Intel Xeon 5570: Smashing SAP records.

Quote:

---

Originally Posted by STEvil

Precedent, as an operating system will have efficiency changes so will a driver release or a chipset.

As I have pointed out there is possibility for system efficiency issues. If you choose to disregard this then that is your issue. Any other review I read based on processors or video cards for enthusiasts is immediately beaten up when systems are not detailed fully enough.

---

IBM posted these results 2 days apart.

Don't you think IBM would have been trying to make it an apples to apples comparison, as this information is surely used by IBM's customers when they purchase equipment like this. Hell IBM may even provide it to prospective customers to make their purchasing decisions easier.

Taking your argument to it's logical conclusion, all SPEC results are worthless.

Quote:

---

If you notice I said HT was both good and bad.

---

Of course I noticed, that is why I provided you with that information to help you out.

Quote:

---

How many applications have the differences been benchmarked in?

---

This review did 6 applications. I am of course unaware of how many applications have been benchmarked by all benchmarkers worldwide at this point. ;)

Quote:

---

I'm starting to wonder actually. What memory timings and subtimings were used? Could the memory loadout difference cause an efficiency issue? Really if you scale the AMD numbers to 2.93ghz you get about 4775.

---

Are you mixing up your comments in respect of the IBM submitted SAP results and the testing done by the geeks at Performance and Architecture Lab (PAL), Los Alamos National Laboratory?

Quote:

---

Its only 225 points short of the Intel numbers which isnt that bad if the benchmark is being limited somewhere.

---

The limit seems to be that AMD only had twice the number of cores that Intel had.

Quote:

---

Originally Posted by Chad Boga

IBM posted these results 2 days apart.

Don't you think IBM would have been trying to make it an apples to apples comparison, as this information is surely used by IBM's customers when they purchase equipment like this. Hell IBM may even provide it to prospective customers to make their purchasing decisions easier.

---

Neither of us works for IBM, do we?

Quote:

---

Taking your argument to it's logical conclusion, all SPEC results are worthless.

---

I guess if you are unwilling to do some research you are right.

Quote:

---

Of course I noticed, that is why I provided you with that information to help you out.

---

If I already knew it why did I need more information?

Quote:

---

This review did 6 applications. I am of course unaware of how many applications have been benchmarked by all benchmarkers worldwide at this point. ;)

---

Here's a hint: not enough.

Quote:

---

Are you mixing up your comments in respect of the IBM submitted SAP results and the testing done by the geeks at Performance and Architecture Lab (PAL), Los Alamos National Laboratory?

---

No, all of my numbers are with respect to the SAP results. If you look at the IBM/Redhat (Dell/Win further down are lower scores) 4-processor-24-core 2.66ghz Intel results also submitted by IBM they follow a similar 1.16x trend (IE: score the same as AMD's 4-processor-16-core 2.7ghz system) and then there's the Fujitsu system at the bottom of the page with is an x5570 (identical to the IBM scoring 5100) which scores 4715. There is no information on the hard disk controller available.

Quote:

---

The limit seems to be that AMD only had twice the number of cores that Intel had.

---

And the other Intel x3850 systems had three times.

Here's are the likely HDD control candidates.

The LS42 probably used the LSI 1064 SAS controller
http://www-03.ibm.com/systems/xbc/co...2_7902hdd.html

The x3650 probably used the ServeRAID-8k-l
http://www-03.ibm.com/systems/xbc/co...0/x3650io.html

The controller for the Fujitsu system is not listed anywhere. It would be interesting to see if how much using SQL server depreciated the the results.



My point that you seem to be missing every time is that there are system differences which you are not researching. If you are not going to research these you are not continuing this discussion in good faith.

Quote:

---

Originally Posted by STEvil

Neither of us works for IBM, do we?

---

So you think it is a reasonable prospect that IBM submitted these results on the AMD system with a HDD controller that performs worse/notably worse than the HDD controller they used on the Nehalem system they submitted for, so that these results should not be viewed as reflective of SAP performance customers could expect?

