 10117 Berlin fax : +49 30 59 00 90 390 Germany -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: 6GB RAM on a Tyan S2875 Tiger K8W - 500 MB missing
Raimund, Before playing with Memory Hole and MTRR settings in your BIOS setup, I strongly suggest to update your BIOS to the latest version. See http://lists.debian.org/debian-amd64/2006/06/msg00344.html Max -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: 6GB RAM on a Tyan S2875 Tiger K8W - 500 MB missing
Lennart Sorensen wrote: Hi Len, * ! >>question is: where is my memory ?! it's too much to be a miscalculation >>of some form (like the HDD manufacturers do it :) could anyone make >>sense of my BIOS-provided physical RAM map if i posted it? what am i >>missing? > > > There is usually a memory hole for BIOS and PCI access, at 3.5 to 4GB. > So unless your bios supports memory holes/memory remapping (most do with > the right bios setting), then you loose that ram. Check your bios for > some settings related to memory holes or something similar. > > According to what I remember, a setting for 'memory hole' in the bios > should be set to 'software' on tyan boards (assuming your bios is new > enough to have the option). Back when I saw this, they were talking > about having to use beta bios releases to get the option, but that was a > while ago (as in last fall). ah, that sounds reasonable. i've seen this option but the short description didnt make sense to me. perhaps my BIOS also isnt new enough. i will check all this once i can reboot the machine again (which is, when everyone else left the office and i'm still around). thanks a bunch, Raimund -- Die Lösung für effizientes Kundenbeziehungsmanagement. Jetzt informieren: http://www.universal-messenger.de Pinuts media+science GmbH http://www.pinuts.de Dipl.-Inform. Raimund Jacob [EMAIL PROTECTED] Krausenstr. 9-10 voice : +49 30 59 00 90 322 10117 Berlin fax : +49 30 59 00 90 390 Germany -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: 6GB RAM on a Tyan S2875 Tiger K8W - 500 MB missing
On Fri, Jul 07, 2006 at 10:25:26AM +0200, Raimund Jacob wrote: > i upgraded the RAM of my workstation yesterday and have some problems > understanding what i'm seeing. > > here is what i did: started with 4 GB of RAM, made up from 4x1GB > registered non-ECC modules. all working well. ripped apart a new server > that came with 2x2GB registered ECC modules (also worked well, so > hardware defect is out of the question). the plan was to put those 2x2 > in my workstation, leaving 2x1 in place. > > that's what i did and i saw only 4 GB, but that was kinda logical > because the 2x1 are non-ECC and i had to tell the BIOS about that. after > disabling ECC alltogether the machines boots with the BIOS reporting > 57xx MB of memory. my kernel says: > Memory: 5746100k/7438336k available (2638k kernel code, 118684k > reserved, 996k data, 196k init) > > here is what i dont get: 6x1024MB are 6144MB but BIOS and kernel report > only 5611MB - so where are my 533MB ?! > > i first though this might be an due to the way i plugged the modules > into the DIMM slots and tried some other patterns. it turns out that the > ECC modules alone only work when put into DIMM1/DIMM2 or DIMM1/DIMM3 - > in combination with the non-ECC modules it only works with the 2x2 in > DIMM1/DIMM2 and the 2x1 in DIMM3/DIMM4. again, ECC is disabled in the > BIOS completely - otherwise it wouldnt use the non-ECC modules at all. > > so i'm thinking if this is some kind of artefact of some "memory hole" > i'm not aware of. > > also, the manual of the board (Tyan Tiger K8W S2875) contains a little > chart that supposedly shows how 64bit (non-interleaved) and 128bit > (interleaved) memory configurations work. but with all i know about > computers i cannot interpret nor understand it:) > > question is: where is my memory ?! it's too much to be a miscalculation > of some form (like the HDD manufacturers do it :) could anyone make > sense of my BIOS-provided physical RAM map if i posted it? what am i > missing? There is usually a memory hole for BIOS and PCI access, at 3.5 to 4GB. So unless your bios supports memory holes/memory remapping (most do with the right bios setting), then you loose that ram. Check your bios for some settings related to memory holes or something similar. According to what I remember, a setting for 'memory hole' in the bios should be set to 'software' on tyan boards (assuming your bios is new enough to have the option). Back when I saw this, they were talking about having to use beta bios releases to get the option, but that was a while ago (as in last fall). Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
6GB RAM on a Tyan S2875 Tiger K8W - 500 MB missing
Hello *, i upgraded the RAM of my workstation yesterday and have some problems understanding what i'm seeing. here is what i did: started with 4 GB of RAM, made up from 4x1GB registered non-ECC modules. all working well. ripped apart a new server that came with 2x2GB registered ECC modules (also worked well, so hardware defect is out of the question). the plan was to put those 2x2 in my workstation, leaving 2x1 in place. that's what i did and i saw only 4 GB, but that was kinda logical because the 2x1 are non-ECC and i had to tell the BIOS about that. after disabling ECC alltogether the machines boots with the BIOS reporting 57xx MB of memory. my kernel says: Memory: 5746100k/7438336k available (2638k kernel code, 118684k reserved, 996k data, 196k init) here is what i dont get: 6x1024MB are 6144MB but BIOS and kernel report only 5611MB - so where are my 533MB ?! i first though this might be an due to the way i plugged the modules into the DIMM slots and tried some other patterns. it turns out that the ECC modules alone only work when put into DIMM1/DIMM2 or DIMM1/DIMM3 - in combination with the non-ECC modules it only works with the 2x2 in DIMM1/DIMM2 and the 2x1 in DIMM3/DIMM4. again, ECC is disabled in the BIOS completely - otherwise it wouldnt use the non-ECC modules at all. so i'm thinking if this is some kind of artefact of some "memory hole" i'm not aware of. also, the manual of the board (Tyan Tiger K8W S2875) contains a little chart that supposedly shows how 64bit (non-interleaved) and 128bit (interleaved) memory configurations work. but with all i know about computers i cannot interpret nor understand it:) question is: where is my memory ?! it's too much to be a miscalculation of some form (like the HDD manufacturers do it :) could anyone make sense of my BIOS-provided physical RAM map if i posted it? what am i missing? thanks for any hint, Raimund -- Die Lösung für effizientes Kundenbeziehungsmanagement. Jetzt informieren: http://www.universal-messenger.de Pinuts media+science GmbH http://www.pinuts.de Dipl.-Inform. Raimund Jacob [EMAIL PROTECTED] Krausenstr. 9-10 voice : +49 30 59 00 90 322 10117 Berlin fax : +49 30 59 00 90 390 Germany -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
> You MUST however have at least one stick of ram installed per cpu in > that cpu's memory area. If you don't put ram in A3 or B3, then it will > not see CPU2 at all. Really? My Tyan board works fine with RAM on only one CPU. The only restriction is that you must have at least one dimm installed in the system (duh!). The converse is not true. ie. you can only install memory in A3/B3 if you have two CPUs installed. Paul -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
Quoting Lennart Sorensen: > Well I was talking to someone on irc last week with that particular asus > board, and he had two dual core cpus, and 2 sticks of ram installed in > A1+B1, and it refused to see the second cpu. I suggested he move B1 to > A3, and when he did, the second cpu showed up. > > Maybe it isn't supposed to be that way, but it certainly acted that way. I'll try to duplicate that. I have 4x1GB sticks and I'll play around with some different configs. I'm not sure if the BIOS has any tools to show memory capacity per channel or not... is there any good way in linux to do this? I have started a wiki for this board at http://tastytronic.net/k8ndlis/ to try and collect data and real-world information on it and I'll put the result of this conversation there. pedro -- Peter A. H. Peterson, technician and musician. ---=[ http://tastytronic.net/~pedro/ ]=--- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
On Mon, Apr 10, 2006 at 04:01:28PM +0100, Paul Brook wrote: > > You MUST however have at least one stick of ram installed per cpu in > > that cpu's memory area. If you don't put ram in A3 or B3, then it will > > not see CPU2 at all. > > Really? My Tyan board works fine with RAM on only one CPU. The only > restriction is that you must have at least one dimm installed in the system > (duh!). > > The converse is not true. ie. you can only install memory in A3/B3 if you > have > two CPUs installed. Well I was talking to someone on irc last week with that particular asus board, and he had two dual core cpus, and 2 sticks of ram installed in A1+B1, and it refused to see the second cpu. I suggested he move B1 to A3, and when he did, the second cpu showed up. Maybe it isn't supposed to be that way, but it certainly acted that way. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
