The Best Server CPUs part 2: the Intel "Nehalem" Xeon X5570
by Johan De Gelas on March 30, 2009 3:00 PM EST- Posted in
- IT Computing
ERP benchmark 1: SAP SD
The SAP SD (sales and distribution, 2-tier internet configuration) benchmark is an interesting benchmark as it is a real world client-server application. We decided to take a look at SAP's benchmark database. The results below are 2-tier benchmarks, so the database and the underlying OS can make a difference. It is best to keep those parameters the same, although the type of database (Oracle, MS SQL server, MaxDb or DB2) only makes a small difference. The results below all run on Windows 2003 Enterprise Edition and MS SQL Server 2005 database (both 64-bit). Every "2-tier Sales & Distribution" benchmark was performed on SAP's "ERP release 2005".
In our previous server oriented article, we summed up a rough profile of SAP S&D:
- Very parallel resulting in excellent scaling
- Low to medium IPC, mostly due to "branchy" code
- Somewhat limited by memory bandwidth
- Likes large caches (memory latency!)
- Very sensitive to sync ("cache coherency") latency
(*) Estimate based on Intel's internal testing
If you focus on the cores only, the differences between the Xeon 55xx "Nehalem" and the previous generation Xeon 54xx "Harpertown" and Xeon 53xx "Clovertown" is relatively small. The enormous differences in SAP scores are solely a result of Hyper-Threading, the "uncore", and the NUMA platform. According to SAP benchmark specialist Tuan Bui (Intel), enabling Hyper-Threading accounts for a 31% performance boost. Using somewhat higher clocked DDR3 (1066 instead of 800 or 1333 instead of 1066) is good for another 2-3%. Enabling the prefetcher provides another 3% and the Turbo mode increased performance by almost 5%. As this SAP benchmark scales almost perfectly with clock speed, that means that the Xeon X5570 2.93GHz was in fact running at 3.07GHz on average.
Consider the following facts:
- The quad-core AMD Opteron 8384 at 2.7GHz has no problem beating the higher clocked 5470 at 3.3GHz.
- It is well known that the Xeon 54xx raw integer power is a lot higher than any of the Opterons (just take a look at SPECint2006).
- Faster memory and thus bandwidth plays only a minor role in the SAP benchmark.
- SAP threads share a lot of data (as is typical for these kind of database driven applications).
It is clear that synchronization (between L2 caches) that happens in the L3 cache, the fast inter-CPU synchronization that happens via dedicated interconnects, is what made the "native quad-cores" of AMD winners in this benchmark. Slow cache synchronization is probably the main reason why the integer crunching power hidden deep inside the "Harpertown" cores did not result in better performance.
Take the same (slightly improved) core and give it the right (L3 as quick syncing point for the L2s) cache architecture and NUMA platform with fast CPU interconnects and all that integer power is unleashed. The result is the Nehalem X5570 Xeon is clock for clock about 66% faster than its predecessor (19000 vs. 11420). Add SMT (Simultaneous Multi-Threading) and you allow the integer core to process a second thread when it is bogged down by one of those pesky branches. The last hurdle for supreme SAP performance is taken: The eight core "Nehalem" server is just as fast as a 24 core "Dunnington" and 80% faster than the competition.
AMD has just launched the Opteron 2389 at 2.9GHz. We estimate that this will bring AMD's best SAP score to about 14800, so Nehalem's advantage will be lowered to ~70%. Unfortunately for AMD, that is still a very large advantage!
Facebook
Twitter
RSS
44 Comments
View All Comments
rkchary - Tuesday, June 16, 2009 - link
We've a customer who is interested in upgrading to Nehalem. He's running on Windows with Oracle database for SAP Enterprise Portals.Could you kindly let us know your recommendations please?
The approximate concurrent users would be around 3000 Portal users.
Keenly looking forward for your response and if you could state any instances of Nehalem installed in SAP environment for production usage, that would be a great deal of help.
Regards,
Chary
Adun - Thursday, April 9, 2009 - link
Hello,I understand the PHP not-enough-threads explanation as to why Dual X5570 doesn't scale up.
But, can anyone please explain why when you add another AMD Opteron 2384 the increase is from 42.9 to 63.9, while when you add another Xeon X5570 there isn't such an increase?