If that is even close to your position, I think that is preposterous and I would suggest that you are not approaching this discussion with good faith.

Quote:

---

I guess if you are unwilling to do some research you are right.

---

Are you just taking "Benchmarks don't matter" to a higher plane?

My belief is that they did not research the issue else there would have been notations.

Only someone who can not research a subject will find benchmarks worthless, which only you have inferred.



The benchmarks presented are less than detailed. I have pointed this out, you have found it a point of contention that this seemingly professionally done benchmark does not need to adhere to. Do you agree or disagree that notation should be supplied?

http://www.eetimes.com/news/semi/sho...cleID=10800621

Quote:

---

Originally Posted by EET

AMD deal with IBM appears to end earlier alliances
David Lammers
EE Times

AUSTIN, Texas — Advanced Micro Devices Inc. has switched partners again, announcing Wednesday (Jan. 8) that it will work with IBM Microelectronics to co-develop process technology at the 65-nanometer and 45-nm nodes.

---

http://www.tomshardware.com/news/amd...ntel,6175.html
AMD First to 22 Nm, Challenges Intel
6:30 PM - August 18, 2008 by Wolfgang Gruener

Quote:

---

Yorktown Heights (NY) - IBM and its chip development partners, including AMD, made a stunning announcement today, apparently beating Intel in the successful production of the first functional 22 nm SRAM cell. 22 nm processors are still three years out in the future, but IBM’s news is a good sign that chip manufacturer will be able to easily scale to this new level by the end of 2011.

---

http://www.internetnews.com/ent-news...le.php/3724636

Quote:

---

That Long-Rumored IBM/AMD Merger: Could it Work?
UPDATED: Speculation continues aloud that a union may in the works. Analysts weigh in on the likelihood -- and the benefits.

---

http://www.intel.com/performance/ser...mp/summary.htm
http://www.intel.com/in/business/hig...htm?iid=SEARCH

Why the links? It seems out of character for IBM to seemingly cut its own throat by over hyping Intel in this manner STEvil. Without IBM, AMD dies years ago or when IBM bailed them out when the extra Hot Thoroughbreds launched at early Prescott like temps. IBM aren't known as weak testers and are trusted enough by the industry. It hasn't or didn't matter when they tested AMD processors and they should get the same respect for the Intel tests. The last Intel resource page is to show AMD is already being hit on the Performance and Price fronts.

NEGATIVE:
No matter how many records are broken, adoption of an Opteron Killer will be slow because companies all around the world are hurting for New loans to finance some of their material replacements, upgrades and in some cases maintenance. That's bad news for Intel and AMD.

http://www.internetnews.com/stats/ar...+for+Intel.htm

One of Intel's better markets.

Quote:

---

Originally Posted by STEvil

My belief is that they did not research the issue else there would have been notations.

Only someone who can not research a subject will find benchmarks worthless, which only you have inferred.


The benchmarks presented are less than detailed. I have pointed this out, you have found it a point of contention that this seemingly professionally done benchmark does not need to adhere to. Do you agree or disagree that notation should be supplied?

---

It would be better to have it than to not have it.

However it is very common to make judgements without 100% of any and every possible variation detailed to the nth degree.

To suggest these results are not meaningful is quite churlish in my view and I will explain why in more detail.

1. The AMD result was not posted by Intel, as you may often see in SPEC results where AMD and Intel post up results of hobbled systems of each other to then use in marketing campaigns.

This result was posted by AMD's most important technological partner, who has a vested interest in portraying the results of AMD based systems that they sell, in the best possible light.

Hell, if I was suspicious of anything, it would be that the hdd controller on the Intel system was less efficient than the AMD system. Afterall, IBM also gave the AMD system more RAM.

2. Give AMD a 10% boost if you must on the basis of perhaps not having as good a hdd controller as the Intel based system(even though I find this percentage to be quite extravagant), and don't even bother to adjust for the extra RAM that the Shanghai system had, do you think this materially effects what an ominous result this is for AMD's likely 2009/2010 share of the SAP/ERP market?

