Waste of CPU resources

freestman
freestman
Joined: 16 Jun 08
Posts: 33
Credit: 1973464002
RAC: 17436
Topic 205377

Whether NVIDIA or AMD, each GPU task must take up a CPU core or thread, resulting in the use of AMD CPU when nearly 90% of the waste of resources. According to the actual situation of the GPU set up a reasonable set of CPU requirements.

sig-1531.png

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109384399472
RAC: 35945454

You really do need 1 CPU core

You really do need 1 CPU core for each GPU task with nvidia.  Whilst the same defaults do apply to AMD, you can change those defaults with an app_config.xml file.  Assuming you wanted to run 2 GPU tasks with the support of just 1 CPU, here is the file you would need to use.  It needs to be named exactly app_config.xml and it needs to be placed in the Einstein project directory.  After placing the file you need to click 'read config files' in BOINC Manager advanced view.


<app_config>
  <app>
    <name>hsgamma_FGRPB1G</name>
    <gpu_versions>
      <gpu_usage>0.5</gpu_usage>
      <cpu_usage>0.5</cpu_usage>
    </gpu_versions>
  </app>
</app_config>

The values for gpu_usage and cpu_usage can be changed to whatever you need for optimum performance.  If you don't have at least one 'free' CPU core, GPU performance is likely to suffer.  When figuring out how many GPU tasks can be run simultaneously, you need to have ~1GB of GPU RAM per task.  Running more than 2 GPU tasks concurrently on the one GPU is not likely to give much further improvement and may actually be counter-productive in some situations.

Cheers,
Gary.

freestman
freestman
Joined: 16 Jun 08
Posts: 33
Credit: 1973464002
RAC: 17436

Thank you!

Thank you!

sig-1531.png

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.