A simple multi-thread test in Ansys Fluent

Published by Fourthbus on

Recently, I have been extremely procrastinative in work and productive in other stuff. Therefore, you might see me switching topics from design to PC setting and back to PhD research in a row. You know a deadline is coming.

I did a brief and simple test on my new i7 12700 (8 p-cores + 4 e-cores, 2 thread per p-core, 1 per e-) to check for the optimum setting for Ansys Fluent. A long story of me fighting against the core affinity, I am not talking about it here. Basically, Windows Server 2022 allocates any process in the background to my e-core by default. This caused me to use 8 cores for web browsing and 4 lame cores for CFD!

Methodology

The methodology is simple, run the same test case with different multi-process settings in Workbench. I used three different values, 8, 12, and 16. This corresponds to 1x, 1.5x, and 2x my core number, remember there are two threads per core. I found 8 processes are not ideal, as it does not produce a consistent 100% CPU optimisation throughout time — each CPU pauses a while for saving flow data. It is a good idea to have a trial-and-error investigation on this before I get too busy is still a good idea.

A very ziggy-zaggy CUP utilisation, and god knows what it is doing.
CPU affinity and priority of Fluent processes.

By setting priority to ‘Real time’ I can ask them to take as much of my core as they want. And check the simulating time in Fluent for 100 time-steps in my transient flow simulation with command parallel>timer>usage. Then I run it again to take an averaged value.

Results and discussions

Here are what I have got, not much difference. 8 and 12 processed take, roughly, the same time, and using 16 processes is not an ideal plan.

Number of processesTime [s]
7 (proper fan, no hyper-thread)151
8168
8 (proper fan, hyper-thread)154
8 (proper fan, no hyper-thread)140
12168
16171
Simulation time for 100 time-steps in Ansys Fluent.

If an Intel fan is not a fan, then I don’t have a fan. Thermal throttle is there to limit performance, all cores can only run around 4.2 GHz. Furthermore, I won’t bring the e-cores into this game unless they solve the utilisation mystery — when the program run in the background, only e-cores are used. I can’t stare at my CFD forever to witness the oscillating pressure curve!

Acknowledgements

There is a deknowledgement (I antonymised acknowledgement) to Amazon as my fan is still not arriving.

(Updated 22 Feb) The fan has arrived. It’s a pain to install it, as the studs are bent with the motherboard and it was disgustingly unenjoyable to install this. This fan also has a specific orientation, so I also need to remember which side up…

The author un-thanks Amazon.

This test is partially funded by leveraged investments in natural gas and oil. Thanks for the gas-price crisis now I can pay the extra 0.1 kWh (2 p in Pound Sterling) for this test.