3. This flogging AMD has been handed in SAP, is pretty consistent with the hiding they have been given in SPEC results too, so it is not like it really comes as any great surprise if the Intel result is so much better.

I have not said that IBM is hyping one or the other Donnie, only that the numbers given are tainted to be because full system specifications are not given.

Thank you for finally acknowledging that, Chad.

Yes, of course it is as I have stated already, no need to repeat that.

Then dont infer that they are meaningless and do not extrapolate that I have said they are as I have repeatedly said that they are not meaningless.

I agree, 10% might be a bit much but really given the non-uniformity of the benchmark standard maybe not..

edit - here's some i7 HT vs no HT numbers for you: http://techreport.com/articles.x/15818/1

Quote:

---

Originally Posted by STEvil

I have not said that IBM is hyping one or the other Donnie, only that the numbers given are tainted to be because full system specifications are not given.

Thank you for finally acknowledging that, Chad.

Yes, of course it is as I have stated already, no need to repeat that.

Then dont infer that they are meaningless and do not extrapolate that I have said they are as I have repeatedly said that they are not meaningless.

I agree, 10% might be a bit much but really given the non-uniformity of the benchmark standard maybe not..

edit - here's some i7 HT vs no HT numbers for you: http://techreport.com/articles.x/15818/1

---

No you didn't and I didn't say you did:up: I'm saying it is kind of strange that IBM wasn't Questioned about ANY of the Opteron Server tests. That my friend is what I questioned you about. Most of the time IBM gives minimum to NO specs of the tested systems.

As per your link-->Because of the presence of Hyper-Threading, you have to look at that final graph to make sense of these results. The benchmark keeps eight threads active all of the time on the Core i7, which reduces per-thread performance. Once we get to the end of the road, though, and estimate the total projected points per day, both the Core i7 and Hyper-Threading prove to be winners. Without Hyper-Threading, the Core i7-965 Extreme is only marginally faster than the QX9770, but with it, the contest becomes a rout.

Now of course Hyper threading will not do anything special with apps designed for one or two cores, nothing new there:up: On one forum, a guy asked one of the better smarter forum members a similar question. How can a Quad core (at that time) not show it is faster during individual tests but still show up much faster at the end of the test. I compared it to Brick Layers. One excellent layer could lay about 40 bricks a minute. He can get beaten by two mediocre layers that can lay about 25 bricks per minute. This, even as each of the layers are slower.

EDIT: This happens on a lot of multithreaded apps!

Singling out one benchmark of the rutt isnt helping your cause Donnie, leave the thread if you're going to do that.

Just a small comment from Donnies post;
They said that the i7 was 'only marginally faster than the QX9770"..
To me thats funny.
The QX9770 has got to be one of the all time monster chips and in my mind if the i7 is beating it by just a CH it's a winner..:D

Quote:

---

Originally Posted by STEvil

Singling out one benchmark of the rutt isnt helping your cause Donnie, leave the thread if you're going to do that.

---

WOW! So your doubting the Validity of IBM and HP is helping? With all due respect, you seem to be the only one stuck in a rut. BTW, it is NOT my cause or case for that matter.

Quote:

---

Originally Posted by Movieman

Just a small comment from Donnies post;
They said that the i7 was 'only marginally faster than the QX9770"..
To me thats funny.
The QX9770 has got to be one of the all time monster chips and in my mind if the i7 is beating it by just a CH it's a winner..

---

Yes and that's two QX9770 running on a Dual Socket Board:up: Then folks are still wondering why 2 X 5570 doing 16 threads smashed SAP records:up: Also impressive is that they can't use FSB for an excuse either since the iV8 has two FSB's to work with.

Quote:

---

Originally Posted by Donnie27

WOW! So your doubting the Validity of IBM and HP is helping? With all due respect, you seem to be the only one stuck in a rut. BTW, it is NOT my cause or case for that matter.



