GPU-only folding Using 100% of a Core

[tarfeef101]

I'm trying to use the `fah-gpu` docker container to get some extra folding done. I managed to get it to run, use the GPU, allow connection to fahcontrol on my desktop, etc. That's all great. The issue I'm running into right now is that I only created 1 slot, for my GPU (which was intentional). However, it seems that 1 core is being pinned at 100%, despite me explicitly (trying to) configure it not to use the CPU. Anyone know how I can avoid this happening? I've included my `config.xml` below

 

<config>
  <!-- Folding Core -->
  <gpu-usage v='80'/>

  <!-- Folding Slot Configuration -->
  <disable-viz v='true'/>

  <!-- GUI -->
  <gui-enabled v='false'/>

  <!-- User Information -->
  <passkey v='REDACTED'/>
  <team v='223518'/>
  <user v='REDACTED'/>

  <!-- allow fahcontrol -->
  <command-allow-no-pass v='REDACTED'/>
  <allow v='REDACTED'/>

  <!-- disable cpu -->
  <cpu v='false'/>
  <cpu-usage v='10'/>

  <!-- Folding Slots -->
  <slot id='0' type='GPU'>
    <pci-bus v='8'/>
    <pci-slot v='0'/>
  </slot>
</config>

 

[CWP]

I remember reading somewhere that each nVidia graphics card will take up a CPU core. I do not recall the reason behind it though...

top - 16:31:41 up 6 days,  2:36,  1 user,  load average: 3.15, 3.23, 3.26
Tasks: 202 total,   4 running, 198 sleeping,   0 stopped,   0 zombie
%Cpu(s):  0.0 us,  0.1 sy, 37.6 ni, 62.2 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem :   7884.5 total,   3702.2 free,   2670.3 used,   1512.0 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.   4877.3 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
  18899 fah       39  19   31.4g 446920 285788 R 100.3   5.5  68:33.75 FahCore_22
  18970 fah       39  19   31.8g   1.0g 291488 R 100.0  12.7  43:55.92 FahCore_22
  19064 fah       39  19   31.4g 446012 285284 R 100.0   5.5   7:39.59 FahCore_22
  18376 fah       39  19   15.1g 798212 115596 S   1.7   9.9   4:09.03 FahCore_22

I give you one guess how many nVidia and how many AMD graphics cards are folding in this system .

[tarfeef101]

6 minutes ago, CWP said:

I remember reading somewhere that each nVidia graphics card will take up a CPU core. I do not recall the reason behind it though...

top - 16:31:41 up 6 days,  2:36,  1 user,  load average: 3.15, 3.23, 3.26
Tasks: 202 total,   4 running, 198 sleeping,   0 stopped,   0 zombie
%Cpu(s):  0.0 us,  0.1 sy, 37.6 ni, 62.2 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem :   7884.5 total,   3702.2 free,   2670.3 used,   1512.0 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.   4877.3 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
  18899 fah       39  19   31.4g 446920 285788 R 100.3   5.5  68:33.75 FahCore_22
  18970 fah       39  19   31.8g   1.0g 291488 R 100.0  12.7  43:55.92 FahCore_22
  19064 fah       39  19   31.4g 446012 285284 R 100.0   5.5   7:39.59 FahCore_22
  18376 fah       39  19   15.1g 798212 115596 S   1.7   9.9   4:09.03 FahCore_22

I give you one guess how many nVidia and how many AMD graphics cards are folding in this system .

Dang. that sucks, my CPU is much warmer than it needs to be now  and the fans are louder as a result. 

 

got any idea how to make those limit variables actually work? i set the CPU core limit to 10% and GPU to 80%, but both are pegged at 100% still

[tarfeef101]

Big oof, my server seems to be crashing after some time running this for whatever reason... worked fine overnight last night, it's been able to train neural nets, mine crypto... not sure what's causing the failures, and don't have time to investigate. guess it's sweater time til I can dig into this (actually can't turn on heating, so folding is my heat in the winter)

[LazyDev]

On 11/12/2020 at 12:41 AM, tarfeef101 said:

I'm trying to use the `fah-gpu` docker container to get some extra folding done. I managed to get it to run, use the GPU, allow connection to fahcontrol on my desktop, etc. That's all great. The issue I'm running into right now is that I only created 1 slot, for my GPU (which was intentional). However, it seems that 1 core is being pinned at 100%, despite me explicitly (trying to) configure it not to use the CPU. Anyone know how I can avoid this happening? I've included my `config.xml` below

Folding at Home will use 1 CPU thread (Or an entire core) per GPU plugged into the system. This is completely normal, as the CPU needs to crunch data for the GPU to process. You will also notice this when looking at the bandwidth of the PCIe lane.

 

[tarfeef101]

10 hours ago, LazyDev said:

Folding at Home will use 1 CPU thread (Or an entire core) per GPU plugged into the system. This is completely normal, as the CPU needs to crunch data for the GPU to process. You will also notice this when looking at the bandwidth of the PCIe lane.

 

I find the 100% all the time nature of it suspicious though. That should be throttleable. Especially since my GPUs in every system I've ever used have been pegged at 100% usage regardless of the cpu being used, even if they have wildly varying single threaded performance. I think the cpu resources are being over consumed. I'd hoped

```

<cpu-usage v='10'/>

```

would have been enough to keep the load more reasonable, but it has done nothing