Hi

I just observed, that when Rosetta fills all my cores and threads, only one of my GPUs is active.
Other projects leave enough CPU overhead allowing both GPUs to be active.

Can someone point me to the specific setting that may undo this bottlenecking?

What I already tried:
Reducing the % of CPUs active (Use at most 90% of the CPUs and and at most 90% of CPU time) for both "When Computer is / is not in use".

This just reduced the active CPU tasks for Rosetta, but still only one GPU active.

Best regards and thanks in advance

---

Hi

I just observed, that when Rosetta fills all my cores and threads, only one of my GPUs is active.
Other projects leave enough CPU overhead allowing both GPUs to be active.

Can someone point me to the specific setting that may undo this bottlenecking?

What I already tried:
Reducing the % of CPUs active (Use at most 90% of the CPUs and and at most 90% of CPU time) for both "When Computer is / is not in use".

This just reduced the active CPU tasks for Rosetta, but still only one GPU active.

Best regards and thanks in advance

Rosetta doesn't use GPUs at all, so it's highly unlikely it's the cause of any issues you may have, though other projects might.
If Rosetta is filling all your cores and threads atm you've been exceptionally lucky and I'm very jealous.

Setting "Use at most 90% of the CPUs" is a legitimate change if you think it might help (I have no idea).
"Use at most 90% of CPU time" does more harm than good in my experience. Rosetta works best with this always set to 100%
"at most 90% of CPU time" is implemented by running 100% for 90% of the time, then 0% for 10% of the time and rapidly switching between the two, not at 90% for 100% of the time as you might more reasonably expect.
You might want to try reducing the %age of CPUs used a little further to see if it makes any difference, though I can't see why it would when Rosetta never calls for any GPU time.

---

Can someone point me to the specific setting that may undo this bottlenecking?

Your best bet is to check the forums for the project that is using the GPU.
You will need an app_config.xml file placed in the

C:/ProgramData/BOINC/projects/ whatever project it is

folder (back slashes replaced with forward slashes as the forum software strips them out).


For Numberfields (depending on the GPU) i use

<app_config>
   <app>
      <name>GetDecics</name>
      <gpu_versions>
         <gpu_usage>0.33</gpu_usage>
         <cpu_usage>1.4</cpu_usage>
      </gpu_versions>
   </app>
</app_config>

GetDecics is the name of the application i want the settings to apply to (so you need to know the name of the application running on your GPUs to replace GetDecics).

0.33 for the GPU usage results in 3 Tasks running on the one GPU. If the application for your GPU project is well optimised, then you'd only want one Task to run which would be 1.0
CPU usage means i reserve 1.4 CPU cores per GPU Task that is running.

With 3 Tasks running, it reserves 4 CPU cores/threads. 3 of them support the GPU Tasks, the other is unused and available to the system.
If you're only running a single Task per GPU, then set that value to 1.0 as well.

That will reserve a CPU core to support the GPU Tasks that are running. If no GPU Tasks are running, then it is released back to any projects running Tasks on the CPU to use.
"Use at most xx % of the CPUs" and "Use at most xx % of CPU time" are best left at 100% in most cases.

---

Grant
Darwin NT