Re: Mixing CPU's
Hello Nathan, Am 2005-07-27 02:06:11, schrieb Nathan Dragun: > Actually they were reffering to the fact that a 1xx would not work > period on a 2xx board. From what I understand the HT pathway is only > used when using multiple cpus (ie: cpu to cpu communications). So even > if you use one lone 1xx cpu, aparently, its supposed to work on a 2xx board. This is what I have already tried. It does not work. > Nathan Michelle -- Linux-User #280138 with the Linux Counter, http://counter.li.org/ Michelle Konzack Apt. 917 ICQ #328449886 50, rue de Soultz MSM LinuxMichi 0033/3/8845235667100 Strasbourg/France IRC #Debian (irc.icq.com) signature.pgp Description: Digital signature
Re: Mixing CPU's
On Wed, Jul 27, 2005 at 02:06:11AM -0400, Nathan Dragun wrote: > I'm not sure if this is CPU related or if its mobo chipset related, but > there is a significant difference between the 939's and 940's, as the > 939's like standard unregistered memory... 940's will only take reg > ecc. Hence the reason they went from the 754 to the 939 since no one > wanted to buy the more expensive ram necessary to run it (possibly other > reasons too). socket 754 and 939 seemed to be released at just about the same time. 754 has single channel memory, 939 has dual channel (like 940, but yes the difference there is unregistered/registered memory use). I forget what the new socket(s) are called that will replace those eventually. > Actually they were reffering to the fact that a 1xx would not work > period on a 2xx board. From what I understand the HT pathway is only > used when using multiple cpus (ie: cpu to cpu communications). So even > if you use one lone 1xx cpu, aparently, its supposed to work on a 2xx board. I wish I had a chance to try it. But no opterons around here. :) Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
Samat Jain wrote: Lennart Sorensen wrote: On Sat, Jul 23, 2005 at 12:01:27PM +0200, Michelle Konzack wrote: I had a MSI K8D Master3 (380 Euro, Two Opteron 240) which drived me crazy and I have tried to run one of those CPU's on a Singel-Opteron mainboard from Tyan and it does not work. Then I have bought two Tyan Dual-Opteron Boards andused it with only one CPU each, which works very fine. Then I have tried to run a Opteron 142 on a Dual-CPU-Board and it does not work. So you NEED the right CPU to your Mainboard. You can not use a 1xx on a Mainboard which was made for a 2xx. I wonder if that is for technical reasons, or just lack of cpu support in the bioses. The Opteron 1xx series of processors are basically conventional Athlon 64 939-pin processors with an extra pin. I'm not sure if this is CPU related or if its mobo chipset related, but there is a significant difference between the 939's and 940's, as the 939's like standard unregistered memory... 940's will only take reg ecc. Hence the reason they went from the 754 to the 939 since no one wanted to buy the more expensive ram necessary to run it (possibly other reasons too). The Opteron 2xx and 8xx series, however, are different--they both have a different count of what is called "coherent HyperTransport links." These links are used for transferring SMP protocol type things (cache coherency checks, cache snooping, etc) between processors. The 2xx series has one of these links (it can connect to one other processor), while the 8xx has three. This is the reason why you can't use a 1xx processor in a 2-CPU system, or a 2xx processor in a 4 or 8-CPU system, but can use an 8xx processor everywhere (the extra coherent links just will not be used). Actually they were reffering to the fact that a 1xx would not work period on a 2xx board. From what I understand the HT pathway is only used when using multiple cpus (ie: cpu to cpu communications). So even if you use one lone 1xx cpu, aparently, its supposed to work on a 2xx board. Nathan Code is poetry. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