On Sun, Apr 09, 2006 at 04:10:26PM -0700, Peter A. H. Peterson wrote: > Quoting Steffen Grunewald: > > Wouldn't this mean that, in a dual CPU system, one should better use > > 4 mem modules? > > I just got an ASUS K8N-DL (because it is NUMA but fits into many ATX > cases). It is a strange board in that CPU2 has two slots and CPU1 has > 4 slots. The slots are also named strangely. Here's a diagram: > > | | [ ] > | | | | | | > | | [ ] | | | | > B A | | | | > 3 3 A A B B > 1 2 1 2 > > Two banks on the left, slot B3 and slot A3. Four banks on the right, > A1, A2, B1, and B2. > > A table of says: > > For CPU1 Sockets > Channel A A1 and A2 > Channel B B1 and B2 > For CPU2 > Channel A A3 > Channel B B3 > > This makes you think that if you have 4GB (4x1GB) you can put two > sticks in A3/B3 and two sticks in A1/A2. > > But then you turn the page and you find this table: > > "You may install 256M, 512M, 1G, 2G or 4G registered ECC DDR DIMMs > into the DIMM sockets using the memory configurations in this > section. > > * For dual-channel configuration, the total size of memory modules > installed per channel must be the same for better performance. > Single CPU: >A1 + A2 = B1 + B2 > Dual CPU: >A1 + A2 = B1 + B2 = A3 + B3 > * When using one DDR DIMM module, install into A1 slot only. > * When using two DDR DIMM modules, install into A1 and A2 slots only. > [snip]" > > THe whole "A1 + A2 = B1 + B2 = A3 + B3" thing doesn't make sense > because it makes it sound like if you want dual-channel memory on both > CPUs you need RAM in all 6 slots, and each needs to be equal, which > seems extremely limiting. > > I put two in A1/A2 and two in A3/B3... but then that means that > channel A has 3 GB and Channel B has 1GB. > > Does anyone know what they're talking about, or has this board and has > done some testing? > > I'm baffled. It is VERY badly written. It seems that the way this board works is this: One cpu has two channels with two slots each. The other cpu has two channels with one slot each. To get the best system performance you want it to be balanced with equal amounts of ram on each channel on each cpu. Some ways to do this would be: A1, A2, B1, B2 = 512M each A3, B3 = 1GB each. Total of 4GB ram. Or you could do: A1, B1, A3, B3 = 1GB ram each. Total of 4GB ram. It does not says that you MUST do it that way, just that you get optimal performance that way. You could also do: A1, A2, A3, B1, B2, B3 = 1GB each. Total of 6GB, but 2/3 of the ram is on cpu1 and 1/3 on cpu2. So in that case if you are lucky the OS will run ram hungry programs on cpu1 and less ram hungry programs on cpu2 most of the time. It is only one extra cycle to go through the hypertransport link to the other cpu so it isn't that big a deal either. You MUST however have at least one stick of ram installed per cpu in that cpu's memory area. If you don't put ram in A3 or B3, then it will not see CPU2 at all. So for a dual cpu system you must have at least 2 sticks of ram, and would be running one stick in A1, B1, A2, or B2, and one in A3 or B3. You would of course loose the dual channel advantage. It would be balanced though if the ram was the same size on both. Asus manuals are usually very well written. This one, at least the section on ram, is not up to their usual standards. Far from it. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
Quoting Steffen Grunewald: > Wouldn't this mean that, in a dual CPU system, one should better use > 4 mem modules? I just got an ASUS K8N-DL (because it is NUMA but fits into many ATX cases). It is a strange board in that CPU2 has two slots and CPU1 has 4 slots. The slots are also named strangely. Here's a diagram: | | [ ] | | | | | | | | [ ] | | | | B A | | | | 3 3 A A B B 1 2 1 2 Two banks on the left, slot B3 and slot A3. Four banks on the right, A1, A2, B1, and B2. A table of says: For CPU1Sockets Channel A A1 and A2 Channel B B1 and B2 For CPU2 Channel A A3 Channel B B3 This makes you think that if you have 4GB (4x1GB) you can put two sticks in A3/B3 and two sticks in A1/A2. But then you turn the page and you find this table: "You may install 256M, 512M, 1G, 2G or 4G registered ECC DDR DIMMs into the DIMM sockets using the memory configurations in this section. * For dual-channel configuration, the total size of memory modules installed per channel must be the same for better performance. Single CPU: A1 + A2 = B1 + B2 Dual CPU: A1 + A2 = B1 + B2 = A3 + B3 * When using one DDR DIMM module, install into A1 slot only. * When using two DDR DIMM modules, install into A1 and A2 slots only. [snip]" THe whole "A1 + A2 = B1 + B2 = A3 + B3" thing doesn't make sense because it makes it sound like if you want dual-channel memory on both CPUs you need RAM in all 6 slots, and each needs to be equal, which seems extremely limiting. I put two in A1/A2 and two in A3/B3... but then that means that channel A has 3 GB and Channel B has 1GB. Does anyone know what they're talking about, or has this board and has done some testing? I'm baffled. pedro -- Peter A. H. Peterson, technician and musician. ---=[ http://tastytronic.net/~pedro/ ]=--- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: OT (was Re: ram/raid1)
Hi Giacomo: It is poll time in Italy. However, I am not sure whether any politician to vote is aware of the situation, or he just likes that privileges are not touched. Politicians seek for votes, not for national wealth. Think about how many privileges exist (untouched) in Italy. I am not making a list not to irritate too many. Francesco On Friday 07 April 2006 11:45, Giacomo Mulas wrote: > On Fri, 7 Apr 2006, Erik Mouw wrote: > > You can order memory everywhere in the EU, that's what the "no > > obstacles for trading goods" rules are for. If you can find memory > > cheaper in (for example) Germany, buy it over there and have it shipped > > to Italy. Be sure to let your Italian vendor know that he missed a > > sale. > > Yes, you can do that if you are buying it for yourself as a private citizen > or company. If you are buying it for a public institution, than you will > require italian invoices and the like, which by and large means you have to > buy in Italy. I know, unfortunately. I had to buy a (way) suboptimal laptop > for my work, despite what I wanted (amd 64 based, large screen resolution, > more than 2GB ram...) being easily available elsewhere, just for this > reason. I had either to fork money out of my own pocket or buy something > worse and more expensive. After many years of complaining (and having > chosen many times before to spend my own money) I finally gave up and > settled for the worse, more expensive solution with the office money (what > the heck!). > > And, of course, companies _know_ this situation and exploit it, making very > different commercial offers for different national markets (see e.g. > differences in HP offers on the web, just for an example, between US and > EU). > > Bye > Giacomo > > -- > _ > > Giacomo Mulas <[EMAIL PROTECTED]> > _ > > OSSERVATORIO ASTRONOMICO DI CAGLIARI > Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) > > Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 > Tel. (UNICA): +39 070 675 4916 > _ > > "When the storms are raging around you, stay right where you are" > (Freddy Mercury) > _ > > -- > Il messaggio e' stato analizzato alla ricerca di virus o > contenuti pericolosi da MailScanner, ed e' > risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram/raid1
Hi Erik: thank you from both the scientific and the commercial point of view Francesco Pietra On Friday 07 April 2006 11:18, Erik Mouw wrote: > On Fri, Apr 07, 2006 at 06:59:29AM +0200, Francesco Pietra wrote: > > n setting up a workstation with > > > > --two amd6a 265 opterons dual core > > --Tyan K8WE S2895SA3NRF main board > > --two 360GB raid1 hd (raid 1 software by debian) > > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any > > reason to prefer two slots of memories 1GB each instead of a single 2GB > > slot? > > Most certainly. > > > The technician here maintains that two slots are needed to have needed > > two channels for raid1; it is unclear to me. > > The memory slots have nothing to do with the RAID. The reason you want > 2x 1GB is that dual (or more) Opteron designs are not SMP (Symmetric > Multi Processor), but NUMA (Non-Uniform Memory Architecture). With SMP > the two CPUs share the same bus to memory, but with NUMA each CPU has > some local memory connected to a local memory bus. The other CPU can > still get to that memory, but it's a bit slower. If you would only put > in 1x 2GB, you will severely slow down the other CPU cause it has to go > through the other CPU to do memory accesses. > > To see what I mean, get the board datasheet at > ftp://ftp.tyan.com/datasheets/d_s2895_101.pdf and look at the block > diagram on the second page. The Linux virtual memory subsystem is NUMA > aware, especially in the latest kernels (i.e.: 2.6.15 and better): it > will take care managing the memory in such a way to minimize the > amount of traffic between the CPUs. > > > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, > > that is more than twice the price in US. This is to recognize that we can > > circumvent the market leader software houses (and be more efficient) but > > we cannot avoid the system in our country which favors handlers against > > citizen (and against scientific research activities). The results of such > > policy are under the eyes. > > You can order memory everywhere in the EU, that's what the "no > obstacles for trading goods" rules are for. If you can find memory > cheaper in (for example) Germany, buy it over there and have it shipped > to Italy. Be sure to let your Italian vendor know that he missed a > sale. > > > Erik -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