Yes and that's two QX9770 running on a Dual Socket Board:up: Then folks are still wondering why 2 X 5570 doing 16 threads smashed SAP records:up: Also impressive is that they can't use FSB for an excuse either since the iV8 has two FSB's to work with.

---

Just a small correction, those would be QX9775's in the dual board..essentially the same as the QX9770's though..

One more small point to you and Stevil:
I know you and stevil both..
Both good people so be decent to each other huh?
Thanks!

Quote:

---

Originally Posted by Movieman

Just a small correction, those would be QX9775's in the dual board..essentially the same as the QX9770's though..

One more small point to you and Stevil:
I know you and stevil both..
Both good people so be decent to each other huh?
Thanks!

---

I wasn't trying to be a jerk or anything and have no grudges against anyone here. I just found what was being said here hard to believe. I didn't try to disrespect him or anyone else. Sure I know "Intel Xeon 5570: Smashing SAP records." will upset some here. Sure this should get questioned and I posted why I thought IBM wouldn't try to pull a fast one. Correct, they are QX9775's, my bad!

Quote:

---

Originally Posted by Donnie27

I wasn't trying to be a jerk or anything and have no grudges against anyone here. I just found what was being said here hard to believe. I didn't try to disrespect him or anyone else. Sure I know "Intel Xeon 5570: Smashing SAP records." will upset some here. Sure this should get questioned and I posted why I thought IBM wouldn't try to pull a fast one. Correct, they are QX9775's, my bad!

---

wasn't mad, just saw a 'tone" in both you and Steve.
I know both of you to be good people and thought to divert any possible issues..
dave the kind..:rofl:

Quote:

---

Originally Posted by Movieman

wasn't mad, just saw a 'tone" in both you and Steve.
I know both of you to be good people and thought to divert any possible issues..
dave the kind..:rofl:

---

No biggie:up: Didn't think you were mad at all.

Quote:

---

Originally Posted by Donnie27

The computer users who buy the most PC's have always known who AMD is=p This old line about "users don't know who AMD is" is just a plain old AMD fanboy myth. Tech Web had a story years about talking New users making up less of the market than DIYers. Families are on the 3rd to 5th computers depending on size. I worked 4 computers today, that's why I had to leave and couldn't answer you sooner.

When computers start costing less than the average Car, the market exploded. Most New Buyers have grown up with computers. They aren't ding bat sheep and are CONSUMERS who'll pick Honda over Ford if Ford sucks more, Dr. Peper over Coke or Pepsi no matter how many ads are shown and etc.........

I say again, when AMD had something worth a damned, they gained market share, when they didn't, according to everyone but AMD Fanboys, they DIDn't! As has been said many times, even when they had great a processor, they still had crappy VIA and early nVidia boards that sucked as well.

I said Anand does more Desktop tests because that's who he caters to. That doesn't mean he'll never run servers against each other.

Anand and Doctor Tom admitted to turning a blind eye to AMD's short comings.

---

Sorry it's taken a while to respond.

I disagree with you.

From what I have been able to find (since I don't have an account with Gartner), Intel has anywhere between 75-87% of the server market. If, according to ZDNet

"AMD has been picking away at Intel's server market share for several years based on the superior performance and power consumption of its Opteron processor. But Intel fired back last month with a new Xeon processor based on its Core microarchitecture that appears to be outperforming current Opteron processors on several tasks."

Then why hasn't AMD already taken the majority market share in the server sgement?

Not to mention that there are many CIO/CTO's and other leaders/managers for IT departments where the person who assigned the budget may not be as technically competent. So, what's your explanation/answer to those who determine the annual budget for IT resources/departments? Are you to say that they're just as equally competent in computers and technology as the department that they're allocating the budget for? (Go ahead, I dare you, ask any of the Fortune 500 CFO's if they've even heard of AMD; and if they have, if they're considered using it within the company, and if they do, what percentage of the systems are AMD, and if not, why not.)

One would logically conclude that if there was a processor that was performing well (yes, admitted, operative word "was"), and was more efficient and competively priced; then wouldn't it stand to reason that people, and thereby companies, would be all over it???

