search

NVIDIA / vid2vid

Pytorch implementation of our method for high-resolution (e.g. 2048x1024) photorealistic video-to-video translation.
Other
8.6k stars 1.2k forks source link

Dockerfile not compatible with Tesla V100 #52

Closed burnzzz closed 6 years ago

burnzzz commented 6 years ago

The Dockerfile you provide is based on cuda 8.0 and cannot be used with current Tesla Cards because they need at least cuda 9.0. If you need the 32GB V100 to run the training effectively at full resolution, then there is a mismatch.

Will the code be updated to pytorch 1.0 in the future? pytorch 0.4.0 has problems with current cuda releases.

dustinfreeman commented 6 years ago

Hey @burnzzz , I constructed the docker file empirically & independently (I don't work at NVIDIA), and was experimenting with P100s on Google Cloud machines. I've been considering experimenting with V100s (and CUDA 9.0), but haven't yet. If you can construct a V100 dockerfile, let me know!

dustinfreeman commented 6 years ago

It "just worked" for me on V100 when I updated the Dockerfile like so: #67

rmitsch commented 5 years ago

Didn't work for me unfortunately, I had to go with P100s instead - see https://github.com/NVIDIA/vid2vid/issues/73.