Processors
Intel® Processors, Tools, and Utilities
14405 Discussions

CPU Comparison for Numerical computing

idata
Employee
2,538 Views

Dear all,

We are working in my lab on numerical modeling for physical problems with Comsol software. Recently, we have bought a new Dell computing server, based on Intel Xeon E7-4820 v3 @ 1.90GHz, 10 cores (2 logical cores per physical). We are surprised to see with some benchmarks that this new server is almost twice slowest than another based on Intel Xeon E5-1620 v2 @ 3.7GHz, 4 cores (although prices are really different!). We do not think it is due to memory, as the simulation only uses about 10 Gb of RAM, far from the 64Gb of the server. Moreover, CPU benchmark on the internet show that the E74820 have largely higher performances thant E51620. Do you have any idea that could explain such result?

Thank you.

Best regards,

0 Kudos
10 Replies
AJana1
Valued Contributor I
1,145 Views

Hello mininger,

http://ark.intel.com/products/84677/Intel-Xeon-Processor-E7-4820-v3-25M-Cache-1_90-GHz?q=%20E7-4820%20v3 Intel® Xeon® Processor E7-4820 v3 (25M Cache, 1.90 GHz) Specifications

http://ark.intel.com/products/75779/Intel-Xeon-Processor-E5-1620-v2-10M-Cache-3_70-GHz?q=E5-1620%20v2 Intel® Xeon® Processor E5-1620 v2 (10M Cache, 3.70 GHz) Specifications

OK, one could get detail specification among those two processors from above links. The most possibly reason why the latter is twice faster than the first one per the results of some benchmark software is that it is not well optimised for multi-thread computing, or benchmark for single thread computing. As is known that they both have the similar underlying microarchitectures (Haswell vs Ivy Bridge), so for the single-threading computing, at the same working frequency, they both would perform more or less equal. But something around the processors would also put positive or negative force to the final benchmark scores (supporting components, O/S and so forth...)

You might use latest version of CPU-Z to make a comparison too! Well, I have to say that the size of cache might not always put the force towards benchmark results, and twice the same frequency processors would not make a twice-frequency(double pumped) processor. Allow me to be frank, those two processors are designed for different working environments, comparison of them both two on the same level would give nothing.

Best Regards,

Aaron Janagewen

0 Kudos
idata
Employee
1,145 Views

Thank you Aaron for your answer. In fact, the results I got with both processors were for my numerical simulations with Comsol Multiphysics with 2 calculation servers we have in the lab, not only benchmark. I need them to go as fast as possible, and the objective of the E74820 purchase was exactly to gain time :-/

We are working with Linux Debian for both servers, so I don't think it is an OS issue. I have thought at the single-thread/multi-thread problem, but with the "htop" linux command I can see that all (virtual) cores are used for the simulations (8 for the E5, 20 for the E7) ! In this case, I do not understand why the E7 (that moreover works with 2 SSD in RAID) is slower than the E5... and I do not know what I can do to find where is the problem, and how to resolve it :-(

When you say "those two processors are designed for different working environments", can you give more details? I am not sure to understand the sentence.

Best regards,

0 Kudos
AJana1
Valued Contributor I
1,145 Views

So many thanks for your reply. Those two (mentioned) processors are different, Xeon E7-4820 v3 is designed for business which needs heavy multi-task computing, in order to get closer to its ideal goal this processor is scalable up to 4 identical processors system. While Xeon E5-1620 v2 is designed for computing performance first field, such as science computing, the higher working frequency enables it to gain even more powerful computing power, but not that scalable. .

I am very sorry I have rare knowledge on Comsol Multiphysics. I suggest you use CPU-Z to make a benchmark of them both in the purpose to test your new processor working correctly or not, not in the purpose to solve your problem. Scientific Computing is a hard job, so you'd better contact with Comsol to gain further support information. And I have also declare that I just express my own opinion, whether E7-4820 is perfect for your computing, I have no ideas!

Best Regards,

Aaron Janagewen

0 Kudos
idata
Employee
1,145 Views

Hello mininger,

 

 

In regard to your request, at this point for you to be able to know more about the results of the numerical comparison that you are using, the best thing to do is to contact the developers of this software.

 

 

You can look more information for Comsol Support at the following link:

 

 

https://www.comsol.com/support https://www.comsol.com/support

 

 

Also, thank you Janagewen for your assistance.

 

 

Regards,

 

 

Angie.
0 Kudos
idata
Employee
1,145 Views

Thank you all for your replies. I will see with them if they can find a solution.

Best regards,

0 Kudos
idata
Employee
1,145 Views

Hello mininger,

 

 

Thank you for letting me know about this matter.

 

 

Regards,

 

Angie

 

0 Kudos
idata
Employee
1,145 Views

Hello mininger,

 

 

I would like to confirm if you were able to solve your concern?

 

 

Hope I can hear from you soon.

 

 

Regards,

 

 

Angie
0 Kudos
idata
Employee
1,145 Views

Solution suggested for such kind of computation: do not use a 20core CPU, but two 4core working in parallel, associated to fast memory bandwith. Link :

https://www.comsol.com/support/knowledgebase/866/ What hardware do you recommend for COMSOL Multiphysics? - 866 - Knowledge Base

0 Kudos
AJana1
Valued Contributor I
1,145 Views

After reading the information you provided, I found that computing models are based on the algorithms heavily optimised for four cores and/or two socket processors system. The memory size and bandwidth are also the key factors to increase the performance. If the system memory configurations on both platforms are the same or similar, the higher working frequency would gain the top performance. That might solve your problem.

Cores more than four, the performance would not buy it. And the lowever frequency could not be made up with more cores. But every coin has two sides, so my suggestion is to partition your new server with virtualisation software. Let your computings done within each vitualised machine, that might help you to fully utilise it.

0 Kudos
idata
Employee
1,145 Views

Hello mininger,

 

 

I am really glad to know that you were able to find the information that you need.

 

 

If you need further assistance with other concerns, do not hesitate to use our forums:

 

 

/news https://communities.intel.com/news

 

 

Regards,

 

Angie.
0 Kudos
Reply