and this is the same rendering with OpenCL:
The very first result is: 6.60 / 5.06 M samples/sec = ~30% faster using exactly the same scene, with exactly the same software and hardware (aside CUDA Vs. OpenCL API). It is a good speed up and I assume there is a HUGE margin for improvement (RTX aside).
I have still a major problem to solve (and probably thousands of minor one) but this is really a milestone.
P.S. when I have most of the stuff working, I will probably go first for out of core support instead of performance because it should be really easy to add. Than I will consider the v2.4 done and I will add RTX support in v2.5.
