GPU being utilized properly? Nvidia drivers give no info

[mrmatthew2k]

Hello.

I have a home server on Ubuntu 19.10. I want to use only GPU folding and I set up my fahclient through the fahcontrol client (the dependencies the package are messed up but whatever).
Anyways, it appears to be working. Except, when I want to verify everything, nvtop and nvidia-smi have nothing to report. Also, the fahclient is using one of my cpu cores at around 90% utilization. So my questions are:

1. Is the high cpu utilization normal behavior even when only GPU folding?
2. Am I correct in saying my GPU is being properly utilized but NVIDIA drivers don't want to tell me anything?

Code: Select all

Environment of process 7057 - /var/lib/fahclient/cores/cores.foldingathome.org/v7/lin/64bit/Core_22.fah/FahCore_22 -dir 00 -suffix 01 -version 704 -lifeline 7053 -checkpoint 15 -gpu 0 -gpu-vendor nvidia

uses 90-100% of one of my cpu core.

Code: Select all

Mon Apr  6 06:51:28 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.50       Driver Version: 430.50       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 670     Off  | 00000000:01:00.0 N/A |                  N/A |
| 40%   59C    P0    N/A /  N/A |    113MiB /  1997MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0                    Not Supported                                       |
+-----------------------------------------------------------------------------+

[HaloJones]

The section under "Volatile GPU-Util" should be showing a percentage utilisation - it does on mine (Mint).

This suggests it's not running.

Code: Select all

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 415.25       Driver Version: 415.25       CUDA Version: 10.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1070    Off  | 00000000:03:00.0 Off |                  N/A |
|  0%   54C    P2   143W / 230W |    171MiB /  8119MiB |     97%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 1070    Off  | 00000000:04:00.0  On |                  N/A |
|  0%   50C    P2   140W / 230W |    445MiB /  8118MiB |     99%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     16118      C   ...7/lin/64bit/beta/Core_22.fah/FahCore_22   159MiB |
|    1      1369      G   /usr/lib/xorg/Xorg                            98MiB |
|    1      1792      G   cinnamon                                      20MiB |
|    1      4797      C   ...7/lin/64bit/beta/Core_22.fah/FahCore_22   321MiB |
|    1      5459      G   /usr/bin/nvidia-settings                       0MiB |
+-----------------------------------------------------------------------------+

[PantherX]

Welcome to the F@H Forum mrmatthew2k,

Regarding the CPU utilization, currently, Nvidia GPUs require a CPU thread per GPU as it uses a spin-wait cycle (polling) which uses 1 CPU. This happens on Windows and Linux since that's the method Nvidia choose to implement OpenCL support.

[scm2000]

Welcome to the F@H Forum mrmatthew2k,

Regarding the CPU utilization, currently, Nvidia GPUs require a CPU thread per GPU as it uses a spin-wait cycle (polling) which uses 1 CPU. This happens on Windows and Linux since that's the method Nvidia choose to implement OpenCL support.

Does it at least call sleep while polling? Or is it just burning cpu cycles for no reason?

[PantherX]

...Does it at least call sleep while polling? Or is it just burning cpu cycles for no reason?

AFAIK, the implementation by the Nvidia Team has it so that it uses 1 CPU worth of cycles for no apparent reason other than it can use polling.

[bruce]

It's not clear what version of OpenCL is installed. FAHCore_22 requires both the NVidia proprietary drivers and the OpneCL drivers.

Please post segments of FAH's log. We need the first couple pages of the log where FAH detects the hardware and sets it up plus segments showing any errors. What messages are you getting and what WUs were active, if any?

See instructions below.