I'm not sure I agree with this. In my view the good alternative would have been OpenCL based machine learning libraries, no?
Nvidia deliberately undermined that at every turn, and pushed a proprietary solution, hard. Almost all gpgpu based machine learning used it, and, what is AMD left with? HIP, the "CUDA wrapper to make it work on ROCm".
What's the alternative here? Say that Nvidia shouldn't do anticompetitive things? That's been an exceedingly successful market strategy for them for quite some time.
Nvidia deliberately undermined that at every turn, and pushed a proprietary solution, hard. Almost all gpgpu based machine learning used it, and, what is AMD left with? HIP, the "CUDA wrapper to make it work on ROCm".
What's the alternative here? Say that Nvidia shouldn't do anticompetitive things? That's been an exceedingly successful market strategy for them for quite some time.