LINPACK: Intel's Nehalem versus AMD Shanghai
by Johan De Gelas on November 28, 2008 12:00 AM EST- Posted in
- IT Computing general
A "beta BIOS update" broke compatibility with ESX, so we had to postpone our virtualization testing on our quad CPU AMD 8384 System.
So we started an in depth comparison of the 45 nm Opterons, Xeons and Core i7 CPUs. One of our benchmarks, the famous LINPACK (you can read all about it here) painted a pretty interesting performance picture.
We had to test with a matrix size of 18000 (2.5 GB of RAM necessary), as we only had 3 GB of DDR-3 on the Core i7 platform. That should not be a huge problem as we tested with only one CPU. We normally need about 4 GB for each quadcore CPU to reach the best performance.
We also used the 9.1 version of Intel's LINPACK, as we wanted the same binary on both platforms. As we have show before, this version of LINPACK performs best on both AMD and Intel platforms when the matrix size is low. The current 10.1 version does not work on AMD CPUs unfortunately.
We don't pretend that the comparison is completely fair: the Nehalem platform uses unbuffered RAM which has slightly lower latency and higher bandwidth than the Xeon "Nehalem" will get. But we had to satisfy our curiousity: how does the new "Shanghai" core compare to "Nehalem"?
Quite interesting, don't you think? Hyperthreading (SMT) gives the Nehalem core a significant advantage in most multi-threaded applications, but not in Linpack: it slows the CPU down by 10%. May we have found the first multi-threaded application that is slowed down by Hyperthreading on Nehalem? That should not spoil the fun for Intel though, as many other HPC benchmarks show a larger gap. AMD has the advantage of being first to the market, Nehalem based Xeons are still a few months away.
Also, the impact of the memory subsystem is limited, as a 50% increase in memory speed results in a meager 6% performance increase. The Math Kernel Libraries are so well optimized that the effect of memory speed is minimized. This in great contrast to other HPC applications where the tripple channel DDR-3 memory system of Nehalem really pays off. More later...
Facebook
Twitter
RSS
60 Comments
View All Comments
ZootyGray - Tuesday, December 2, 2008 - link
First you bollox the test w a bios floppp. convenient/Then you can't afford enough DDR3.
Nobody seems to know if bozo turbo is off.
You don't know that a real IT would not use turbo.
Your readers all hate you for being so smug.
You have reinforced why I don't come here anymore.
You are conning and lying about AMD.
You use pro-Intel conditions and pretend you are fair.
You discredit a very detailed test posted with real facts and also it bears the names of the testers - and their source. That test, in the 'comments' is far better than anything on this website.
You insult people who refuse to consume the manure you spread.
You treat us all like mushrooms.
You will refute all of the above.
In short, you suck so bad.
ANYONE SEEKING REAL INFO RE AMD SERVERS OR DESKTOPS GOTO AMDZONE.
and if I wanted to know about intel I would not come here for that either.
Do you actually expect any better treatment than this?
I am sure also that you will tell us all that
the i7 TLB BUG IS JUST OK TOO.
You will probably del this post. Like I care. Ship of fools.
ZootyGray - Tuesday, December 2, 2008 - link
Oh I forgot/running the amd with old slow ram - that's rich.
Thx for revealing yourself. You wanna call me an amd fanboy? It's really true.
Your forum kikked me for being pro-amd. or was it my bad attitude.
- spit
.
I find it really hard to even be here. I love truth. out.
s1ugh34d - Saturday, November 29, 2008 - link
Toms hardware has gotten better, but 25 page reviews that fit on two pages printed, thats still pushing it especially for the ads. Anandtech is better on my phone( which I am posting from)anyhow, and has a heck of a lot less pages and ads.I see the issue everyone has with the recent Intel king of the hill performance. But I hate to be the one throwing laymen terms out, Amd is not as good as intel. Multi socket systems aside, every bit of Intels consumer market outdue AMD.
Server tests also take alot more time, and more documentation, before they can be put into a chart for our eyes. I think this was mainly focused on the HT differences in the intel chip, and AMD scores were posted as refrence points.
The DDR3 itself being unregistered(as noted) should show the readers that this is not an apples to apples compairson. I'm just lost on why i7 chips are almost strictly consumer right now, so why not just wait for the server chips and server memory.
I can't believe that the new Phenom II X4 hasn't even got a simple blog post about its ridiculous OC claims on the net. It's a week in and nothing even small shows up here.
wwswimming - Saturday, November 29, 2008 - link
i like websites that have 1 or 2 ads per page.
TH has one of the most ad-bombarded visual presentations i've
ever seen on a website. ads between posts in the forums.
if this was TH, there would be ads here in the "post comments
to blog" section !
BLaber - Friday, November 28, 2008 - link
Please let us all know whether the TURBO BOOST feature was in use when you were performing the above tests.Thanksthebeastie - Friday, November 28, 2008 - link
Despite the countless millions of dollars companies like Intel have injected into the Linux kernel over the years a simple Linpack test shows its flaws as it can't deal with the extra HT logical cores.It shows too much money is injected into one spot and should be more evenly spread into other OS projects like FreeBSD. Just because one project has a nicer installer shouldn't be its justification.
Griswold - Saturday, November 29, 2008 - link
Do you have any evidence to back your claims up or do you (foolishly) expect everyone to take your word for granted?thebeastie - Monday, December 1, 2008 - link
LOL there is no doubt its a flaw in the Linux kernel, thats not to say they wont fix it.I would be pretty sure it will be fixed pretty soon, its just that MS had it all going fine straight out of the bat.
I can't believe you would even think its anything else then the Linux kernel poorly dealing with the HT, what would you think it is? the power of God?
narlzac85 - Friday, November 28, 2008 - link
Did you disable the nehelem turbo boost or whatever it was called? If you are only using 1 core (not a realistic scenario I assume), then nehelem will clock that core 1 or 2 multipliers higher. However, that is unlikely to be happening in a heavily loaded server right?JohanAnandtech - Saturday, November 29, 2008 - link
LINPACK is extremely well threaded. That means that all cores are used to their full potential. But to be sure we disabled turbo booster, as we speculate it will probably not be used on the server products. That is why I labeled the Core i7 at it's clockspeed. (and also because I absolutely hate Intel's and AMD's numbering systems)