Hi Len: a lucid lesson from you. Thank you Francesco Pietra On Friday 07 April 2006 15:05, Lennart Sorensen wrote: > On Fri, Apr 07, 2006 at 09:02:45AM -0400, Lennart Sorensen wrote: > > On Fri, Apr 07, 2006 at 06:57:06AM +0200, Francesco Pietra wrote: > > > In setting up a workstation with > > > > > > --two amd6a 265 opterons dual core > > > --Tyan K8WE S2895SA3NRF main board > > > --two 360GB raid1 hd (raid 1 software by debian) > > > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any > > > reason to prefer two slots of memories 1GB each instead of a single 2GB > > > slot? > > > > > > The technician here maintains that two slots are needed to have needed > > > two channels. > > > > > > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, > > > that is more than twice the price in US. This is to recognize that we > > > can circumvent the market leader software houses (and be more > > > efficient) but we cannot avoid the system in our country which favors > > > handlers against citizen (and against scientific research activities). > > > The results of such policy are under the eyes. > > > > With 2 * 1GB, you get 6.4GB/s memory bandwidth. With 1 * 2GB, you get > > 3.2GB/s memory bandwidth. I think that is a reason to prefer the > > pair 1GB sticks > > > > Socket 939/940 AMD's have dual channel memory controllers to get double > > memory bandwidth, but only if you put in at least two sticks of memory > > (and into the right slots on the board). > > Actually since you are using 2 cpus, you really would want at least 4 > sticks of identical memory to get full performance from the system since > each cpu has a dual channel memory controller. You need need memory in > each channel of each cpu if you want the maximum performance. > > Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
On Fri, Apr 07, 2006 at 09:02:45AM -0400, Lennart Sorensen wrote:
> On Fri, Apr 07, 2006 at 06:57:06AM +0200, Francesco Pietra wrote:
> > In setting up a workstation with
> >
> > --two amd6a 265 opterons dual core
> > --Tyan K8WE S2895SA3NRF main board
> > --two 360GB raid1 hd (raid 1 software by debian)
> > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason
> > to
> > prefer two slots of memories 1GB each instead of a single 2GB slot?
> >
> > The technician here maintains that two slots are needed to have needed two
> > channels.
> >
> > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that
> > is
> > more than twice the price in US. This is to recognize that we can
> > circumvent
> > the market leader software houses (and be more efficient) but we cannot
> > avoid
> > the system in our country which favors handlers against citizen (and
> > against
> > scientific research activities). The results of such policy are under the
> > eyes.
>
> With 2 * 1GB, you get 6.4GB/s memory bandwidth. With 1 * 2GB, you get
> 3.2GB/s memory bandwidth. I think that is a reason to prefer the
> pair 1GB sticks
>
> Socket 939/940 AMD's have dual channel memory controllers to get double
> memory bandwidth, but only if you put in at least two sticks of memory
> (and into the right slots on the board).
Wouldn't this mean that, in a dual CPU system, one should better use 4 mem
modules?
S
--
Steffen Grunewald * MPI Grav.Phys.(AEI) * Am Mühlenberg 1, D-14476 Potsdam
Cluster Admin * http://pandora.aei.mpg.de/merlin/ * http://www.aei.mpg.de/
* e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
No Word/PPT mails - http://www.gnu.org/philosophy/no-word-attachments.html
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
On Fri, Apr 07, 2006 at 09:02:45AM -0400, Lennart Sorensen wrote: > On Fri, Apr 07, 2006 at 06:57:06AM +0200, Francesco Pietra wrote: > > In setting up a workstation with > > > > --two amd6a 265 opterons dual core > > --Tyan K8WE S2895SA3NRF main board > > --two 360GB raid1 hd (raid 1 software by debian) > > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason > > to > > prefer two slots of memories 1GB each instead of a single 2GB slot? > > > > The technician here maintains that two slots are needed to have needed two > > channels. > > > > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that > > is > > more than twice the price in US. This is to recognize that we can > > circumvent > > the market leader software houses (and be more efficient) but we cannot > > avoid > > the system in our country which favors handlers against citizen (and > > against > > scientific research activities). The results of such policy are under the > > eyes. > > With 2 * 1GB, you get 6.4GB/s memory bandwidth. With 1 * 2GB, you get > 3.2GB/s memory bandwidth. I think that is a reason to prefer the > pair 1GB sticks > > Socket 939/940 AMD's have dual channel memory controllers to get double > memory bandwidth, but only if you put in at least two sticks of memory > (and into the right slots on the board). Actually since you are using 2 cpus, you really would want at least 4 sticks of identical memory to get full performance from the system since each cpu has a dual channel memory controller. You need need memory in each channel of each cpu if you want the maximum performance. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram
On Fri, Apr 07, 2006 at 06:57:06AM +0200, Francesco Pietra wrote: > In setting up a workstation with > > --two amd6a 265 opterons dual core > --Tyan K8WE S2895SA3NRF main board > --two 360GB raid1 hd (raid 1 software by debian) > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason > to > prefer two slots of memories 1GB each instead of a single 2GB slot? > > The technician here maintains that two slots are needed to have needed two > channels. > > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that is > more than twice the price in US. This is to recognize that we can circumvent > the market leader software houses (and be more efficient) but we cannot avoid > the system in our country which favors handlers against citizen (and against > scientific research activities). The results of such policy are under the > eyes. With 2 * 1GB, you get 6.4GB/s memory bandwidth. With 1 * 2GB, you get 3.2GB/s memory bandwidth. I think that is a reason to prefer the pair 1GB sticks Socket 939/940 AMD's have dual channel memory controllers to get double memory bandwidth, but only if you put in at least two sticks of memory (and into the right slots on the board). Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: OT (was Re: ram/raid1)
On Fri, Apr 07, 2006 at 11:45:17AM +0200, Giacomo Mulas wrote: > On Fri, 7 Apr 2006, Erik Mouw wrote: > > >You can order memory everywhere in the EU, that's what the "no > >obstacles for trading goods" rules are for. If you can find memory > >cheaper in (for example) Germany, buy it over there and have it shipped > >to Italy. Be sure to let your Italian vendor know that he missed a > >sale. > > Yes, you can do that if you are buying it for yourself as a private citizen > or company. If you are buying it for a public institution, than you will > require italian invoices and the like, which by and large means you have to > buy in Italy. I know, unfortunately. I had to buy a (way) suboptimal laptop > for my work, despite what I wanted (amd 64 based, large screen resolution, > more than 2GB ram...) being easily available elsewhere, just for this > reason. I had either to fork money out of my own pocket or buy something > worse and more expensive. After many years of complaining (and having chosen > many times before to spend my own money) I finally gave up and settled for > the worse, more expensive solution with the office money (what the heck!). That sounds like an abuse of EU rules. A German/Dutch/French/Spanish/etc company should have no problem selling stuff to Italian public institutions. If Italian public institutions require invoices from Italian companies, that is an unnecessary burden for equal access to markets. There is however a workaround, we sometimes used it at our university in order to work around silly internal accounting rules (invoices over 5k EUR had to be OK'ed by the dean, even if the money came from an EU RACE project): one of my colleagues with an own company bought the complete stuff, and resold it (in quantities less than 5 kEUR) to the university. > And, of course, companies _know_ this situation and exploit it, making very > different commercial offers for different national markets (see e.g. > differences in HP offers on the web, just for an example, between US and > EU). Complain to the EU, this is not supposed to happen. Erik -- +-- Erik Mouw -- www.harddisk-recovery.com -- +31 70 370 12 90 -- | Lab address: Delftechpark 26, 2628 XH, Delft, The Netherlands -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
OT (was Re: ram/raid1)
On Fri, 7 Apr 2006, Erik Mouw wrote: You can order memory everywhere in the EU, that's what the "no obstacles for trading goods" rules are for. If you can find memory cheaper in (for example) Germany, buy it over there and have it shipped to Italy. Be sure to let your Italian vendor know that he missed a sale. Yes, you can do that if you are buying it for yourself as a private citizen or company. If you are buying it for a public institution, than you will require italian invoices and the like, which by and large means you have to buy in Italy. I know, unfortunately. I had to buy a (way) suboptimal laptop for my work, despite what I wanted (amd 64 based, large screen resolution, more than 2GB ram...) being easily available elsewhere, just for this reason. I had either to fork money out of my own pocket or buy something worse and more expensive. After many years of complaining (and having chosen many times before to spend my own money) I finally gave up and settled for the worse, more expensive solution with the office money (what the heck!). And, of course, companies _know_ this situation and exploit it, making very different commercial offers for different national markets (see e.g. differences in HP offers on the web, just for an example, between US and EU). Bye Giacomo -- _ Giacomo Mulas <[EMAIL PROTECTED]> _ OSSERVATORIO ASTRONOMICO DI CAGLIARI Str. 54, Loc. Poggio dei Pini * 09012 Capoterra (CA) Tel. (OAC): +39 070 71180 248 Fax : +39 070 71180 222 Tel. (UNICA): +39 070 675 4916 _ "When the storms are raging around you, stay right where you are" (Freddy Mercury) _ -- Il messaggio e' stato analizzato alla ricerca di virus o contenuti pericolosi da MailScanner, ed e' risultato non infetto. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: ram/raid1
