Hello, I created a program for benchmarking CPU and GPU performance, it uses OpenCL to effectively render a Mandelbrot set and does so for around 30 seconds, afterwards it computes the total number of pixels that have been rendered and computes an average amount of pixels/s as a score.
The issue Im facing is extremely poor performance with the AMD OpenCL driver, resulting in throughput of about 4 Gpixels/s for FP32 on my Ryzen 7 1700 (3.8 GHz OC), I figured this out when a couple of my friends ran the benchmark and got much higher scores with their fairly old i7 CPUs. I then installed the Intel OpenCL Driver, which is obviously not optimal for Ryzen, but even then its averaging more than 3x higher scores, 12.6 Gpixels precisely. my Radeon R7 360 GPU is doing 13.5 Gpixels/s, which taking into account the performance of my CPU on the Intel driver seems rather low, but I dont have an equivalent green GPU to test against.
To me it seems rather odd that the AMD Driver would be that much slower, but same things happened on the other systems that were tested, so it is not only my system. Not only does it basically make my benchmark worthless due to driver differences, but I feel like this could pose serious issues in the OpenCL based compute region.
Screenshot of my results:
Source code/program can be provided if needed, I dont feel like throwing a link for my own site in my first post is a great idea