Why One But Not The Other?

[BartJY]

Hi Folks,

The GPU is churning away doing computations for the cause. But the CPUs have not done anything, what's up with that?

Thanks
Bart

[Jesse_V]

There's been a 20x increase in F@h userbase and workunits are being consumed faster than they can be generated, which has led to the queue sometimes being empty. I'd recommend keeping everything running and the CPU should pick up work when its available. The research labs are rapidly trying to get new projects online and more servers into place to keep up with demand, so hopefully the situation can be fixed pretty soon.

[BartJY]

Also my cpu is a 6 core 12 thread verity. Will all of the threads be use separately or in combination for just one workunit?

Thanks
Bart

[Tuna_Ertemalp]

I'm also a new user, so don't know the exact technically correct answer to your question, but all threads on all my CPUs (some up to 32) are being used at 95-100% capacity, so I don't wonder "why" or on "what" exactly they are used. As long as they are used at capacity, it is a moot point if they are churning on one WU or N WUs.

Tuna

[Jesse_V]

Also my cpu is a 6 core 12 thread verity. Will all of the threads be use separately or in combination for just one workunit?

Thanks
Bart

A single workunit can use the entire CPU or part of it. The exact number of cores/threads used is set by the software and can be configured in the "Advanced Control" tool called FAHControl.

I'm also a new user, so don't know the exact technically correct answer to your question, but all threads on all my CPUs (some up to 32) are being used at 95-100% capacity, so I don't wonder "why" or on "what" exactly they are used. As long as they are used at capacity, it is a moot point if they are churning on one WU or N WUs.

Tuna

A workunit, a piece of the molecular simulation under study, can be divided across many different CPU cores and processed all at once that way. So a 32-core CPU can get the workunit done 32 times faster than a 1-core CPU. As a reward for this speed, there's a non-linear point system, so your PPD (points/day) is pretty huge for running a single workunit on 32 cores instead of 32 different workunits in parallel.