FAH underperforming on new system

[arisu]

I thought there was an effort in GROMACS to improve the load balancer so that it could distribute load effectively even among a mix of E- and P-cores.

The issue is not with Gromacs but with OS scheduler (mostly Windows one) that will assign all threads to E core because they run at low priority ...

And yes, allowing FAH to handle all combinations automatically will be a pain.

Yeah it's not a GROMACS issue, although in theory it could be solved within GROMACS (but not easily). Having the client assign threads to P cores would be better and like I said I do that on my systems, but I don't know Windows development so it's currently Linux-only.

[foxpy]

When I pin all threads but one to a single core, the remaining thread drops to around 70%. I believe that is because it exceeds the limit of the load balancer. I'll check the source code to make sure, but I don't think it is using spinlocks.

Maybe it doesn't use spinlocks indeed, but still something weird happens under the hood. 70% usage is quite high in this scenario.

[arisu]

I can only imagine that the 70% is due to it performing load balancing, but not well enough to achieve 100%.