How to set GPU high capacity

[enlaihe]

Hi nvidia team:

I want to test the power dissipation of GPU on high capacity .
I kown a open source tools gpu-burn:GitHub - enlaihe/gpu-burn: Multi-GPU CUDA stress test
so download and make
exec ./gpu_burn -d 3600

but there is an error: sh：1： nvidia-smi:not found.
and when set time 3600s ,the test over soon.

[kayccc]

I don’t know if any suitable tool for GPU stress test, may other developers help to share experiences.

[linuxdev]

Added note: nvidia-smi is a PCI tool (a “discrete” dGPU is needed). The GPU in the Jetson is integrated (an iGPU), and so nvidia-smi based applications will not work on a Jetson.

[Honey_Patouceul]

Seems gpu-burn only uses nvidia-smi for monitoring GPU temperature. You may find similar temperature information from tegrastats or from sysfs (PMIC temperature in tegrastats output - #4 by Honey_Patouceul), so you may adapt gpu-burn code and rebuild for Jetson.
Note that there may be other adaptations, for example memory management which is different because iGPU on Jetson shares the same physical memory than CPUs.

[enlaihe]

Hi linuxdev:

do you know any other tools can set GPU High capcity and take a test?

thanks.

[linuxdev]

Sorry, I do not know of such a case. You would probably have to write a program to do this, or at least port the other nvidia-smi-based tool (which would not be easy).

[linuxdev]

Followup: As mentioned by @Honey_Patouceul, you can get the temperature with sysfs. If you can find a way to load the GPU, then you probably don’t need the nvidia-smi tools.

[enlaihe]

hi linuxdev:

I had follow @Honey_Patouceul mentioned.
In fact ,I can get the temperature and other info from jtop,now i remove the source code about nvidia-smi,it’ ok now.
tks.