Hi, there is no short answer to that question. I found that only and the best implementation of OpenCL is on macOS and I compared Nvidia Titan RTX 24GB GDDRX VRAM ($3000) on GNU/Linux, on CUDA, on OpenCL… and to beat it I chose AMD Radeon VII 16GB HBM2 VRAM ($700), but on GNU/Linux is slow, very slow. However, on the installed by me macOS on Hania PC, this AMD GPU computation time beat easily the performance of Titan RTX… here it is the comparison results below.
In conclusion, I can only say that properly implemented GPU driver and OpenCL, even if this technology has 20+ years is amazing good on Apple macOS. And even if deprecated I believe it is a future of AI/ML and C/C++ software industry.