[magjohan]

Hi all,

With respect to the three following specifications:
How does the amount of cores influence simulation time and overall performance?

Which specification would you choose and why?

[flotus1]

Reading some of the threads in this subforum you will see that it is generally accepted that memory bandwith is what counts for CFD performance.
With this in mind, option 1 can be ruled out because the single processor only has 4 memory channels and 4 of the 8 DIMMs in the setup are wasted (in terms of bandwith).
So if you can afford it go for a two-processor setup. Which one depends on what you can afford and how many licenses you have available. But if your budget is limited, I would rather choose number 2 and spend the rest of the money on some fast SSDs. A few additional cores wont make much of a difference when the memory bandwith is capped out.

Some additional information on the software you are using and the typical model size might be helpful.

[magjohan]

Thx guys for your input.

In this case, all systems has an Ssd drive and a standard 2TB - difference is the CPU.

So the clock speed in this case won't influence simulation time that much compared to number of cores? The CPU in option 1 with its 10 cores is a monster according to cpubenchmark.net. Google it and you'll find a lot of info about it and it's performance.

Is this the final answer, 2cpus are always better in these 3 options?

[magjohan]

Originally Posted by flotus1

Reading some of the threads in this subforum you will see that it is generally accepted that memory bandwith is what counts for CFD performance.
With this in mind, option 1 can be ruled out because the single processor only has 4 memory channels and 4 of the 8 DIMMs in the setup are wasted (in terms of bandwith).
So if you can afford it go for a two-processor setup. Which one depends on what you can afford and how many licenses you have available. But if your budget is limited, I would rather choose number 2 and spend the rest of the money on some fast SSDs. A few additional cores wont make much of a difference when the memory bandwith is capped out.

Some additional information on the software you are using and the typical model size might be helpful.

Fluent/CFX/openfoam is the main CFD software thatll be used!

[Hrushi]

Specification with dual Xeon processor is better than first option in terms of memory bandwidth.

Amongst Spec2 and Spec 3, Spec 3 with E5-2660v3 seems to be superior based on memory bandwidth. E5-2640 v3 has got 59 GB/s whereas E5-2660 has got memory bandwidth of 68 GB/s. Also E5-2660 v3 (9.4 GT/s) has got higher OPI speed than E5-2640 v3 (8 GT/s).

For both Spec2 and Spec3, cache per core works out to be same. Also there is not much difference in terms of clock speed for E5-2640 v3 and E5-2660 v3.

[Eike]

After using a Dell T3500 w/ 6-core CPU and getting slower calculations using more than 3 cores I switched to a Dell T7500 w/ 2x 4-core CPU. Now I can use all cores :-)

STAR-CCM+, btw.

Btw, by now we like to build custom 64GB Desktop PCs w/ Intel 4-core CPUs and oc them to 5 GHz (air cooled). You are getting pretty fast Workstations on a Budget that way ...