I wish them all the best, but I haven't seen much adoption yet. It's not particularly clear that this is the right path yet because folks are building special-purpose neural net accelerators, and those may not have the same programming model and may make this whole HIP thing irrelevant for ML.
I'm also not totally convinced software developers are ready to take advantage of something like this; developers are barely taking advantage of multiple CPU cores, let alone the more limited GPGPU environment.
I wish them all the best, but I haven't seen much adoption yet. It's not particularly clear that this is the right path yet because folks are building special-purpose neural net accelerators, and those may not have the same programming model and may make this whole HIP thing irrelevant for ML.
I'm also not totally convinced software developers are ready to take advantage of something like this; developers are barely taking advantage of multiple CPU cores, let alone the more limited GPGPU environment.