On Fri, Apr 07, 2006 at 06:59:29AM +0200, Francesco Pietra wrote: > n setting up a workstation with > > --two amd6a 265 opterons dual core > --Tyan K8WE S2895SA3NRF main board > --two 360GB raid1 hd (raid 1 software by debian) > --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason > to > prefer two slots of memories 1GB each instead of a single 2GB slot? Most certainly. > The technician here maintains that two slots are needed to have needed two > channels for raid1; it is unclear to me. The memory slots have nothing to do with the RAID. The reason you want 2x 1GB is that dual (or more) Opteron designs are not SMP (Symmetric Multi Processor), but NUMA (Non-Uniform Memory Architecture). With SMP the two CPUs share the same bus to memory, but with NUMA each CPU has some local memory connected to a local memory bus. The other CPU can still get to that memory, but it's a bit slower. If you would only put in 1x 2GB, you will severely slow down the other CPU cause it has to go through the other CPU to do memory accesses. To see what I mean, get the board datasheet at ftp://ftp.tyan.com/datasheets/d_s2895_101.pdf and look at the block diagram on the second page. The Linux virtual memory subsystem is NUMA aware, especially in the latest kernels (i.e.: 2.6.15 and better): it will take care managing the memory in such a way to minimize the amount of traffic between the CPUs. > Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that is > more than twice the price in US. This is to recognize that we can circumvent > the market leader software houses (and be more efficient) but we cannot avoid > the system in our country which favors handlers against citizen (and against > scientific research activities). The results of such policy are under the > eyes. You can order memory everywhere in the EU, that's what the "no obstacles for trading goods" rules are for. If you can find memory cheaper in (for example) Germany, buy it over there and have it shipped to Italy. Be sure to let your Italian vendor know that he missed a sale. Erik -- +-- Erik Mouw -- www.harddisk-recovery.com -- +31 70 370 12 90 -- | Lab address: Delftechpark 26, 2628 XH, Delft, The Netherlands -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
ram
In setting up a workstation with --two amd6a 265 opterons dual core --Tyan K8WE S2895SA3NRF main board --two 360GB raid1 hd (raid 1 software by debian) --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason to prefer two slots of memories 1GB each instead of a single 2GB slot? The technician here maintains that two slots are needed to have needed two channels. Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that is more than twice the price in US. This is to recognize that we can circumvent the market leader software houses (and be more efficient) but we cannot avoid the system in our country which favors handlers against citizen (and against scientific research activities). The results of such policy are under the eyes. Thanks a lot for advice Francesco Pietra -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
ram/raid1
n setting up a workstation with --two amd6a 265 opterons dual core --Tyan K8WE S2895SA3NRF main board --two 360GB raid1 hd (raid 1 software by debian) --ram 2GB (Kingston KVR400D4R3A/2G - DDR 400 Ecc Registered), is any reason to prefer two slots of memories 1GB each instead of a single 2GB slot? The technician here maintains that two slots are needed to have needed two channels for raid1; it is unclear to me. Incidentally, the 2GB Kingston is charged in Italy six hundred euros, that is more than twice the price in US. This is to recognize that we can circumvent the market leader software houses (and be more efficient) but we cannot avoid the system in our country which favors handlers against citizen (and against scientific research activities). The results of such policy are under the eyes. Thanks a lot for advice Francesco Pietra -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: i don't think its the RAM
On 10/6/05, Peter Sheldrick <[EMAIL PROTECTED]> wrote: > i just ran Memtest86+ v1.51 over the RAM and after two > passes there where no Errors - so its unlikley that > there is anything wrong with it. > > so: > (1) the RAM is ok > (2) its compatible with my mobo > > its not the RAM either! When I first started trying to install, I could run MEMTEST86 literally for days without any detected errors. However the installs failed early with an assortment of errors, mostly related to inability to read files from the disk or CD. I finally discovered that there was a problem with the board I was using when RAM was installed in slots 3,4. I could underclock the RAM and the system was rock solid or I could move the RAM to slots 1,2 and that also solved the problem. IMO if MEMTEST reports problems, you definitely have problems, but if it does not find any problem, the results are not conclusive. HTH, hank -- Beautiful Sunny Winfield, Illinois
Re: i don't think its the RAM
Could you try using the daily installer from amd64.debian.net from sept. 22? This is the one I used, I have the same mobo as you, original bios, and I have no problems. I wonder if they changed something in the kernel that is part of that build. The sata_uli there should work fine. On 10/6/05, Lennart Sorensen <[EMAIL PROTECTED]> wrote: > On Thu, Oct 06, 2005 at 02:55:50PM +0200, Peter Sheldrick wrote: > > i just ran Memtest86+ v1.51 over the RAM and after two > > passes there where no Errors - so its unlikley that > > there is anything wrong with it. > > > > so: > > (1) the RAM is ok > > (2) its compatible with my mobo > > > > its not the RAM either! > > Well doing some searches on the error message, it seems many RHEL users > have seen it too, and they supposedly made a kernel fix in late > september for it. > > A few possible things to look at: > Try booting with acpi=off since apparently that often avoids the > problem. > > There seems to be general concensus that this is a bios bug, and an > updated bios should be able to fix it (if the board maker knows to fix > it). Turning off acpi apparently avoids the buggy part of the bios on > many systems. > > Another case where the message occours is in using an em64t kernel on an > amd, or using a k8 kernel on an intel. The generic kernel should work > on either. > > Len Sorensen > > > -- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] > >
Re: i don't think its the RAM
On Thu, Oct 06, 2005 at 02:55:50PM +0200, Peter Sheldrick wrote: > i just ran Memtest86+ v1.51 over the RAM and after two > passes there where no Errors - so its unlikley that > there is anything wrong with it. > > so: > (1) the RAM is ok > (2) its compatible with my mobo > > its not the RAM either! Well doing some searches on the error message, it seems many RHEL users have seen it too, and they supposedly made a kernel fix in late september for it. A few possible things to look at: Try booting with acpi=off since apparently that often avoids the problem. There seems to be general concensus that this is a bios bug, and an updated bios should be able to fix it (if the board maker knows to fix it). Turning off acpi apparently avoids the buggy part of the bios on many systems. Another case where the message occours is in using an em64t kernel on an amd, or using a k8 kernel on an intel. The generic kernel should work on either. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: i don't think its the RAM
On Thu, Oct 06, 2005 at 02:55:50PM +0200, Peter Sheldrick wrote: > i just ran Memtest86+ v1.51 over the RAM and after two > passes there where no Errors - so its unlikley that > there is anything wrong with it. > > so: > (1) the RAM is ok > (2) its compatible with my mobo > > its not the RAM either! My experience is that 2 passes is nowhere near enough. I've been using memtest from the sysutils package recently (which runs within linux rather than as a seperate boot, and therefore can't test all your RAM at once), and I've seen errors occur on the 10th pass or later. If it passes 20, or 24 hours, I'd be more confident. Hamish -- Hamish Moffatt VK3SB <[EMAIL PROTECTED]> <[EMAIL PROTECTED]> -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
i don't think its the RAM
i just ran Memtest86+ v1.51 over the RAM and after two passes there where no Errors - so its unlikley that there is anything wrong with it. so: (1) the RAM is ok (2) its compatible with my mobo its not the RAM either! - Peter ___ Gesendet von Yahoo! Mail - Jetzt mit 1GB Speicher kostenlos - Hier anmelden: http://mail.yahoo.de -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Tyan s2885 dual 246, 2Gb RAM crashes on kernel 2.6.10
Forgot one bit of information, the stock 2.6.10 kernel source tree from kernel.org work OK. Thanks Valerio -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Tyan s2885 dual 246, 2Gb RAM crashes on kernel 2.6.10
Hello,
I have a Tyan s2885, (latest BIOS 2.04) with 2 246 opterons and 2Gb of
RAM.
