On 2018-11-16 06:54, Mo Zhou wrote: > Please hold on the CUDA version for a while. Some important applications > are still not fully ready for CUDA 10. Since recently Nvidia started to > produce buggy software and buggy hardware, I think we should slow down > our pace and only update CUDA when necessary, or when the target version > has been extensively tested by it's user communities such as TensorFlow, > PyTorch, etc.
I have pushed branches 10.0 and 10.0+ to the git repository. It builds, but I have not tested it at all. Especially the nsight-compute package needs more work. Andreas
