Clever Geek Handbook

How to explain this figure about nvidia-smi with NVIDIA GPUs?

I just started simpleMultiGPU

on a server with two GPUs installed K20m

. Then run the command nvidia-smi

to display the state of the GPUs. The result is as follows:

enter image description here

The problems are as follows:

  • GPU memory usage doesn't seem to be the same.
  • Unstable GPU-Util device 1

    is 99%.

Any advice?

+3


source to share


1 answer


  • There is no concept that memory usage should be the same when running applications on multiple GPUs. In addition, memory usage can vary at different points in the application, depending on the main activity (for example, the point at which the allocations are made) when the memory usage is sampled.

  • The use of the Voltage GPU of device 1 is likely an indication that the application and / or is nvidia-smi

    causing activity on device 1.



It's not clear why you call these problems. You control the activity on the device.

+4


source






More articles:


All Articles