I hear you, but like I've stated a nr of times now, without a proper static analysis resulting in a theoretical design that we know will run profitably on AMD GPUs, there is nothing that can justify spending the enormous amount of implementation time necessary here. Spending 2+ months on an implementation of something that earns 25% of a decent Ryzen CPU is clearly suboptimal use of our time. And, note that the available CPU benchmarks for RandomX are based on the available proof of concept code. It's not bad, but if anything, CPU performance will increase over time. As an indicative comparison, SCernykh threw what he could at the problem with AMD GCN asm and on-gpu program compilation and ended up with 1100 h/s on his Vega 64, this compared to the listed benchmark for a Ryzen 7 1700 at 4100 h/s.

So if this hasn't been clear already from the specs for RandomX, ongoing Reddit discussions etc: for all gpu miners who still believe you'll be mining XMR after the October fork, the odds are stacked heavily against you. The nrs and comments the RandomX people for some reason are putting out for GPUs are completely irrelevant, measuring their performance as a pct of their CN/r hashrate. Who cares if a Nvidia gpu (which is useless for CN/r to begin with) can produce roughly the same hashrate after the fork when a CPU increases its hashrate 7-10x?