Yet, "history" as most would call it, would show that AMD only had small gains in server market share. So, what's your explanation for this illogical result/reaction?

Consumers, yes, I would agree with you would (in your analogy) buy a Honda over Ford if Ford sucks (although I'd dare you to ask any Ford employee that question!), but IT departments often don't think like consumers do.

If that were the case, then why is it that according to the 11/08 Top500 list that 75.8% of the systems are using Intel EM64T processors and 87.5% of all quad-core systems?

Yet, if you analyze only the Top50, 40% of the processors are IBM POWER, 32% are AMD, and only 28% are Intel.

So, how do you explain such discrepancies that prove contrary to your argument?

I don't think that I've ever seen a VIA chipset show up in the server/HPC world. (nVidia possible, but that's the nVidia 2000 and 3000 series chipsets). Other integrators/implementations are all custom ICs and chipsets. Don't forget that there's also the AMD 8000 series chipset as well.

And if AMD had such a crappy processor (e.g. Barcelona Rev.B2, with the TLB errata, such that they stopped production entirely, which also BTW, was known to be only under a specific HPC condition), why did TACC order a system with 15,000 processors, and 45,000 processors for the ORNL Cray Jaguar?

(That doesn't include the IBM Roadrunner, which is #1 right now on the Top500 list; albeit they're only using dual-core Opterons and they're not compute processors, but I/O processors.)

Anand (et. al.) haven't been the only people that are calling for Core i7 to be put to the test for server and HPC apps. Many "normal" desktop computer users (ethusiasts and "regular Joe") have been asking for it as well.

And as I clearly stated, you can't really or reliably extrapolate the server (Nehalem Xeon) performance based on the Core i7. (Because the Core i7 doesn't have multi-socket platform and I don't think that it can support as much memory as servers do (either physically or practically).

Sources:
http://www.edn.com/article/CA6553931.html
http://www.xbitlabs.com/news/cpu/dis...rocessors.html
http://news.zdnet.com/2100-9584_22-149047.html
http://www.top500.org/lists/2008/11/highlights

Quote:

---

Originally Posted by Chad Boga

I'm not sure if this is the sort of thing you are referring to or not, but here a Dual Socket(8 cores) 2.93Ghz Nehalem with 48gig of RAM is able to set a SAP-SD benchmark of 5100 users.

For comparison, a Quad Socket(16 cores) 2.7Ghz Shanghai with 64gig of RAM falls short of the Nehalem score with a benchmark result of 4386 users.

http://www.sap.com/solutions/benchmark/sd2tier.epx

---

Just so you know, your intepretation of the results is inaccurate/incorrect.

The actual benchmark results is SAPS.

And there are a number of things that I find a bit interesting about this benchmark result (both of them done by IBM, albeit at different facilities).

For the Nehalem Xeon (X5570), it's 8.5% faster in clock speed compared to the Opteron 8384, but the Opteron system has more memory than the Xeon.

According to SAP, a (benchmark) user is defined as "A simulated user that runs the specific dialog benchmark scenario in a loop, repeating the actions with a think time of 10 seconds between each user interaction.".

In terms of SAPS/user, the Opteron system is 0.2% faster at 5.015 SAPS/user while the Xeon is at 5.005 SAPS/user.

Also another interesting thing is that the CPU utilization of the Opteron system wasn't as high as the Xeon (95% vs. 99%), which would tend to indicate that either a) it could have supported more users (I haven't been able to find out HOW they get that number actually), or b) that the Opteron is completing the transactions/requests/operations for the benchmark in less than one full second.

MY guess is that either it means that IBM wanted to achieve similiar SAPS by reducing the number of users (which seems likely given the "odd" number), and that the slight performance advantage (0.2%, and a 95% CPU utilization) could be brought on by the memory (architecture and amount).

IF the SAPS/user is correct, then for the Opteron system, my projected SAPS for 5100 users would be 25581 which would only be 51 points higher than the Xeon.

It's interesting.

