search

Syllo / nvtop

GPU & Accelerator process monitoring for AMD, Apple, Huawei, Intel, NVIDIA and Qualcomm
Other
7.79k stars 287 forks source link

fatal bug when install in docker #199

Closed sliverp closed 1 year ago

sliverp commented 1 year ago

docker image is nvidia offical image "nvidia/cuda 11.6.2-base-ubuntu20.04" ,you can repeat this bug easily

use apt-get to install apt-get install nvtop

then successfully installed. but next， nvidia-smi said Failed to initialize NVML: Driver/library version mismatch

next,use apt-get autoremove nvtop to uninstall, nvidia-smi was available again

The bug does not appear every time. sometimes nvtop can be used correctly for a period of time,and suddenly docker cuda was broken. I have to run a new container from image,and dont know why. But this time this bug occur derectly so i find it.

alexander-cheplev commented 1 year ago

Have the same issue. uninstalling as @sliverp mentioned solves the problem.

Syllo commented 1 year ago

Yes, Ubuntu 20.04 packaged nvtop with a fixed nvidia driver for some reason. What I would suggest is build nvtop from source, which never had a dependency on a specific nvidia driver whatsoever.

Syllo commented 1 year ago

You can also download the AppImage, which does not need any compilation.