Lennart Sorensen wrote: On Sat, Jul 23, 2005 at 12:01:27PM +0200, Michelle Konzack wrote: I had a MSI K8D Master3 (380 Euro, Two Opteron 240) which drived me crazy and I have tried to run one of those CPU's on a Singel-Opteron mainboard from Tyan and it does not work. Then I have bought two Tyan Dual-Opteron Boards andused it with only one CPU each, which works very fine. Then I have tried to run a Opteron 142 on a Dual-CPU-Board and it does not work. So you NEED the right CPU to your Mainboard. You can not use a 1xx on a Mainboard which was made for a 2xx. I wonder if that is for technical reasons, or just lack of cpu support in the bioses. Len Sorensen The Opteron 1xx series of processors are basically conventional Athlon 64 939-pin processors with an extra pin. The Opteron 2xx and 8xx series, however, are different--they both have a different count of what is called "coherent HyperTransport links." These links are used for transferring SMP protocol type things (cache coherency checks, cache snooping, etc) between processors. The 2xx series has one of these links (it can connect to one other processor), while the 8xx has three. This is the reason why you can't use a 1xx processor in a 2-CPU system, or a 2xx processor in a 4 or 8-CPU system, but can use an 8xx processor everywhere (the extra coherent links just will not be used). If processors are different speeds I'd imagine there'd be problem with the cache coherency checking. A processor that is significantly faster could in theory change the contents of the cache in the mean time the other processor is receiving information about the first's now defunct state. SMP has safeguards for this, but AFAIK they rely on the processors being the same speed. There are probably engineering things you can do to avoid problems with mismatched speeds, but since very few people use mismatched processors I don't think this is a manufacturer's concern. I'd have to agree with the consensus about dual-core and single-core mixing--there doesn't appear to be any technical barrier for it not to work. But I know I don't have the money or need to try it. I'm not sure how things will behave when the number of processors are not a power of 2, though. Samat -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
Am 2005-07-25 10:34:08, schrieb Lennart Sorensen: > I wonder if that is for technical reasons, or just lack of cpu support > in the bioses. AFAIK a technical reason... > Len Sorensen Greetings Michelle -- Linux-User #280138 with the Linux Counter, http://counter.li.org/ Michelle Konzack Apt. 917 ICQ #328449886 50, rue de Soultz MSM LinuxMichi 0033/3/8845235667100 Strasbourg/France IRC #Debian (irc.icq.com) signature.pgp Description: Digital signature
Re: Mixing CPU's
On Sat, Jul 23, 2005 at 12:01:27PM +0200, Michelle Konzack wrote: > I had a MSI K8D Master3 (380 Euro, Two Opteron 240) which drived me > crazy and I have tried to run one of those CPU's on a Singel-Opteron > mainboard from Tyan and it does not work. > > Then I have bought two Tyan Dual-Opteron Boards andused it with only > one CPU each, which works very fine. > > Then I have tried to run a Opteron 142 on a Dual-CPU-Board and it does > not work. > > So you NEED the right CPU to your Mainboard. You can not use a 1xx on > a Mainboard which was made for a 2xx. I wonder if that is for technical reasons, or just lack of cpu support in the bioses. Len Sorensen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
On Sat, Jul 23, 2005 at 12:33:01AM -0400, Nathan Dragun wrote: > Has anyone tried mixing the processor types of multi-cpu systems? What > was the outcome? I have two opteron 244's in my machine. One is stepping 8, the other is 10. 10 supports powernow and is 90nm. Works fine. Matching clockrates is essential. Even if the bios supported mismatched clocks, the kernel assumes all cpu's are the same speed. I don't know if one dual-core and one single core can be used together. I've been wondering that myself. Some kernel things make be tuned (or even assume) the number of cpu's is a power of 2, or at least divisible by 2. Though, the cpu hot plug project should have removed any of those assumptions. -- Tom Vier <[EMAIL PROTECTED]> DSA Key ID 0x15741ECE -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
RE: Mixing CPU's
Correct! :-) > -Original Message- > From: Nathan Dragun [mailto:[EMAIL PROTECTED] > Sent: Saturday, July 23, 2005 11:03 PM > To: Latchezar Dimitrov; debian-amd64@lists.debian.org > Subject: Re: Mixing CPU's > > Latchezar Dimitrov wrote: > > >Well, it's true it's theoretical until you try it, however there has > >been a discussion recently I believe Len Sorensen gave a good > >description. Also AMD is your friend - go and check it out. > You'll see > >why the cpu are called dual-CORE not dual-cpu > > > >Latchezar > > > Awesome, found the thread...very insightful. I was under the > impression that there was more than one HT pipeline in the > dual cores, but I guess > not. Seems like the only issue would be finding how to > match them, and > as you mentioned before with the AMD site, probably easiest > way is to spec out the stats and see what matches. > > Thanks, > Nathan > > > Code is poetry. > >
