Folding Forum

I do just about understand how hyperthreading works, my main confusion was as to why SMP 6 with a GPU slot running as well is folding faster than SMP 8 on its own? In this case surely with SMP 6 it would effectively have access to 3 cores with one core left to deal with the GPU slot and anything else that's using up resources. But when using SMP 8 it would effectively have access to all 4 cores and would then have to share a thread with any small jobs that the OS throws up. Is this correct? Maybe it's these small jobs that impede performance sufficiently enough to slow it down to a slower pace than SMP 6?

Actually, in Resource Monitor, when SMP 6 is used it shows that 7 threads are being used at an average of 80-90% load and another is being used at around 5% load. Is this normal? Have I completely misunderstood hyperthreading?

Thanks for the help.

With -smp 6, it's up to the scheduler in the OS to decide whether the two unused threads are on the same real CPU or on different ones. Chances are that all four real cores would be used, two running one thread and two running two each. How it assigns the work will change very, very rapidly so you won't see it happening unless you force it to work a particular way with affinity settings.

When running -smp 6 and nothing else, Task Monitor will probably show what looks like a random allocation between various virtual CPUs but at any instant, they add up to 75%. Mentally stacking the 8 graphs pairwise into four stacks will still give you random times when both half of a pair is working and times when only one half of the pair is working -- but it's way too fast to really perceive.

Okay thank you very much for the help! After all this I'm now going to try and move over to a VM with linux to fold in.

The help is greatly appreciated!