Ok, to be honest, I haven’t gotten to this part of the course yet but I heard that Swift for Tensorflow may become a huge part of fastai in the future.
And Swift is huge with Mac programming.
According to this, Swift builds on top of LLVM which supports AMD Cards: https://github.com/tensorflow/swift/blob/master/docs/WhySwiftForTensorFlow.md
Putting 2 and 2 together, does this mean that GPU support for Fastai may finally be coming to the Mac (which typically uses AMD Gpus) via S4TF?
S4TF uses Tensorflow, which doesn’t have great support for AMD yet. So it’s dependent on that.
AMD has it’s ROCm TensorFlow port (up to date with latest TF 1.14), but MacOS GPU is not listed there.
Linux GPU should work and i know companies that use AMD Radeon VII (16GiB memory) in their machines and it works with TensorFlow! (no licence restriction from AMD to use only expensive GPU in servers)
I would definitely support more competition in the deep learning GPU space. If the goal is to make deep learning more available to a wider audience, cost is still an important factor and if we can bring that down by getting AMD into the space, I’m all for it.