Is there some sort of compatibility chart for tensorflow versions relating to CUDA/cuDNN? "Use the latest install guide" doth butter no parsnips if you're working with an older version of tensorflow.
Does someone know what's the current state of TF on Windows?
I've used it for some image classification homework 1.5y ago and back then I had some issues but it kind of worked.
Does someone have more recent experiences in some hobby project or even in a production env?
Or are there other NN frameworks when using Windows?
It's pretty easy to get running on Windows since 1.0 or whatnot, in my experience. I just ran `pip3 install --upgrade tensorflow-gpu` and do that on every release and it works just dandy. You get a binary wheel of course, so the install is quick and painless.
I did have to install a new cudnn 7 DLL recently after moving to 1.5 (or 1.6) with CUDA 9. That's about it.
I don't use all the fancier tools though (e.g. loading from GCS/S3 buckets etc, just local data for me). But everything seems to works just fine these days, it seems.
15 comments
[ 2.6 ms ] story [ 43.1 ms ] threadGlad to see it since I think that means I can use CUDA 9.1.
Does someone have more recent experiences in some hobby project or even in a production env? Or are there other NN frameworks when using Windows?
I did have to install a new cudnn 7 DLL recently after moving to 1.5 (or 1.6) with CUDA 9. That's about it.
I don't use all the fancier tools though (e.g. loading from GCS/S3 buckets etc, just local data for me). But everything seems to works just fine these days, it seems.