The 2.6.10 kernel
(kernel-image-2.6.10-9-amd64-k8-smp_2.6.10-4_amd64.deb)
crashes on startup, while loading the initrd disk.
This is a short excerpt captured from the serial port
[...]
Attached scsi disk sdb at scsi1, channel 0, id 0, lun 0
SCSI device sdc: 488397168 512-byte hdwr sectors (250059 MB)
SCSI device sdc: drive cache: write back
SCSI device sdc: 488397168 512-byte hdwr sectors (250059 MB)
SCSI device sdc: drive cache: write back
/dev/scsi/host3/bus0/target0/lun0: p1 p2
Attached scsi disk sdc at scsi3, channel 0, id 0, lun 0
kjournald starting. Commit interval 5 seconds
EXT3-fs: mounted filesystem with ordered data mode.
INIT: version 2.86 booting
divide error: [1] SMP
CPU 0
Modules linked in: sr_mod cdrom sd_mod unix fbcon font bitblit vesafb
cfbcopyarea cfbimgblt cfbfillrect ext2 ext3 jbd mbcache sata_sil libata
scsi_mod
Pid: 0, comm: swapper Not tainted 2.6.10-9-amd64-k8-smp
RIP: 0010:[]
{smp_local_timer_interrupt+135}
RSP: 0018:803b9f78 EFLAGS: 00010046
RAX: 00030a72 RBX: RCX:
RDX: RSI: 8010c780 RDI: 0001
RBP: 80436148 R08: 8040c000 R09: 0001
R10: 0060 R11: 0001 R12: 8040dec8
-
I've tried all possible kernel option (numa=off, noacpi, nolapic, notsc,
acpi=ht). Nothing worked.
Kernels 2.6.9* and 2.6.8* are all fine.
Regards,
Valerio
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Do I need 64Bit if RAM is more than 4 GB?
Frederik Schueler wrote: > Bob Proulx wrote: > > Our own benchmarks at work show about a 7%-8% decrease in performance > > when you turn on CONFIG_HIGHMEM64G. This was for a configuration like > > this: xw8000 dual P4 3.06GHz w/ 6GB ram running two 3GB processes. > > Maybe this is related to the missing IOMMU on em64t systems? The Intel Xeon is 32-bit only. I have no benchmarks for the em64t yet. Bob pgpbGHSbAfTga.pgp Description: PGP signature
Re: Do I need 64Bit if RAM is more than 4 GB?
Hi, On Thu, Sep 09, 2004 at 07:23:24PM -0600, Bob Proulx wrote: > Our own benchmarks at work show about a 7%-8% decrease in performance > when you turn on CONFIG_HIGHMEM64G. This was for a configuration like > this: xw8000 dual P4 3.06GHz w/ 6GB ram running two 3GB processes. Maybe this is related to the missing IOMMU on em64t systems? Greetings Frederik Schueler -- ENOSIG
Re: Do I need 64Bit if RAM is more than 4 GB?
Peter Cordes wrote: > Dirk H. Schulz wrote: > > I want to run a server with more than 4 GB of RAM. I do not need > > applications/processes to address more than 4 GB each. Let's say I > > want to have 2 instances of apache on the machine, and each > > instance should address a max of 4 GB. I think if you really look at memory use of apache what you describe does not match the reality of an apache web server. You are expecting much more memory than will actually be used. Let me suggest some research to understand exactly what your memory needs will be. > > Do I need a 64Bit Linux then? Or can I install a 32Bit Linux on a Server > > with 8 GB RAM, set up my 2 instances of apache, and that`s it? > Ok, here are your options: A nice summary. Thanks for providing that. > 32bit kernel, 32bit user space: Each process gets 3GB of virtual address > space. The kernel can divvy that up however it wants. You'll need to > enable highmem support for that. If CPU performance isn't your bottleneck, > it won't matter that you don't get to use the extra registers. You lose > maybe 15% CPU performance, depending on what you're doing. (Compare SPEC > scores breakdowns for AMD's submission, if you're curious.) Our own benchmarks at work show about a 7%-8% decrease in performance when you turn on CONFIG_HIGHMEM64G. This was for a configuration like this: xw8000 dual P4 3.06GHz w/ 6GB ram running two 3GB processes. In that configuration you get a good return from the 64G mode. Or needing twice as many machines to run the jobs upon. The non-alternative is swapping and running at disk drive speeds which as you can imagine is very slow and a waste of a fast cpu. But otherwise I tend to think the 64G mode is good to avoid. But it is very stable and we have had no issues with it. > This will be the most stable configuration, because you can just > install i386 Sarge, and forget all about 64bit. (Yes, Opterons > support PAE and all that's needed for a 32bit kernel to use lots of > RAM.) The kernel has to use bounce buffers to move data around, > because it can't map all the memory. Nicely stating why there is a performance penalty. > 64bit kernel, 32bit userspace: ... Good points but let me add that this. It is a configuration you should be somewhat knowledgable about as an admin because it will be different than your buddy and you will have to be confident of your own system debugging skills. This is actually a mode I am contemplating very seriously for a work configuration. It is very similar to some of the commercial UN*X offerings. But until multi-arch realizes it is thwart with policy issues such as where things should go. > 64bit kernel, 64bit userspace: ... I currently have an amd64 server running in this mode. I have installed a full i386 chroot. If this evaluates well then this is the simplest to administer. Two machines in one running apt in parallel. (64-bit superior and 32-bit inferior in the chroot.) This is a rock solid configuration. Bob pgp6MVuiSWDdn.pgp Description: PGP signature
Re: Do I need 64Bit if RAM is more than 4 GB?
On Thu, 2004-09-09 at 17:46 -0400, Raul Miller wrote: > On Thu, Sep 09, 2004 at 03:44:32PM -0500, Ron Johnson wrote: > > Segmentation allowed a smoother software upgrade for existing > > 8080 programs, whereas the 68000 is forward-thinking, a clean break > > with the 6809. > > > > From rom a business perspective, Intel's segmented method is better, > > but from a technical point, the 68K-way is better. > > The business advantage of the 8086/8088 had little to do with the > segmented architecture and everything to do with being able to run > programs written for the 8080. > > If the 68000 was able to run 6809 programs, it would have had a similar > business advantage -- even if it achieved this without "segments". I think that's what I said. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B 484,246 sq mi (1,254,197 sq km) are needed for 6 billion people to live, 4 persons per lot, in lots that are 60'x150' (a nice suburban US plot). That is ~ California, Texas and Missouri. Alternatively, France, Spain and The United Kingdom. signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Thu, Sep 09, 2004 at 03:44:32PM -0500, Ron Johnson wrote: > Segmentation allowed a smoother software upgrade for existing > 8080 programs, whereas the 68000 is forward-thinking, a clean break > with the 6809. > > From rom a business perspective, Intel's segmented method is better, > but from a technical point, the 68K-way is better. The business advantage of the 8086/8088 had little to do with the segmented architecture and everything to do with being able to run programs written for the 8080. If the 68000 was able to run 6809 programs, it would have had a similar business advantage -- even if it achieved this without "segments". -- Raul
Re: Do I need 64Bit if RAM is more than 4 GB?
