5
Deceptive nvidia-smi output
source link: https://scaleoutsean.github.io/2023/12/06/deceptive-nvidia-smi.html
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
Deceptive nvidia-smi output
06 Dec 2023 -
1 minute read
I’m curious if I should buy a more expensive GPU so I use nvidia-smi to see how busy the current one is.
$ sudo nvidia-smi
Wed Dec 6 13:09:01 2023
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 545.23.08 Driver Version: 545.23.08 CUDA Version: 12.3 |
|-----------------------------------------+----------------------+----------------------+
| 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 NVIDIA GeForce RTX 2080 ... On | 00000000:2B:00.0 On | N/A |
| 0% 46C P8 16W / 250W | 1136MiB / 8192MiB | 39% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
Whoa, 39%!
BUY, BUY, BUY!
What’s in fact happening the GPU is so damn idle that GPU clocked down to a fraction of the maximum clock speeds.
GPU memory clock and processor clocks are each around 400 MHz, compared to the maximum values of 7750 and 1650 MHz, respectively.
$ sudo nvidia-smi dmon
# gpu pwr gtemp mtemp sm mem enc dec jpg ofa mclk pclk
# Idx W C C % % % % % % MHz MHz
0 15 44 - 12 9 0 0 0 0 405 375
0 15 44 - 25 17 0 0 0 0 405 375
So my 39% is really just a small fraction of maximum GPU performance I can get from this hardware.
Recommend
About Joyk
Aggregate valuable and interesting links.
Joyk means Joy of geeK