(I will also like to add however, that given that one's a 2P/8C/16T configuration and the other's a 4P/16C/16T configuration, that the initial cost might be able to the same for both systems, and that I'd also expect the power consumption to be about the same as well.)

So, to those that are saying/thinking that the Core i7/Nehalem is an Opteron killer...at least for this benchmark...not quite.

(There's more to the benchmark than just the SAPS number. Because if it were ONLY just the SAPS number, than the Sun/Fujitsu-Siemens SPARC Enterprise Server M9000 is your baby at 195,000+ SAPS. THAT would blow anything/everything out of the water with 64P/256C/512T and 1 TB of RAM.) And even then, it's CPU utilization is only a measily 67%.

Quote:

---

Originally Posted by alpha754293

(I will also like to add however, that given that one's a 2P/8C/16T configuration and the other's a 4P/16C/16T configuration, that the initial cost might be able to the same for both systems, and that I'd also expect the power consumption to be about the same as well.)

So, to those that are saying/thinking that the Core i7/Nehalem is an Opteron killer...at least for this benchmark...not quite.

---

Wouldn't the Nehalem 2P system likely sell for a lot less than the Shanghai 4P system?

Quote:

---

(There's more to the benchmark than just the SAPS number. Because if it were ONLY just the SAPS number, than the Sun/Fujitsu-Siemens SPARC Enterprise Server M9000 is your baby at 195,000+ SAPS. THAT would blow anything/everything out of the water with 64P/256C/512T and 1 TB of RAM.) And even then, it's CPU utilization is only a measily 67%.

---

I suspect that the Nehalem 2P system will be a direct rival to the Shanghai 4P system and be a more mainstream deployment, whereas a system like this is for a completely different category of customer(more niche like by comparison to 4P and below) and would be taking on POWER and Itanium.

Quote:

---

Originally Posted by Chad Boga

Wouldn't the Nehalem 2P system likely sell for a lot less than the Shanghai 4P system?



I suspect that the Nehalem 2P system will be a direct rival to the Shanghai 4P system and be a more mainstream deployment, whereas a system like this is for a completely different category of customer(more niche like by comparison to 4P and below) and would be taking on POWER and Itanium.

---

Not necessarily.

The platform is expensive. Expensive processors. Expensive boards. Expensive RAM. (And yes, like anything it'll get cheaper over time, but for now...)

When you add everything up, you'd think that a 4P system would be more than a 2P system, but because they're so new, it's going to be pretty much about the same.

Nehalem 2P would be able to enter into the 1U and 2U only markets. Mostly challenging the 16-core Shanghai with 8-core workloads on the premise that HTT will work.

For anything that's computationally intensive, Shanghai will win out in the long run otherwise you'd have to OC the Xeon X5570 at least 50% in order to come close to the 16-core Shanghai 4P system's level of FPU performance.

Swap the Shanghais for faster CPUs and you're going to run out of room to OC the Xeon (and that's on the presumption that you can OC it to begin with, which in all reality, you probably can't).

The other really interesting thing is that Shanghai was a drop-in replacement (as demonstrated by Cray's XT4 Jaguar system). You can't do that with Nehalem.

I haven't done the calculations/analysis yet, but I know for certain that with Shanghais now...you can hit 100 MFLOPS/W efficiency. I'm not entirely sure you can do that with X5570 yet.

Quote:

---

Originally Posted by alpha754293

Not necessarily.

The platform is expensive. Expensive processors. Expensive boards. Expensive RAM. (And yes, like anything it'll get cheaper over time, but for now...)

---

Whilst that is all true on the desktop, are the platform costs in the Server market for Nehalem, a significant increase over Intel's Penryn derived Server products?

Quote:

---

Originally Posted by Chad Boga

Whilst that is all true on the desktop, are the platform costs in the Server market for Nehalem, a significant increase over Intel's Penryn derived Server products?

---

What are Penryn derived server products? From what I've been able to find, Penryn isn't even used in the desktops (laptops only), so unless I'm missing something...