On Thu, 2004-09-09 at 12:37 -0700, Karl Hegbloom wrote: > On Thu, 2004-09-09 at 00:33 -0500, Ron Johnson wrote: > > On Wed, 2004-09-08 at 20:48 -0700, Karl Hegbloom wrote: > > > On Wed, 2004-09-08 at 08:45 -0500, Ron Johnson wrote: > > > > Back when segments were 16 bits wide, yes it was a pain. I'm old > > > > enough to have done assembly programming on the 8088. (Now that I > > > > have the wisdom of time, I understand why Intel did what they did, > > > > even though the 68K was much cleaner.) > > > > > > ? Well, so why did they do it that way? > > > > They? Motorola or Intel? > > Well, both then. Why did each make the design choice they made wrt > segmentation vs a 'flat' address model? Segmentation allowed a smoother software upgrade for existing 8080 programs, whereas the 68000 is forward-thinking, a clean break with the 6809. From rom a business perspective, Intel's segmented method is better, but from a technical point, the 68K-way is better. For example, IBM was able to easily use existing 8 bit peripheral parts in the design of the PC, but Motorola had to design all new support parts for the 68K. IOW, low-cost and time-to-market was valuable, even then. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B SpaceShipOne powered flight: #1 : 17 December 2003 - 1497k/h, 20400m, 15 sec thrust burn #2 : 8 April 2004 - Mach 2, 31500m, 40 sec thrust burn #3 : 13 May 2004 - Mach 2.5, 63420m, 55 sec thrust burn #4 : 21 June 20004 - ~2400k/h, 100125m, 70 sec thrust burn signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Thu, 2004-09-09 at 00:33 -0500, Ron Johnson wrote: > On Wed, 2004-09-08 at 20:48 -0700, Karl Hegbloom wrote: > > On Wed, 2004-09-08 at 08:45 -0500, Ron Johnson wrote: > > > Back when segments were 16 bits wide, yes it was a pain. I'm old > > > enough to have done assembly programming on the 8088. (Now that I > > > have the wisdom of time, I understand why Intel did what they did, > > > even though the 68K was much cleaner.) > > > > ? Well, so why did they do it that way? > > They? Motorola or Intel? Well, both then. Why did each make the design choice they made wrt segmentation vs a 'flat' address model? > > Let me guess... it's just so that you can get more than 16 bits of > > address space for one program? > -- Karl Hegbloom (o_ mailto:[EMAIL PROTECTED] //\ jabber:[EMAIL PROTECTED] V_/_ yahoo:karlheg
Re: Do I need 64Bit if RAM is more than 4 GB?
On Thu, 2004-09-09 at 09:20 -0400, Stephen Frost wrote: > * Mattias Wadenstein ([EMAIL PROTECTED]) wrote: > > >Even in 1995 4GB would have been a rather expensive amount of ram even > > >for a high end sparc or power machine. > > > > Well, instead of searching for prices, go find an old manual of the > > largest sun sparc32 smp? The one I can think of right now is the ss1000, > > were there any bigger ones? Before the ultrasparc days that is. > > SparcCenter 2000, we had one here actually. 10 system boards, up to 2 > procs each (as I recall we had 16 processors in ours), 0, 8 or 16 SIMMs > per board which were 8M or 32M. I believe we had the 8 system boards > that had CPUs full with 32M SIMMs for a total of 4G. I'm pretty sure > the system was capable of going to 32 CPUs w/ the dual-CPU modules. I'm > not sure if you could get more than 4G of memory in to it though. Well, I guess that answers my question, then. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B "A great many open minds should be closed for repairs." Toledo Blade Newspaper signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
* Mattias Wadenstein ([EMAIL PROTECTED]) wrote: > >Even in 1995 4GB would have been a rather expensive amount of ram even > >for a high end sparc or power machine. > > Well, instead of searching for prices, go find an old manual of the > largest sun sparc32 smp? The one I can think of right now is the ss1000, > were there any bigger ones? Before the ultrasparc days that is. SparcCenter 2000, we had one here actually. 10 system boards, up to 2 procs each (as I recall we had 16 processors in ours), 0, 8 or 16 SIMMs per board which were 8M or 32M. I believe we had the 8 system boards that had CPUs full with 32M SIMMs for a total of 4G. I'm pretty sure the system was capable of going to 32 CPUs w/ the dual-CPU modules. I'm not sure if you could get more than 4G of memory in to it though. Stephen signature.asc Description: Digital signature
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 17:30 -0400, Lennart Sorensen wrote: > On Wed, Sep 08, 2004 at 04:09:59PM -0500, Ron Johnson wrote: > > http://www3.sk.sympatico.ca/jbayko/cpu5.html#Sec5Part4 > > "Part IV: IBM RS/6000 POWER chips (1990). . . ." > > "Thirty two 32-bit registers were defined for the POWER1 integer > > unit, which also included certain string operations, as well as > > all load/store operations." > > Blah blah blah POWER2 > > "It was superceded by the POWER3 (Early 1998), with eight functional > > units (two FPU, three integer (two single cycle, one multicycle), > > two load/store, and branch unit), but capable of operating at much > > higher clock speeds. In addition, a 64 bit version, the PowerPC > > A35 (Apache), was designed for the AS/400 E series" > > > > So, the first 64 bit POWER chips arrived 8 years after the 32 > > bit versions. > > 1998 seems like a fairly resonable time to start getting into 64bit. I > guess it does indicate the power wasn't designed as 64bit to begin with, > but seems to have been designed well enough that extending it later was > reasonable to do. As Ben mentioned, Alpha and DEC OSF/1 (a.k.a. Digital Unix a.k.a tru64) have been 64 bit since 1992. We've been running Alpha/VMS since 1995. > > > Well, sparc64 has been around an awful long time. Adding PAE-like hacks > > > seems > > > > Since 1995. > > > > There were largish SMP SPARC32 boxen for many years before the > > SPARC64 came into existence. I can't find any references on the > > web, but some of those big boxen had to have more than 4GB RAM. > > I wonder how much 4GB ram would have cost in 1995 or even 1998. I > remember getting 16M for a 486 for $600 in 1992. I think it was 1996 > when I got 128M for about the same amount. The price lists I found once > for Decstation 5000 boxes had ram listed at around $5 for 128M in > 1991. > > Even in 1995 4GB would have been a rather expensive amount of ram even > for a high end sparc or power machine. If you're a big company buying a box with 16 or 32 CPUs... -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B "What has a tiny brain, a big mouth, and an opinion nobody cares about? You!" from Murphy Brown signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 23:49 +0200, Mattias Wadenstein wrote: > On Wed, 8 Sep 2004, Lennart Sorensen wrote: > > > On Wed, Sep 08, 2004 at 04:09:59PM -0500, Ron Johnson wrote: [snip] > Ehm. There is no 64-bit version of the POWER ISA, it was > extended/fixed/replaced by the PowerPC ISA which was designed with 32 and > 64 bit implementations to begin with (I think). POWER3 is a ppc64 > implementation selling under the POWER brand, not a 64-bit POWER > implementation. > > Before the POWER3 (and other ppc64 implementations), the SMP rs6000 > machines where 32-bit ppcs and had address limitations which meant that > the maximum ammount of memory supported was around 3-3.5 gigs. This is in > place even for the ppc smp sp2 node called "silver", which I happen to run > a couple of for ftp.se.debian.org. These were the high-end computational > resources that were replaced by the POWER3, and couldn't handle more that > 4 gigs of ram. > > The IBM sales manuals are around and pretty good at telling you exactly > what hardware combinations are/were supported, I think you'll notice that > the support for more than 4GB came at the launch of the POWER3 (or the > RS64(?) chip, another ppc64 implementation used by ibm for the commercial > computing segment rather than technical computing). Well, I guess that answers the original question regarding whether POWER32 could handle more than 4GB... -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B "He was about as useful in a crisis as a sheep." Dorothy Eden signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 20:48 -0700, Karl Hegbloom wrote: > On Wed, 2004-09-08 at 08:45 -0500, Ron Johnson wrote: > > Back when segments were 16 bits wide, yes it was a pain. I'm old > > enough to have done assembly programming on the 8088. (Now that I > > have the wisdom of time, I understand why Intel did what they did, > > even though the 68K was much cleaner.) > > ? Well, so why did they do it that way? They? Motorola or Intel? > Let me guess... it's just so that you can get more than 16 bits of > address space for one program? -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B Remember when environmental doom-and-gloomers said that it would take 10 years to put out the 750 post-GW1 oil fires? Yet they were all out in 6 months. Remember when environmental doom-and-gloomers said in ~1975 that the oil would run out in 50 years? signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 08:45 -0500, Ron Johnson wrote: > Back when segments were 16 bits wide, yes it was a pain. I'm old > enough to have done assembly programming on the 8088. (Now that I > have the wisdom of time, I understand why Intel did what they did, > even though the 68K was much cleaner.) ? Well, so why did they do it that way? Let me guess... it's just so that you can get more than 16 bits of address space for one program? -- Karl Hegbloom (o_ mailto:[EMAIL PROTECTED] //\ jabber:[EMAIL PROTECTED] V_/_ yahoo:karlheg
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wednesday 08 September 2004 23:13, Peter Cordes wrote: [Snip a good description of 32-vs 64-bit] > 3D acceleration is only possible with 64/64 kernel/user, or 32/32, if that > matters to you. The nvidia drivers provide 3D acceleration for both 64 and 32 bit apps on 64 bit kernels. Paul
Re: Do I need 64Bit if RAM is more than 4 GB?
