Does directml support multi-GPU training #406
Description
My AMD GPUs have more VRAM than their NVIDIA counterparts, so I've been interested in trying directml lately. I was curious if directml, either tenosflow or pytorch versions, support mulit-GPU training? tensorflow.distribute mostly seems to lean on nccl, Nvidia specific, so I wasn't sure if directml had some special way it dealt with this for AMD cards? I saw someone asked a similar question in 2020 and you guys said it was part of your backlog (#70) , so I'm wondering if in 2023 that backlog has been filled...