[EternalSeekerX]

Hello everyone,

Has the consensus on cfd and cfd like performance on these big Little x86_64 processors changed? It seems mainstream intel and now AMD is moving towards a performance and efficient core chiplet design. Desktop AMD still uses the full performance core, but with strix point apu in mobile, they may move to something similar on Desktop (like how intel does it).

From anecdotal evidence, Forums posts, benchmarks and even videos like this:

https://youtu.be/tgzh_O5UI5o?feature=shared

It seems that cfd struggles a bit on these architecture. This is most likely due to cfd being able to leverage multiple full cpu cores, and is reliant on memory bandwidth and also cache. It is very sensitive to inter-core communication and any delay. It seems it's the same for any cfd solver (fluent, openfoam, su2, starccm+, comsol, power/Xflow, etc.).

Most of these p-core and e-core (or equivalent zen and zen-c cores for AMD), have clock rate, cache and memory interconnect differences. Using a mix of cores causes the higher speed cores to function at slower clocks or adds delay for inter-CCD communications. This effectively means for cfd that these cpu are best performant under only p-cores. So the shiny 16/24 core intel is only really 6/8 cores, while strixpoint are really just 4 core systems.

I know heavy cfd is done using server cpu that don't have this problem yet, but for consumer desktop that can be used for heavy cfd, this sucks if this remains true. Even more so for mobile workstation for students to.

It seems this type of cpu are best for tasks that just ignore intercommunication or don't need to talk to cache to much. I don't know if this also effects visualization of cfd simulation too.