On Tue, Sep 07, 2004 at 09:24:31PM +0200, Dirk H. Schulz wrote: > Hi folks, > > I hope this is the right place for this kind of question: > > I want to run a server with more than 4 GB of RAM. I do not need > applications/processes to address more than 4 GB each. Let's say I want to > have 2 instances of apache on the machine, and each instance should address > a max of 4 GB. > > Do I need a 64Bit Linux then? Or can I install a 32Bit Linux on a Server > with 8 GB RAM, set up my 2 instances of apache, and that`s it? > > Sorry for that kind of basic question but I did not find any docs on that > googling around. Ok, here are your options: 32bit kernel, 32bit user space: Each process gets 3GB of virtual address space. The kernel can divvy that up however it wants. You'll need to enable highmem support for that. If CPU performance isn't your bottleneck, it won't matter that you don't get to use the extra registers. You lose maybe 15% CPU performance, depending on what you're doing. (Compare SPEC scores breakdowns for AMD's submission, if you're curious.) This will be the most stable configuration, because you can just install i386 Sarge, and forget all about 64bit. (Yes, Opterons support PAE and all that's needed for a 32bit kernel to use lots of RAM.) The kernel has to use bounce buffers to move data around, because it can't map all the memory. The kernel uses the remaining 1GB of virtual address space for itself, and maps all the RAM it can. What's left is highmem. 64bit kernel, 32bit userspace: Each process gets 4GB of virtual address space. Disadvantage: you need a module-init-tools, iptables, and so on that can talk to the kernel. All the normal system calls by 32bit programs go through a translation layer (not much overhead, don't worry) so you can boot a 64bit kernel with root=/dev/path-to-i386-Sarge. You can install some 64bit libraries, or make some statically linked binaries, so you can run a few things 64bit. statically linked AMD64 iptables might be the best way to go. (you can debootstrap a 64bit chroot so you can apt-get install the stuff you need... Use dchroot to make your chroot convenient). The ia32->amd64 kernel translation layer works well, so while it might not be as stable as a fully i386 system, and you have to worry about special kernel interfaces that don't get 32bit translated, you don't have to worry about bugs in userspace programs like storing a pointer in an int variable. On an SMP system, the kernel will know about NUMA and be able to allocate memory that's attached to the CPU the requesting process is running on. 64bit kernel, 64bit userspace: Each process gets 64bit virtual address space. Same as above, but you have to worry about user-space too. Not all packages are available, and some of them have bugs because they truncate pointers to 32bits in some places. Just all around less stable still, not to mention that AMD64 Debian might not release with sarge, so security updates won't come from security.debian.org. You can install libraries so that you can run i386 binaries if you have any binary-only programs. (32bit code can't link to 64bit code at all, ever, on amd64. (not counting special translation layers like the kernel-userspace boundary).) 3D acceleration is only possible with 64/64 kernel/user, or 32/32, if that matters to you. 32bit kernel, 64bit userspace: not possible. -- #define X(x,y) x##y Peter Cordes ; e-mail: X([EMAIL PROTECTED] , des.ca) "The gods confound the man who first found out how to distinguish the hours! Confound him, too, who in this place set up a sundial, to cut and hack my day so wretchedly into small pieces!" -- Plautus, 200 BC
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 8 Sep 2004, Lennart Sorensen wrote: On Wed, Sep 08, 2004 at 04:09:59PM -0500, Ron Johnson wrote: http://www3.sk.sympatico.ca/jbayko/cpu5.html#Sec5Part4 "Part IV: IBM RS/6000 POWER chips (1990). . . ." "Thirty two 32-bit registers were defined for the POWER1 integer unit, which also included certain string operations, as well as all load/store operations." Blah blah blah POWER2 "It was superceded by the POWER3 (Early 1998), with eight functional units (two FPU, three integer (two single cycle, one multicycle), two load/store, and branch unit), but capable of operating at much higher clock speeds. In addition, a 64 bit version, the PowerPC A35 (Apache), was designed for the AS/400 E series" So, the first 64 bit POWER chips arrived 8 years after the 32 bit versions. 1998 seems like a fairly resonable time to start getting into 64bit. I guess it does indicate the power wasn't designed as 64bit to begin with, but seems to have been designed well enough that extending it later was reasonable to do. Ehm. There is no 64-bit version of the POWER ISA, it was extended/fixed/replaced by the PowerPC ISA which was designed with 32 and 64 bit implementations to begin with (I think). POWER3 is a ppc64 implementation selling under the POWER brand, not a 64-bit POWER implementation. Before the POWER3 (and other ppc64 implementations), the SMP rs6000 machines where 32-bit ppcs and had address limitations which meant that the maximum ammount of memory supported was around 3-3.5 gigs. This is in place even for the ppc smp sp2 node called "silver", which I happen to run a couple of for ftp.se.debian.org. These were the high-end computational resources that were replaced by the POWER3, and couldn't handle more that 4 gigs of ram. The IBM sales manuals are around and pretty good at telling you exactly what hardware combinations are/were supported, I think you'll notice that the support for more than 4GB came at the launch of the POWER3 (or the RS64(?) chip, another ppc64 implementation used by ibm for the commercial computing segment rather than technical computing). Well, sparc64 has been around an awful long time. Adding PAE-like hacks seems Since 1995. There were largish SMP SPARC32 boxen for many years before the SPARC64 came into existence. I can't find any references on the web, but some of those big boxen had to have more than 4GB RAM. I wonder how much 4GB ram would have cost in 1995 or even 1998. I remember getting 16M for a 486 for $600 in 1992. I think it was 1996 when I got 128M for about the same amount. The price lists I found once for Decstation 5000 boxes had ram listed at around $5 for 128M in 1991. Even in 1995 4GB would have been a rather expensive amount of ram even for a high end sparc or power machine. Well, instead of searching for prices, go find an old manual of the largest sun sparc32 smp? The one I can think of right now is the ss1000, were there any bigger ones? Before the ultrasparc days that is. /Mattias Wadenstein
Re: Do I need 64Bit if RAM is more than 4 GB?
