search

dattalab / keypoint-moseq

https://keypoint-moseq.readthedocs.io
Other
70 stars 28 forks source link

NVCC version issue #161

Open simonlsp opened 2 weeks ago

simonlsp commented 2 weeks ago

In "conda_envs\environment.win64_gpu.yml", the version of cudatoolkit is 11.8.0, but the version of cuda-nvcc is not defined.

  - conda-forge::cudatoolkit=11.8.0
  - conda-forge::cudnn=8.8.0.121
  - nvidia::cuda-nvcc

Unlike "conda_envs\environment.linux_gpu.yml"

  - cuda-nvcc=11.8.89

This leads to different cuda version after windows install.

cuda-nvcc                 12.4.131                      0    nvidia
cuda-version              11.8                 hcce14f8_3    defaults
cudatoolkit               11.8.0              h09e9e62_13    conda-forge
cudnn                     8.8.0.121            h84bb9a4_4    conda-forge

Which further leads to such error

jaxlib.xla_extension.XlaRuntimeError: UNKNOWN: no kernel image is available for execution on the device

Soution: define nvcc version in environment.win64_gpu.yml

calebweinreb commented 2 weeks ago

Thanks for the heads up! So were you able to fix this issue by modifying conda_envs\environment.win64_gpu.yml? If so, could you just clarify exactly what the change was or create a PR with your new version?