Yeah, at this point we're not planning on investing time and effort into RandomX. The current GPU implementation for AMD GCN can be improved, but it's too far from being profitable vs CPUs anyway.
@kerney666 I think what the people are asking for is that whether you can do RandomX optimisations for Ryzen as you did for the GPU algos on CGN. They are not necessarily asking for RandomX optimisations on CGN. Is what I’m saying making any sense?
Well, we do certainly have the chops to pull off some serious x86-64 low level work, but the guys behind RandomX are skilled in that area as well. Moreover, the algorithm itself is (on purpose) an exercise in creating a free-from-patents instruction set that by "random" chance maps very very well to x86-64+SSE2. So, the random code is translated into x86-64+SSE2 ops, and that's about it. I don't think there's any low hanging fruit there really, but I haven't spent more than 15 mins looking at a few RandomX programs so can't say for sure. I'm sure there are a number of optimizations left to be uncovered that applies to special case situations in the RandomX programs, but it's not going to happen right now, we have some other ongoing work we need to finish firs before we can even consider it.
Gotta love it though!
