workstation for CFD
 

Hello

I have a question about workstation's performance for CFD. Which of the following solutions do you think is the best one :
- 2 Xeon x 6 cores @ 3.70 Ghz / 64Gb RAM
- 2 Xeon x 8 cores @ 3.40 Ghz / 64Gb RAM
- 2 Xeon x 10 cores @ 3.00Ghz / 64Gb RAM
- 2 Xeon x 12 cores @ 2.70 Ghz / 64Gb RAM

This is for aero/thermal model with cfx, typical model size ranging from 10 to 20 millions elements.
If you have any advice relating to that topic do not hesitate to share please !

Thanks in advance

Greetings Xeon and welcome to the forum!

Can you specify which specific CPU models you're referring to?
Because there is a particular detail about the current Intel CPUs: when not using HyperThreading, they can go up a bit on the clock frequency, because the stock speed is for when using all cores with HT turned on at full throttle.

For example, an L5640 has stock speed of 2.26 GHz, but a max speed of 2.8 GHz. This means that if HT is turned off, it should be able to handle the mid range speed, namely (2.26+2.8)/2 = ~2.53 GHz.

The other detail to keep in mind is which memory speed you'll be using and how many modules per CPU socket.

Best regards,
Bruno

I'm guessing he is talking about new IVY-Bridge-E XEONs, (E5-v2), and will turn hyper-threading off.
What kind of HPC licenses do you have access to?
CFX is very memory bandwidth intensive, and is not going to scale well across 24 cores with only 8 memory channels. You may be better off going with multiple machines with lower core counts.
When you say "best" that could mean a lot. The dual dodeca core machine would be the "best" with an unlimited number of HPC licenses, but would also be the most inefficient use of those licenses. It may also be only slightly faster than a dual hexa, or even a dual quad-core machine. I've read getting dual E5-2643v2 quad cores would be the best choice, as it will be the most efficient use of cores.
make sure you get the fastest memory you can, which I believe is 1866 MHz now on the server boards, that will be more important than core count.

Hi, thanks for your replies.

Actually I was wondering what is the optimal configuration between high number of CPU and high frequency of proc..and by "best" I meant shortest time to run a simulation !
I've seen some benchmark comparing Xeon and last I7 ..now I may be interested by the i7 but I wonder if one could have 2 i7, must find the right motherboard..what do you think of i7 vs Xeon ?

Regards

You can't have two i7s on one motherboard. Only XEONs are capable of that. You would have to use some sort of network connection (gigE, Infiniband) to distribute to multiple i7 machines.
You still haven't answered my first question about HPC Licenses. How many processors can you run? If only 1 in serial, then a 24 core machine won't be a wise investment.

We plan to have 3 HPC Packs so I could potentially run a calculation on 128 cores but I think 64 cores will be enough just for me.

Wow, 3 HPC packs, That sure is nice. Well it sounds like you will have to set up a cluster to use all those processors. Is that what you are interested in? (making a cluster for 32, 64, or 128 core parallel?) Or are you just going to have a few stand-alone machines?

a cluster or stand alone machines..., don't know yet !