Re: Mixing CPU's
Latchezar Dimitrov wrote: Well, it's true it's theoretical until you try it, however there has been a discussion recently I believe Len Sorensen gave a good description. Also AMD is your friend - go and check it out. You'll see why the cpu are called dual-CORE not dual-cpu Latchezar Awesome, found the thread...very insightful. I was under the impression that there was more than one HT pipeline in the dual cores, but I guess not. Seems like the only issue would be finding how to match them, and as you mentioned before with the AMD site, probably easiest way is to spec out the stats and see what matches. Thanks, Nathan Code is poetry. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
RE: Mixing CPU's
Well, it's true it's theoretical until you try it, however there has been a discussion recently I believe Len Sorensen gave a good description. Also AMD is your friend - go and check it out. You'll see why the cpu are called dual-CORE not dual-cpu Latchezar > -Original Message- > From: Nathan Dragun [mailto:[EMAIL PROTECTED] > Sent: Saturday, July 23, 2005 8:47 PM > To: Latchezar Dimitrov; debian-amd64@lists.debian.org > Subject: Re: Mixing CPU's > > Latchezar Dimitrov wrote: > > >All that being true however from the arch of duals one > should be able > >to put together dual-core 2-way and a "regular", i.e., > single-core cpu > >on a 2-way opteron mobo. > > > >Latchezar > > > > > I guess thats what I'm the most curious about; mixing these > dual and single core processors. Do the required Hyper > Transport pipelines match up when you use a dual with a single? > > If so... how do you match the duals with the comparable > single cores since it seems they use a different naming > scheme (ex: single 244 = = dual 265). It all seems > 'theoretically' plausable, I'm just wondering if anyone has > actually tried yet. > > Nathan > > > Code is poetry. > >
Re: Mixing CPU's
Latchezar Dimitrov wrote: All that being true however from the arch of duals one should be able to put together dual-core 2-way and a "regular", i.e., single-core cpu on a 2-way opteron mobo. Latchezar I guess thats what I'm the most curious about; mixing these dual and single core processors. Do the required Hyper Transport pipelines match up when you use a dual with a single? If so... how do you match the duals with the comparable single cores since it seems they use a different naming scheme (ex: single 244 = = dual 265). It all seems 'theoretically' plausable, I'm just wondering if anyone has actually tried yet. Nathan Code is poetry. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
RE: Mixing CPU's
All that being true however from the arch of duals one should be able to put together dual-core 2-way and a "regular", i.e., single-core cpu on a 2-way opteron mobo. Latchezar > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] > Sent: Saturday, July 23, 2005 9:06 AM > To: Nathan Dragun > Cc: Paolo Alexis Falcone; debian-amd64@lists.debian.org > Subject: Re: Mixing CPU's > > Hi Nathan, > > May I ask what is the purpose of the original question ? > > Do I understand correctly that you want to mix a single and a > double CPU, perhaps even a quad or 8-way Opteron CPU in the > same system ? > > Do you understand that f.e. the 2 way CPUs have an extra > "Bus" between them to speed up communication and share memory > to the other CPU. > In 4 and 8-way CPUs you will find 3 of those extra "Buses". > > These means f.e. that in a 4-way setup you can share the > memory between those CPUs, processes etc. according to your needs. > > I believe that mixing up different CPU types and /or speeds > is basically like putting for different wheels on your car > and see how far you can go with them ;-). > > Does that make things clearer ? > > I am not sure yet if your background is to make "a one size > fits all computer" or you just have nothing to do. Anyway, > either way good luck ;-) > > Best regards > > Nils Valentin > > > Paolo Alexis Falcone wrote: > > > >>On 7/23/05, Nathan Dragun <[EMAIL PROTECTED]> wrote: > >> > >> > >>>Has anyone tried mixing the processor types of multi-cpu systems? > >>>What was the outcome? > >>> > >>>Has anyone tried mixing dual-core and single-core cpus together? > >>> > >>> > >>>I'm truely interested in knowing if its necessary to shell out all > >>>that money for 2+ chips... > >>> > >>> > >> > >>Generally it's not a good idea. You've got timing issues to > consider, > >>among other things. > >> > >> > >> > > What kind of timing issues? What other things > > Just because it isn't recomended dosen't mean it won't > work. On the > > contrary, half the things 'they' tell you won't work will > work, they > > just want to make more money off you. > > > > Nathan > > > > Code is poetry. > > > > > > -- > > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > > with a subject of "unsubscribe". Trouble? Contact > > [EMAIL PROTECTED] > > > > > > > -- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact > [EMAIL PROTECTED] > >
