GPU Starved?

A forum for discussing FAH-related hardware choices and info on actual products (not speculation).

Moderator: Site Moderators

Forum rules
Please read the forum rules before posting.
Post Reply
trleith
Posts: 18
Joined: Tue Mar 24, 2020 7:19 pm

GPU Starved?

Post by trleith »

I'm folding in a Dell T7400 with two Xeon processors (8 cores) and an NVIDIA 1060. I have a 6 core CPU slot, and one GPU slot.

I'm seeing the GPU running at 85%. What would be reasons it isn't closer to 100%?

t

Code: Select all

[trl@svr2 ~]$ nvidia-smi
Fri Mar  5 15:32:03 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.39       Driver Version: 460.39       CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  GeForce GTX 106...  Off  | 00000000:02:00.0 Off |                  N/A |
| 62%   75C    P2   108W / 120W |    105MiB /  3019MiB |     92%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A     50332      C   ...13/Core_22.fah/FahCore_22      103MiB |
+-----------------------------------------------------------------------------+
[trl@svr2 ~]$

[trl@svr2 ~]$ top

top - 15:33:16 up 18:15,  2 users,  load average: 7.29, 7.19, 7.16
Tasks: 232 total,   5 running, 227 sleeping,   0 stopped,   0 zombie
%Cpu(s):  0.3 us,  5.7 sy, 83.2 ni, 10.4 id,  0.0 wa,  0.3 hi,  0.0 si,  0.0 st
MiB Mem :  50142.7 total,  45245.8 free,   1641.9 used,   3255.0 buff/cache
MiB Swap:  25216.0 total,  25216.0 free,      0.0 used.  47923.9 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
  33853 root      39  19  598436 184004  12228 R 598.3   0.4   2788:48 FahCore+
  50332 root      39  19   14.9g 384576 189296 R  99.7   0.7  53:39.15 FahCore+
   4278 root      20   0       0      0      0 R  10.3   0.0 112:22.63 nv_queue
   1867 root      20   0  125724   9932   7456 S   3.0   0.0   3:12.56 pmdaproc
   1259 root      20   0  224996  41148  39404 S   1.0   0.1   1:18.38 sssd_nss

...
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: GPU Starved?

Post by Joe_H »

GPU usage is partly influenced by the size of the WU. WUs with few atoms will usually show a lower load than those with many.

Another influence is the number of PCIe lanes and their speed. Faster, wider cards will see some throttling running on narrow connections such as 4x.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Frontiers
Posts: 50
Joined: Thu Sep 13, 2012 3:23 pm
Hardware configuration: Ryzen 5 5600x
G.Skill 2x16 GB 3200@3333
GTX 1070
Lancool II Mesh Perf.
Linux Mint 21.3
some drives, some cooler, some peripherials

Re: GPU Starved?

Post by Frontiers »

You have 92% utilization with this low atom count 13439, or similar to it, WU.
With Windows F@H Client - for same small WUs you will have ~75-80% (according to GPU-Z) utilization, with 20-30% less PPD.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: GPU Starved?

Post by bruce »

Every calculation involves a portion which is parallelizable and a portion which is necessarily serial (except a factious benchmark). All small WUs fail to keep a large number of shaders busy while they're processing the serial portions of the WU's calculation. A GPU with a large number of shaders will not approach 100% parallel calculation as closely compared to a GPU with a small number of shaders but it still runs faster.
Post Reply