CUDA `11.x` images cannot work with non-Tesla cards due to the `cuda-compat` package
cuda-compat
Context on See the first paragraph of #182.
11.x
upstream image
Non-Tesla GPU cards (except for some RTX cards) can't run any Forward compatibility is only supported for Tesla (data center) cards and some RTX cards: https://docs.nvidia.com/deploy/cuda-compatibility/index.html#forward-compatible-upgrade
Forward Compatibility is applicable only for systems with NVIDIA Data Center GPUs or select NGC Server Ready SKUs of RTX cards. It’s mainly intended to support applications built on newer CUDA Toolkits to run on systems installed with an older NVIDIA Linux GPU driver from different major release families. This new forward-compatible upgrade path requires the use of a special package called “CUDA compat package”.
See also:
- https://github.com/NVIDIA/nvidia-docker/issues/1515#issuecomment-872962686
- https://github.com/NVIDIA/nvidia-docker/issues/1515#issuecomment-872974709
Therefore, CUDA 11.x
images can currently only work with:
- all Tesla cards (only with some specific driver versions, see #181 (closed))
- Kepler cards (only tags < 11.6)
I have not tested this, but it should hold according to the NVIDIA docs, and I understand your team wants to be pedantic w.r.t. accurately following what the docs support.