Comment by bee_rider
16 days ago
I wonder if they really need a CUDA rival.
This AI stuff has progressed a bit. Intel has been working on interesting stuff with OneAPI. It might be the case that things have progressed to the point where the primitives are well enough understood that you need something more like a good library rather than a good compiler.
In the end, more people seem to love BLAS than Fortran, after all.
That library (Triton) sits on top of the compiler and drivers (ROCm). If the driver kernel panics, no high-level library can fix that.
I don’t have direct experience, so I could be wrong. But, I believe a lot of the nice stuff that CUDA brings along is profiling and performance related, that is, most useful if you are writing the code yourself. Plus, if the ecosystem is not quite as stable, but it is mostly AMD’s engineers writing the library that have to deal with it, they have more latitude to just not go down the buggy or bad-performance code-paths.