Re: Intel EM64T vs. Opteron

2005-10-18 Thread Tim Cutts


On 14 Oct 2005, at 8:06 am, mike wrote:


Basically, I'd be looking at an Intel 830D (3.0ghz, dual-core, EM64T)
processor-based system, with 1 gig of ram and SATA HD vs. a
single-core Opteron 1.8ghz (or a dual-processor NOT dual-core Opteron
1.8ghz) system, same HD and same RAM. Has anyone had the opportunity
to benchmark this, or have any real experience with changing the
underlying platform?


Our experience with IBM HS20 blade servers shows the 2.4 GHz Opteron  
blows the 3.2 GHz EM64T out of the water on our bioinformatics codes,  
which are predominantly limited by integer performance and memory  
bandwidth.  And uses a lot less power (this is significant when you  
have about 1000 processors in the cluster!).  No more EM64T for us,  
at least for a while.


One thing to be careful with, if it's a dual CPU opteron machine that  
you're benchmarking, is that you need a really up to date kernel that  
gets the NUMA topology right; our initial benchmarks were  
disappointing, and it turned out that the the numa topology stuff in  
the kernel was exactly backwards, so every process was always  
accessing memory local to the other CPU.  This has been fixed in  
recent kernels.


Tim


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Re: Intel EM64T vs. Opteron

2005-10-15 Thread Pete Harlan
On Fri, Oct 14, 2005 at 01:59:16PM -0700, mike wrote:
> The main question is 64-bit support using Debian-amd64/kernel 2.6.x
> [on EM64T]. Is it any different than installing Debian-amd64 on an
> Opteron box?

In my experience, no, it's no different at all.  Ignoring
speed/whatever, running Sarge 64-bit Debian on our one (dual
processor, hyperthreading) EM64T box has been no different than on the
numerous Opteron boxes we run.

I can't comment on the debian _kernels_, as we compile from upstream,
but we've had no problems with the box.

--Pete


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Re: Intel EM64T vs. Opteron

2005-10-14 Thread mike
On 10/14/05, Lennart Sorensen <[EMAIL PROTECTED]> wrote:
>
> Do you know what chipset/motherboard the p4 system would be using?
>

LGA775 socket
E7230 chipset it appears - i think i saw another model or two that may
have slightly different chipsets, but i can't find them now.

anyway, from the discussions it sounds like Opterons are just a better
choice for now, based on this and some other collateral and forum
posts i've been able to google up.



Re: Intel EM64T vs. Opteron

2005-10-14 Thread Lennart Sorensen
On Fri, Oct 14, 2005 at 01:59:16PM -0700, mike wrote:
> Please note the "I have a chance" line - I have some discounts that I
> can apply against Intel 830D's which would take the overall cost down
> to less than an Opteron machine - and I'd be getting dual-core for
> that lower price as well.
> 
> Pentium 4 830D's have EM64T support; it's not just Xeons.

Well discounts are nice.

> You can compare anything you want. It's a comparison for that reason.
> My two options are an 830D or an Opteron-based system.  The main
> question is 64-bit support using Debian-amd64/kernel 2.6.x. Is it any
> different than installing Debian-amd64 on an Opteron box? I already
> run it on 7 Opteron-based systems and an Athlon64 system.

Installing should be exactly the same as long as the chipset and such
in the machine is supported in 2.6.8 kernel (or 2.6.12 if you use my
version of the installer).

> >From the replies after this, it sounds like the EM64T 64-bit arch is a
> little slower than AMD64. However, maybe that's just a bottleneck that
> will be fixed soon, much like when Hyperthreading came out?

No, I don't think the current em64t design will ever run faster in 64bit
than 32bit.  The opteron/a64 was designed from scratch with 64bit in
mind.  The netburst architecture wasn't.  It had to be modified to
handle 64bit without a complete redesign from scratch.  It is however
from what I gather not a particularly noticeable difference, and if the
program has any advantages from 64bit (like using sse for floating piint
instead of x87) and using more than 4GB memory mapped, then it should be
worth the slight drop in speed.

Do you know what chipset/motherboard the p4 system would be using?