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 what about Alpha.. Alpha has been a 64bit since the begining: http://www3.sk.sympatico.ca/jbayko/cpu5.html#Sec5Part5 - -ben "Unix is user friendly, Its just picky about its friends." -BEGIN PGP SIGNATURE- Version: GnuPG v1.2.4 (GNU/Linux) iD8DBQFBP33vflzKmtpiQEMRArqPAJ9ztkv+/Ea+GDXxcQfappm9dxHm8ACaAoxW DMELdkITSmcwwafmt3Ylm7g= =B8xy -END PGP SIGNATURE-
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, Sep 08, 2004 at 04:09:59PM -0500, Ron Johnson wrote: > http://www3.sk.sympatico.ca/jbayko/cpu5.html#Sec5Part4 > "Part IV: IBM RS/6000 POWER chips (1990). . . ." > "Thirty two 32-bit registers were defined for the POWER1 integer > unit, which also included certain string operations, as well as > all load/store operations." > Blah blah blah POWER2 > "It was superceded by the POWER3 (Early 1998), with eight functional > units (two FPU, three integer (two single cycle, one multicycle), > two load/store, and branch unit), but capable of operating at much > higher clock speeds. In addition, a 64 bit version, the PowerPC > A35 (Apache), was designed for the AS/400 E series" > > So, the first 64 bit POWER chips arrived 8 years after the 32 > bit versions. 1998 seems like a fairly resonable time to start getting into 64bit. I guess it does indicate the power wasn't designed as 64bit to begin with, but seems to have been designed well enough that extending it later was reasonable to do. > > Well, sparc64 has been around an awful long time. Adding PAE-like hacks > > seems > > Since 1995. > > There were largish SMP SPARC32 boxen for many years before the > SPARC64 came into existence. I can't find any references on the > web, but some of those big boxen had to have more than 4GB RAM. I wonder how much 4GB ram would have cost in 1995 or even 1998. I remember getting 16M for a 486 for $600 in 1992. I think it was 1996 when I got 128M for about the same amount. The price lists I found once for Decstation 5000 boxes had ram listed at around $5 for 128M in 1991. Even in 1995 4GB would have been a rather expensive amount of ram even for a high end sparc or power machine. Len Sorensen
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 16:18 +0100, Paul Brook wrote: > On Wednesday 08 September 2004 15:42, Ron Johnson wrote: > > On Wed, 2004-09-08 at 15:03 +0100, Paul Brook wrote: > > > On Wednesday 08 September 2004 14:45, Ron Johnson wrote: > > > > In fact, it seems to me that *any* 32 bit processor (SPARC, HPPA, > > > > Power) that wants to be able to use more than 4GB of total RAM > > > > would have to use such a segmentation scheme. > > > > > > Err, all of the above have 64-bit variants. > > > > Ye, but they didn't *start* with 64 bit variants. > > Actually Power did. It was designed as a 64-bit architecture that can also be > run/implemented with only 32-bits. http://www3.sk.sympatico.ca/jbayko/cpu5.html#Sec5Part4 "Part IV: IBM RS/6000 POWER chips (1990). . . ." "Thirty two 32-bit registers were defined for the POWER1 integer unit, which also included certain string operations, as well as all load/store operations." Blah blah blah POWER2 "It was superceded by the POWER3 (Early 1998), with eight functional units (two FPU, three integer (two single cycle, one multicycle), two load/store, and branch unit), but capable of operating at much higher clock speeds. In addition, a 64 bit version, the PowerPC A35 (Apache), was designed for the AS/400 E series" So, the first 64 bit POWER chips arrived 8 years after the 32 bit versions. > > >I don't know if the > > > 32-bit variants support more than 4GB ram, but I doubt it. > > > > Oh come on. You think the SPARC32s, Powers & PA-RISCs that ran > > big Solaris, AIX and HP-UX SMP boxen in big shops *never* had more > > than 4GB of RAM? > > > > I find it supremely hard to believe that Intel is the only company > > to have a 32 bit chip that can address more than 4GB of RAM. > > Well, sparc64 has been around an awful long time. Adding PAE-like hacks seems Since 1995. > a strange decision when you have backwards compatible 64-bit CPUs. There were largish SMP SPARC32 boxen for many years before the SPARC64 came into existence. I can't find any references on the web, but some of those big boxen had to have more than 4GB RAM. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B "Peace is a daily, a weekly, a monthly process, gradually changing opinions, slowly eroding old barriers, quietly building new structures." John F Kennedy signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wednesday 08 September 2004 15:42, Ron Johnson wrote: > On Wed, 2004-09-08 at 15:03 +0100, Paul Brook wrote: > > On Wednesday 08 September 2004 14:45, Ron Johnson wrote: > > > In fact, it seems to me that *any* 32 bit processor (SPARC, HPPA, > > > Power) that wants to be able to use more than 4GB of total RAM > > > would have to use such a segmentation scheme. > > > > Err, all of the above have 64-bit variants. > > Ye, but they didn't *start* with 64 bit variants. Actually Power did. It was designed as a 64-bit architecture that can also be run/implemented with only 32-bits. > > I don't know if the > > 32-bit variants support more than 4GB ram, but I doubt it. > > Oh come on. You think the SPARC32s, Powers & PA-RISCs that ran > big Solaris, AIX and HP-UX SMP boxen in big shops *never* had more > than 4GB of RAM? > > I find it supremely hard to believe that Intel is the only company > to have a 32 bit chip that can address more than 4GB of RAM. Well, sparc64 has been around an awful long time. Adding PAE-like hacks seems a strange decision when you have backwards compatible 64-bit CPUs. Paul
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 15:03 +0100, Paul Brook wrote: > On Wednesday 08 September 2004 14:45, Ron Johnson wrote: > > In fact, it seems to me that *any* 32 bit processor (SPARC, HPPA, > > Power) that wants to be able to use more than 4GB of total RAM > > would have to use such a segmentation scheme. > > Err, all of the above have 64-bit variants. Ye, but they didn't *start* with 64 bit variants. >I don't know if the > 32-bit variants support more than 4GB ram, but I doubt it. Oh come on. You think the SPARC32s, Powers & PA-RISCs that ran big Solaris, AIX and HP-UX SMP boxen in big shops *never* had more than 4GB of RAM? I find it supremely hard to believe that Intel is the only company to have a 32 bit chip that can address more than 4GB of RAM. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B "Sometime they'll give a war and nobody will come." Carl Sandburg Oh, come on. Sure they will. That's what testosterone is for... signature.asc Description: This is a digitally signed message part
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wednesday 08 September 2004 14:45, Ron Johnson wrote: > In fact, it seems to me that *any* 32 bit processor (SPARC, HPPA, > Power) that wants to be able to use more than 4GB of total RAM > would have to use such a segmentation scheme. Err, all of the above have 64-bit variants. I don't know if the 32-bit variants support more than 4GB ram, but I doubt it. The 64-bit Power CPUs don't have a real 32-bit mode. They always operate on 64-bit registers. 32-bit code is executed by automatically sign- or zero-extending 32-bit memory loads, and certain instructions (eg. comparisons) ignore the top 32 bits. I suspect SPARC does something similar, I don't know about PARISC. Paul
Re: Do I need 64Bit if RAM is more than 4 GB?
On Wed, 2004-09-08 at 08:23 -0400, Lennart Sorensen wrote: > On Wed, Sep 08, 2004 at 02:29:09AM -0500, Ron Johnson wrote: > > I'd say "hack" is a strong word for PAE, which is just an extension > > of the segmented memory concept. > > I think intel's messy segmented memory model is quite a hack. At least > with the 386 in protected mode you could treat memory as flat (except > for that hole at 640 to 1024k, but you could just start at 1M and forget > about it). With PAE now you are back to having segments and mapping and > such going on again. Having done a bit mf programing at the OS level on > a 486, I sure felt like intel's memory segments were a hack, which even > made the pagetables in protected mode somewhat messy to create. Back when segments were 16 bits wide, yes it was a pain. I'm old enough to have done assembly programming on the 8088. (Now that I have the wisdom of time, I understand why Intel did what they did, even though the 68K was much cleaner.) Now (actually since the 386), though, the segments are 32 bits wide, and the need to manipulate segments has migrated into the kernel, while each userland app sees a 32 bit address space. To me, that's an acceptable compromise. In fact, it seems to me that *any* 32 bit processor (SPARC, HPPA, Power) that wants to be able to use more than 4GB of total RAM would have to use such a segmentation scheme. -- - Ron Johnson, Jr. Jefferson, LA USA PGP Key ID 8834C06B NAMBLA - Nat'l Assoc of Marlon Brando Look-Alikes (Yes, it's a South Park reference.) signature.asc Description: This is a digitally signed message part
