Tensorflow/Pytorch with GPU on ARM64


I wish to launch some ML python scripts on arm64 and I have a question how one could use tensorflow or pytorch with gpu acceleration using nvidia drivers. As I can see, for ARM64 NVIDIA supports only CUDA 11. However tensorflow is only compatible with version 10  (unless you use tf-nightly-gpu 2.5.0.dev20201029 version of tensorflow which is not yet available on arm64). The same goes for pytorch: I cannot find a way to build version 1.7 on arm64 (which is the only one that supports CUDA 11).

More questions in this forum