Even if there were, DDR3, plus the whole tri-/triple-channel memory (which from what I've been able to find out means that you have to install the memory in triplicates) would add a fair bit of cost to it.

I couldn't even find a motherboard manufacturer or a chipset (anywhere) that supports the Nehalem Xeons (prolly pending NDA release).

The processor itself, from what I've been able to find starts at around $1300. (x2). That's $2600.

Kingston KVR 1066 3x2GB kit ECC Reg. (which would be the minimum presumed requirements for the server) is $458 a pop. 12 GB of RAM would be $916 (closest to 16 GB without being over.)

So, between the processors and the memory alone, that's nearly $3600.

4 AMD 8378 are $1200 a pop ($4800). 16 GB of DDR2-667 FBDIMM is $128 a pop ($512). So that's $5300 between procs and mem.

BUT...that's JUST procs and mem.

When you add the motherboard, chassis, hard drives, power supply and all to it; there's still potential for it to all work out to be about the same.

And at least with AMDs right now, you can get a Shanghai barebone system starting from $1422 (including a 1 kW PSU).

So, without really trying, you can probably get a 4P Shanghai system up and running for around $7000.

I'd expect that the bare minimal Nehalem configuration to be at least about that, if not slightly more.

And of course, the higher up you go in specs, the more expensive it will be; but I think for the time being, AMD is only SLIGHTLY better in the sense that in terms of TCO; the Socket F was planned all along for this kind of upgrade/adaptation.

(Although there really isn't any reason why a system that isn't certified to run with the Shanghai cores wouldn't work; other than it won't be able to take advantage of the core-by-core throttling that's in these new processors.)

I'd suspect that at least in theory, if you're someone who constantly keeps the system at 100% load, than any of the older Socket F barebones will work too.

An update to help those still in denial.

================================================== =======

Nehalem Xeon EP update: too good but true
http://www.anandtech.com/weblog/showpost.aspx?i=554

We were quite amazed, even slightly suspicious, when HP and Fujitsu-Siemens Published their SAP numbers. These numbers showed that the newest Xeon X5570 (Nehalem EP) series offer an enormous performance boost over the Xeon X5470 (Harpertown). After all, an almost 100% improvement at a slightly lower speed (2.93 GHz vs 3.3 GHz) is nothing short of amazing. Turns out that the real clockspeed is 3.2 GHz (2.93 GHz + 266 MHz turbo) but that does not alter the fact that these are truly incredible performance numbers.

I can now confirm that there are no tricks behind these numbers: they paint the right picture about the Xeon Nehalem EP. Talking to SAP benchmarking specialists, it became clear that few tuning tricks exist that are not know to the big OEM. The benchmark has been analyzed and tuned so well, that even the use of a different database (for example MS SQL instead of DB2) only makes a 2 to 3% difference most of the time. So you might even compare SAP numbers which are obtained on different databases. To resume, the SAP numbers can only be really boosted by better hardware (CPU-memory).

Now why I am talking so much about SAP benchmarking numbers? It is not like the expensive ERP software is run by everyone.

Well, the SAP numbers are showing a dual 2.93 GHz (or 3.2 GHz) Xeon beating the only quad AMD 8384 (Shanghai at 2.7 GHz) score of 22000 we have so far. Granted, a blade server is most of the time a bit slower. But four AMD 8384 2.7 GHz will be in the same league as a dual Xeon X5570, which will be out very soon now.

Even worse for AMD is that the SAP benchmark is not some exotic exceptional benchmarking case for the Xeon 55xx series. It shall be no surprise that the HPC numbers will be very impressive too. So it looks like AMD is in a tough spot.
.
.
The lack of any form of multi-threading is hurting AMD badly. It is well known that most of these business applications achieve very low IPC (0.2-0.6) and that modern superscalar CPUs have ample execution resources for running two threads in these applications. The results is Simultaneous Multi Threading offers a typical 20 tot 40% performance advantage. And that is huge, considering that you need 25 to 50% more clockspeed to counter that. It is basically a mission impossible for a modern CPU without SMT to outperform a similar superscalar CPU with SMT in OLTP, Java, webserver, rendering and ERP workloads. AMD really dropped the ball there, SMT should have been part of the K10 architecture.