Thank you for the article,
Adun.
stimudent - Thursday, April 2, 2009 - link
Was it really too much effort to clean off the processor before posting a picture of it? Or were they trying to show that it was used, tested?LizVD - Friday, April 3, 2009 - link
Would you perhaps like us to draw a smiley face on it as well? ;-)GazzaF - Wednesday, April 1, 2009 - link
Well done on an excellent review using as many real-world tests as possible. The VMWare test is a real eye opener and shows how the 55xx can match double the number of CPUs from the last generation of Xeons *AND* crucially save $$$$ on licensing from Windows and MS SQL and other per-socket licensed software, plus the power saving which is again a financial saving if you hire rack space in a datacentre.I eagerly await your own in-house VM tests. Please consider also testing using Windows 2008 Hyper-V which I think doesn't have the 55xx optimisations that the latest release of VMWare has (and might not have until R2?).
Thanks for the time you put in to running the endless tests. The results make a brilliant business case for anyone wanting to upgrade their servers. You must have had the chips a good week before Intel officially launched them. :-) I do feel sorry for AMD though. I'm sure they have plenty of motivation to come back with a vengeance like they did a few years ago.
JohanAnandtech - Thursday, April 2, 2009 - link
Thanks! Good to hear from another professional. I believe the current Hyper Beta R2 already has some form of support for EPT.Our virtualization testing is well under way. I'll give an update soon on our blog page.
Lifted - Wednesday, April 1, 2009 - link
You mention octal servers from Sun and HP for VM's, but does anybody really use these systems for VM's? I can't imagine why anybody would, since you are paying a serious premium for 8 sockets vs. 2 x 4 socket servers, or even 4 x 2 socket servers. Then the redundancy options are much lower when running only a few 8 socket servers vs many 2 or 4 socket servers when utilizing v-motion, and the expansion options are obviously far less w/ NIC's and HBA's. From what I've seen, most 8 socket systems are for DB's.Veteran - Wednesday, April 1, 2009 - link
What i mentioned after reading the review is there are very few benches on benchmarks a little bit favored by AMD.For example, only 1 3DSmax test (so unusefull) at least 2 are needed
Only 1 virtualization benchmark, which is really a shame....
Virtualization is becoming so important and you guys only throw in one test?
Besides that, the review feels a bit biased towards intel, but i will check some other reviews of the xeon 5570
duploxxx - Wednesday, April 1, 2009 - link
Virtualization benchmark come from the official Vmmark scores.However there is something real strange going on in the results...
HP HP ProLiant DL370 G6
VMware ESX Build #148783 VMmark v1.1
23.96@16tiles
View Disclosure 2 sockets
8 total cores
16 total threads 03/30/09
Dell Dell PowerEdge R710
VMware ESX Build #150817 VMmark v1.1
23.55@16tiles
View Disclosure 2 sockets
8 total cores
16 total threads 03/30/09
Inspur Inspur NF5280
VMware ESX Build #148592 VMmark v1.1
23.45@17tiles
View Disclosure 2 sockets
8 total cores
16 total threads 03/30/09
Intel Intel Supermicro 6026-NTR+
VMware ESX v3.5.0 Update 4 VMmark v1.1
14.22@10 tiles
View Disclosure 2 sockets
8 total cores
16 total threads 03/30/09
So lets see all the prebuilds of esx3.5 update 4 get a real high score of 16 tiles almost as much as a 4s shanghai while Vmware performance team themselves stated that we should never see the HT core as a real cpu in Vmware (even with the new code for HT) while yet the benchmark shows a high performance increase, no not like anandtech is stating that this is due to the more available memory and its bandwith, those Vmmarks are not memory starving. Now look at the official Intel benchmark with ESX update 4, it provides 10 tiles and a healthy increase, that from a technical point of view seems much more realistic. All other marketing stuff like switching time etc, all nice, but then again is within the same line of current shanghai.
JohanAnandtech - Wednesday, April 1, 2009 - link
What kind of tests are you looking for? The techreport guys have a lot of HPC tests, we are focusing on the business apps."very few benches on benchmarks a little bit favored by AMD."
That is a really weird statement. First of all, what is a test favored by AMD?
Secondly, this new kind of testing with OLTP/OLAP testing was introduced in the Shanghai review. And it really showed IMHO that there was a completely wrong perception about harpertown vs Shanghai. Because Shanghai won in the tests that mattered the most to the market. While many tests (inclusive those of Intels) were emphasizing purely CPU intensive stuff like Blackscholes, rendering and HPC tests. But that is a very small percentage of the market, and that created the impression that Intel was on average faster, but that was absolutely not the case.
"Only 1 virtualization benchmark, which is really a shame..."
Repeat that again in a few weeks :-). We have just succesfully concluded our testing on Nehalem.
Personally I am a bit shocked about the "not enough tests" :-). Because any professional knows how hard these OLTP/OLAP tests are to set up and how much time they take. But they might not appeal to the enthousiast, I am not sure.