Parallel processing with Intel performance and effciency cores?

Talder · May 16, 2024, 04:39

The OpenFOAM documentation always said parallel processing cannot make use of 'hyper-threading' capability of multicore processors. You only gain efficiency with 'true' performance cores. No benefit from hyper-threading functionality.

My own experience with an 8-core 16-thread processor (i9900K) confirms that.

Reconsidering new hardware in 2024, I couldn't find any information on how OpenFOAM handles more recent Intel multicore processors with a (non-symmetrical) mix of performance and efficiency cores. (Like 8 performance and 12 efficiency cores)

Can OpenFOAM parallel processing benefit from this?
Or any other hyper-threading functionality?

It keeps feeling like underusage of what the CPU is capable of.

Any thoughts/experience on this?

Best Regards,
Tom

Talder · May 16, 2024, 10:22

Thinking further about the question I found a partial answer:
The parallel processing tool used by OpenFOAM is obviously OpenMPI, so the question is how efficient OpenMPI can utilize recent hyperthreading options.
That seems to require more digging into OpenMP handles, and may depend on how OpenMP is compiled to handle "P" and "E" cores in the newer generation Intel CPU's.

Sorry for the crosspost, but it is obviously the same question (and rather about OpenMPI):

https://stackoverflow.com/questions/...on-new-machine

May 16, 2024, 04:39	Parallel processing with Intel performance and effciency cores?	#1
Talder New Member Tom Alderweireldt Join Date: Jan 2015 Location: Schilde, Belgium Posts: 16 Rep Power: 11	The OpenFOAM documentation always said parallel processing cannot make use of 'hyper-threading' capability of multicore processors. You only gain efficiency with 'true' performance cores. No benefit from hyper-threading functionality. My own experience with an 8-core 16-thread processor (i9900K) confirms that. Reconsidering new hardware in 2024, I couldn't find any information on how OpenFOAM handles more recent Intel multicore processors with a (non-symmetrical) mix of performance and efficiency cores. (Like 8 performance and 12 efficiency cores) Can OpenFOAM parallel processing benefit from this? Or any other hyper-threading functionality? It keeps feeling like underusage of what the CPU is capable of. Any thoughts/experience on this? Best Regards, Tom

May 16, 2024, 10:22		#2
Talder New Member Tom Alderweireldt Join Date: Jan 2015 Location: Schilde, Belgium Posts: 16 Rep Power: 11	Thinking further about the question I found a partial answer: The parallel processing tool used by OpenFOAM is obviously OpenMPI, so the question is how efficient OpenMPI can utilize recent hyperthreading options. That seems to require more digging into OpenMP handles, and may depend on how OpenMP is compiled to handle "P" and "E" cores in the newer generation Intel CPU's. Sorry for the crosspost, but it is obviously the same question (and rather about OpenMPI): https://stackoverflow.com/questions/...on-new-machine