Difficult times ahead for AMD

Even if AMD is able to speed up beyond 3 GHz, chances are slim that AMD will be able to compete with the new Nehalem Xeons. Add Turbo mode, hyperthreading, a lower latency memory controller and a better integer core together and you get a performance gap the size of the "Grand Canyon".

So does AMD have any chance at all beyond a new architecture in 2011? Is it over and out for AMD in 2009 and 2010? Adding 2 cores at the end of 2009 is a good step in the right direction. But even if AMD executes flawlessly the 32 nm Xeon Westmere will only give a window of a few months to the AMD hexacore "Istanbul". Istanbul should appear at the end of 2009, the Westmere Xeon is scheduled for very early 2010.
.
.
It is sure that we'll see better (lower) virtualization switching from virtual machine to hypervisor time and some small tweaks in AMD's Istanbul CPU, but it remains unclear if there are any significant performance boosters in the core. So it looks like Intel will own the dual socket space throughout 2009 and 2010, if we may believe the current roadmaps.

As the SAP numbers indicate, even the slowest Intel Xeons will show a large performance gap with the best AMD Opteron's.

Is AMD doomed completely?

In a large part of the market, yes.

AMD's istanbul will make the gap a bit smaller but probably not small enough.

There are some unknown factors that together with one of the few remaining weaknesses (or rather less strong points) of Nehalem that might make it possible that AMD's opteron comes close enough in a particular area of the market. In my next post, I will clarify the one and only opportunity that I see for AMD in the next two years. Until then, don't shoot the messenger :-).

It's looking quite gloomy for AMD, and frankly, all enthusiasts for the sake of innovation and price.

Quote:

---

Originally Posted by alpha754293

Just so you know, your intepretation of the results is inaccurate/incorrect.

The actual benchmark results is SAPS.

And there are a number of things that I find a bit interesting about this benchmark result (both of them done by IBM, albeit at different facilities).

For the Nehalem Xeon (X5570), it's 8.5% faster in clock speed compared to the Opteron 8384, but the Opteron system has more memory than the Xeon.

According to SAP, a (benchmark) user is defined as "A simulated user that runs the specific dialog benchmark scenario in a loop, repeating the actions with a think time of 10 seconds between each user interaction.".

In terms of SAPS/user, the Opteron system is 0.2% faster at 5.015 SAPS/user while the Xeon is at 5.005 SAPS/user.

Also another interesting thing is that the CPU utilization of the Opteron system wasn't as high as the Xeon (95% vs. 99%), which would tend to indicate that either a) it could have supported more users (I haven't been able to find out HOW they get that number actually), or b) that the Opteron is completing the transactions/requests/operations for the benchmark in less than one full second.

MY guess is that either it means that IBM wanted to achieve similiar SAPS by reducing the number of users (which seems likely given the "odd" number), and that the slight performance advantage (0.2%, and a 95% CPU utilization) could be brought on by the memory (architecture and amount).

IF the SAPS/user is correct, then for the Opteron system, my projected SAPS for 5100 users would be 25581 which would only be 51 points higher than the Xeon.

It's interesting.

(I will also like to add however, that given that one's a 2P/8C/16T configuration and the other's a 4P/16C/16T configuration, that the initial cost might be able to the same for both systems, and that I'd also expect the power consumption to be about the same as well.)

So, to those that are saying/thinking that the Core i7/Nehalem is an Opteron killer...at least for this benchmark...not quite.

(There's more to the benchmark than just the SAPS number. Because if it were ONLY just the SAPS number, than the Sun/Fujitsu-Siemens SPARC Enterprise Server M9000 is your baby at 195,000+ SAPS. THAT would blow anything/everything out of the water with 64P/256C/512T and 1 TB of RAM.) And even then, it's CPU utilization is only a measily 67%.

---

The Sun M9000 is a beast. x64 systems can only dream of such greatness. ;)