Re: Mixing CPU's
Hi Nathan, May I ask what is the purpose of the original question ? Do I understand correctly that you want to mix a single and a double CPU, perhaps even a quad or 8-way Opteron CPU in the same system ? Do you understand that f.e. the 2 way CPUs have an extra "Bus" between them to speed up communication and share memory to the other CPU. In 4 and 8-way CPUs you will find 3 of those extra "Buses". These means f.e. that in a 4-way setup you can share the memory between those CPUs, processes etc. according to your needs. I believe that mixing up different CPU types and /or speeds is basically like putting for different wheels on your car and see how far you can go with them ;-). Does that make things clearer ? I am not sure yet if your background is to make "a one size fits all computer" or you just have nothing to do. Anyway, either way good luck ;-) Best regards Nils Valentin Paolo Alexis Falcone wrote: > >>On 7/23/05, Nathan Dragun <[EMAIL PROTECTED]> wrote: >> >> >>>Has anyone tried mixing the processor types of multi-cpu systems? What >>>was the outcome? >>> >>>Has anyone tried mixing dual-core and single-core cpus together? >>> >>> >>>I'm truely interested in knowing if its necessary to shell out all that >>>money for 2+ chips... >>> >>> >> >>Generally it's not a good idea. You've got timing issues to consider, >>among other things. >> >> >> > What kind of timing issues? What other things > Just because it isn't recomended dosen't mean it won't work. On the > contrary, half the things 'they' tell you won't work will work, they > just want to make more money off you. > > Nathan > > Code is poetry. > > > -- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact > [EMAIL PROTECTED] > > -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
Am 2005-07-23 01:14:51, schrieb Nathan Dragun: > What kind of timing issues? What other things > Just because it isn't recomended dosen't mean it won't work. On the > contrary, half the things 'they' tell you won't work will work, they > just want to make more money off you. I had a MSI K8D Master3 (380 Euro, Two Opteron 240) which drived me crazy and I have tried to run one of those CPU's on a Singel-Opteron mainboard from Tyan and it does not work. Then I have bought two Tyan Dual-Opteron Boards andused it with only one CPU each, which works very fine. Then I have tried to run a Opteron 142 on a Dual-CPU-Board and it does not work. So you NEED the right CPU to your Mainboard. You can not use a 1xx on a Mainboard which was made for a 2xx. > Nathan > > Code is poetry. Greetings Michelle -- Linux-User #280138 with the Linux Counter, http://counter.li.org/ Michelle Konzack Apt. 917 ICQ #328449886 50, rue de Soultz MSM LinuxMichi 0033/3/8845235667100 Strasbourg/France IRC #Debian (irc.icq.com) signature.pgp Description: Digital signature
Re: Mixing CPU's
Paolo Alexis Falcone wrote: On 7/23/05, Nathan Dragun <[EMAIL PROTECTED]> wrote: Has anyone tried mixing the processor types of multi-cpu systems? What was the outcome? Has anyone tried mixing dual-core and single-core cpus together? I'm truely interested in knowing if its necessary to shell out all that money for 2+ chips... Generally it's not a good idea. You've got timing issues to consider, among other things. What kind of timing issues? What other things Just because it isn't recomended dosen't mean it won't work. On the contrary, half the things 'they' tell you won't work will work, they just want to make more money off you. Nathan Code is poetry. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Mixing CPU's
On 7/23/05, Nathan Dragun <[EMAIL PROTECTED]> wrote: > Has anyone tried mixing the processor types of multi-cpu systems? What > was the outcome? > > Has anyone tried mixing dual-core and single-core cpus together? > > > I'm truely interested in knowing if its necessary to shell out all that > money for 2+ chips... Generally it's not a good idea. You've got timing issues to consider, among other things. -- Paolo Alexis Falcone [EMAIL PROTECTED]