Len Sorensen


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Re: Intel EM64T vs. Opteron

2005-10-14 Thread mike
On 10/14/05, Lennart Sorensen <[EMAIL PROTECTED]> wrote:

> > I have a chance soon to switch out my servers, and going the Intel
> > route I would be able to get a lot more bang for my buck. The website
> > says that EM64T works fine with Debian-amd64, I just want to make sure
> > there's no gotchas, compared to using it on an Opteron-based system.
>
> When did xeon servers become cheaper than opteron servers? :)
>

Please note the "I have a chance" line - I have some discounts that I
can apply against Intel 830D's which would take the overall cost down
to less than an Opteron machine - and I'd be getting dual-core for
that lower price as well.

Pentium 4 830D's have EM64T support; it's not just Xeons.

>
> Remember you can not compare an 830D to an opteron.  That's like
> comparing an 830D to a xeon.  Rather unreasonable comparison.  Compare
> it to an Athlon 64 X2 system.
>

You can compare anything you want. It's a comparison for that reason.
My two options are an 830D or an Opteron-based system.  The main
question is 64-bit support using Debian-amd64/kernel 2.6.x. Is it any
different than installing Debian-amd64 on an Opteron box? I already
run it on 7 Opteron-based systems and an Athlon64 system.

>From the replies after this, it sounds like the EM64T 64-bit arch is a
little slower than AMD64. However, maybe that's just a bottleneck that
will be fixed soon, much like when Hyperthreading came out?



Re: Intel EM64T vs. Opteron

2005-10-14 Thread Lennart Sorensen
On Fri, Oct 14, 2005 at 11:58:56AM -0400, Karl Magdsick wrote:
> Along the same lines, do the EMT64 chips still actually run slower in
> 64-bit mode as compared to 32-bit mode?

If the xeon em64t users on this list in the past are anything to go by,
the answer is yes.  Not usually much but a bit.  Oposit of the amds in
any case.

Len Sorensen


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Re: Intel EM64T vs. Opteron

2005-10-14 Thread Lennart Sorensen
On Fri, Oct 14, 2005 at 12:06:24AM -0700, mike wrote:
> I recently (re-)subscribed to the mailing list, I have a couple
> specific questions, and can't seem to find any good data about it, and
> would like to tap people who probably have some real-world experience
> with it.
> 
> First off, I am a user of Debian-amd64 right now - and I love it. I
> run it on Opterons. It screams.
> 
> I have a chance soon to switch out my servers, and going the Intel
> route I would be able to get a lot more bang for my buck. The website
> says that EM64T works fine with Debian-amd64, I just want to make sure
> there's no gotchas, compared to using it on an Opteron-based system.

When did xeon servers become cheaper than opteron servers? :)

> Basically, I'd be looking at an Intel 830D (3.0ghz, dual-core, EM64T)
> processor-based system, with 1 gig of ram and SATA HD vs. a
> single-core Opteron 1.8ghz (or a dual-processor NOT dual-core Opteron
> 1.8ghz) system, same HD and same RAM. Has anyone had the opportunity
> to benchmark this, or have any real experience with changing the
> underlying platform? I can build an 830D based system with dual-cores
> for cheaper than a single-core Opteron box right now (I have some
> discounts) but if the performance isn't as good as it seems it should
> be, or there's some odd things that must be done to get EM64T to work
> properly under Debian-amd64, I'd like to know that before committing
> to the hardware choice :)

I know the athlon 64 is very fast and certainly faster than the p4 I have
around here (at a much higreh clock rate).  From what I have gathered
reading this list, the P4/xeon slows down a little in general when you
enable 64bit mode.  AMDs speed up.  Given the AMD is usually faster
at mosts tasks already in 32 bit mode, the situation is not improved
(for intel) in 64bit mode.

Remember you can not compare an 830D to an opteron.  That's like
comparing an 830D to a xeon.  Rather unreasonable comparison.  Compare
it to an Athlon 64 X2 system.

> I plan on running the latest Linux kernel 2.6.x - two of my servers
> will be LVS machines and I will have a handful of webservers running
> this platform, in case you're wondering about specific usage.
> 
> Thanks in advance for any feedback!

