You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Unfortunately, I cannot seem to get docker-cuda build running.
The old 18.04/3.6 image (11.3.1-cudnn8-runtime-ubuntu18.04) fails because opencv-python-headless won't build.
With 11.3.1-cudnn8-runtime-ubuntu20.04:
ldconfig
apt-get -y install --no-install-recommends cuda-runtime-10-0 cuda-runtime-10-1 cuda-runtime-10-2 cuda-runtime-11-0 cuda-runtime-11-1 cuda-runtime-11-3 libcudnn7
Reading package lists...
Building dependency tree...
Reading state information...
E: Unable to locate package cuda-runtime-10-0
E: Unable to locate package cuda-runtime-10-1
E: Unable to locate package cuda-runtime-10-2
E: Unable to locate package libcudnn7
make: *** [Makefile:234: cuda-ubuntu] Error 100
From running nvidia/cuda:11.3.1-cudnn8-runtime-ubuntu20.04 directly, I can see that we need another apt-get update before installing. But:
root@b4480b7d3280:/# apt-cache search cuda-runtime
cuda-runtime-11-0 - CUDA Runtime 11.0 meta-package
cuda-runtime-11-1 - CUDA Runtime 11.1 meta-package
cuda-runtime-11-2 - CUDA Runtime 11.2 meta-package
cuda-runtime-11-3 - CUDA Runtime 11.3 meta-package
cuda-runtime-11-4 - CUDA Runtime 11.4 meta-package
cuda-runtime-11-5 - CUDA Runtime 11.5 meta-package
cuda-runtime-11-6 - CUDA Runtime 11.6 meta-package
cuda-runtime-11-7 - CUDA Runtime 11.7 meta-package
cuda-runtime-11-8 - CUDA Runtime 11.8 meta-package
cuda-runtime-12-0 - CUDA Runtime 12.0 meta-package
cuda-runtime-12-1 - CUDA Runtime 12.1 meta-package
and
root@b4480b7d3280:/# apt-cache search libcudnn
libcudnn8 - cuDNN runtime libraries
libcudnn8-dev - cuDNN development libraries and headers
I am not sure about the ramifications of updating libcudnn and not supporting CUDA 10 runtime here - @bertsky?
I am not sure about the ramifications of updating libcudnn and not supporting CUDA 10 runtime here - @bertsky?
Indeed, this will have grave implications for our TF 1.15 processors, probably also Detectron2. I recommend keeping core-cuda on the old (now defunct) image, tracking this in a new issue and merging here – we need the new release!
The text was updated successfully, but these errors were encountered:
#986 (comment)
The text was updated successfully, but these errors were encountered: