Changing resources can cause other WUs to restart

[Frogging101]

My observation

• Changing a nonzero CPU count to zero, or vice versa, will cause GPU units to restart. This wastes progress (potentially a lot of progress, depending on the WU)
• Disabling a GPU will cause running CPU WUs to stop and start again. This isn't too much of an issue because Gromacs can checkpoint whenever it wants to.
• Changing a nonzero CPU count to another nonzero value doesn't appear to disrupt anything.
• I don't have a multi-GPU system, but the behaviour with that case should be examined as well. Disruption to other WUs should be minimized.

Why this affects me

Sometimes I want to stop folding on either my CPU or GPU, to manage heat output or because I need that resource for something else. And in v8, since there are no slots, the only way to do this without pausing everything is to change the resource count.

In v8, we have also lost the ability to set the "finish" state per-resource, which I miss because I can no longer let my GPU finish its current WU and then go idle (i.e. I want it to pause, but would prefer that it finishes first). Ideally, I would like to see a replacement for v7's per-WU pause and finish states. But that is tangential to this issue.

Edit: It was pointed out that splitting the CPU/GPU into resource groups covers the above. Thanks.

cheers

[muziqaz]

V8 has Resource Groups, which are similar to Slots on v7

[Frogging101]

V8 has Resource Groups, which are similar to Slots on v7

Yes, I know :)

[muziqaz]

So half of the statements in your original post are not correct.
"Sometimes I want to stop folding on either my CPU or GPU" - yes you can
"in v8, we have also lost the ability to set the "finish" state per-resource" - this is incorrect
"And in v8, since there are no slots" - that is incorrect
Here is my Web UI:

As you can see each device has its own resource group, on each of the computers.
Changing CPU count mid folding is not recommended.

[Frogging101]

Oh, I see now. Thanks for that tip. I did not realize you could use resource groups that way.

I think it may still be a bug that changing the CPU or GPU count interrupts unrelated slots.

[muziqaz]

Yes, the first part of your comment is valid and possible bug:
Changing CPU number restarts GPU folding within the same resource group, or if there are no resource groups, within same computer