To me the netburst architecture looked wrong when it came out.  I still
think it looks wrong and I will be happy when intel finally dumps it for
the pentium-m based architecture in their server and desktop lineups.
Until then I will buy what is the fastest (and cheapest) systems you can
get, which is AMD based systems.  Of course I haven't been buying dual
core, which do seem rather pricey relate to single core.

for example:
Asus A8V Deluxe $132cdn
AMD Athlon 64 X2 3800 $440cdn
AMD Athlon 64 X2 4800 $1066cdn
2 x 512MB DDR400 $122cdn
2 x 1GB DDR400 $330cdn
Total is $694cdn (3800 1GB) to $1528cdn (4800 2GB)

Asus P5LD2 $153cdn
Intel Pentium D 820 $314cdn (2.8GHz)
Intel Pentium D 840 $670cdn (3.2GHz)
(I am ignoring the extreme edition as being silly expensive)
2 x 512MB DDR2-533 $140cdn
2 x 1GB DDR2-533 $280cdn
Total is $607cdn (D820 1GB) to $1103cdn (840D 2GB)

Video, HD, case, etc should be similar between systems and hence not
relevant.

An X2 3800 = 2xAthlon64 3200 (which in my experience is easily faster
than a 3.2GHz P4).  An X2 4800 = 2xAthlon64 4000.  The 4600 is the same
but with half the cache.

Given the more efficient memory interface of the athlon64 x2 than the
shared FSB 830D, and that the 3800 is a dual of a cpu that already is as
fast as the single core in the 840D, it is actually fair to compare the
X2 3800 to the 840D (which costs more of course).  The 820D and 830D
might be cheaper, but they will also be slower for probably just any any
task you throw at them.

Enough ranting from me. :)

Len Sorensen


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Re: Intel EM64T vs. Opteron

2005-10-14 Thread Karl Magdsick
Along the same lines, do the EMT64 chips still actually run slower in
64-bit mode as compared to 32-bit mode?


-Karl



Re: Intel EM64T vs. Opteron

2005-10-14 Thread mikepolniak
On 00:06 Fri 14 Oct , mike wrote:
> Hi all,
> 
> I recently (re-)subscribed to the mailing list, I have a couple
> specific questions, and can't seem to find any good data about it, and
> would like to tap people who probably have some real-world experience
> with it.
> 
> First off, I am a user of Debian-amd64 right now - and I love it. I
> run it on Opterons. It screams.
> 
> I have a chance soon to switch out my servers, and going the Intel
> route I would be able to get a lot more bang for my buck. The website
> says that EM64T works fine with Debian-amd64, I just want to make sure
> there's no gotchas, compared to using it on an Opteron-based system.
> 
> Basically, I'd be looking at an Intel 830D (3.0ghz, dual-core, EM64T)
> processor-based system, with 1 gig of ram and SATA HD vs. a
> single-core Opteron 1.8ghz (or a dual-processor NOT dual-core Opteron
> 1.8ghz) system, same HD and same RAM. Has anyone had the opportunity
> to benchmark this, or have any real experience with changing the
> underlying platform? I can build an 830D based system with dual-cores
> for cheaper than a single-core Opteron box right now (I have some
> discounts) but if the performance isn't as good as it seems it should
> be, or there's some odd things that must be done to get EM64T to work
> properly under Debian-amd64, I'd like to know that before committing
> to the hardware choice :)

Just a heads up about the big drop in Opteron pricing:
http://www.theinquirer.net/?article=26927


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



Intel EM64T vs. Opteron

2005-10-14 Thread mike
Hi all,

I recently (re-)subscribed to the mailing list, I have a couple
specific questions, and can't seem to find any good data about it, and
would like to tap people who probably have some real-world experience
with it.

First off, I am a user of Debian-amd64 right now - and I love it. I
run it on Opterons. It screams.

I have a chance soon to switch out my servers, and going the Intel
route I would be able to get a lot more bang for my buck. The website
says that EM64T works fine with Debian-amd64, I just want to make sure
there's no gotchas, compared to using it on an Opteron-based system.

Basically, I'd be looking at an Intel 830D (3.0ghz, dual-core, EM64T)
processor-based system, with 1 gig of ram and SATA HD vs. a
single-core Opteron 1.8ghz (or a dual-processor NOT dual-core Opteron
1.8ghz) system, same HD and same RAM. Has anyone had the opportunity
to benchmark this, or have any real experience with changing the
underlying platform? I can build an 830D based system with dual-cores
for cheaper than a single-core Opteron box right now (I have some
discounts) but if the performance isn't as good as it seems it should
be, or there's some odd things that must be done to get EM64T to work
properly under Debian-amd64, I'd like to know that before committing
to the hardware choice :)

I plan on running the latest Linux kernel 2.6.x - two of my servers
will be LVS machines and I will have a handful of webservers running
this platform, in case you're wondering about specific usage.

Thanks in advance for any feedback!

- mike