Radeon RX 6800 Series Has Excellent ROCm-Based OpenCL Performance On Linux
When it came to global memory bandwidth, to little surprise the Radeon VII with its HBM2 memory was the winner... The RX 6800 series global memory bandwidth rated by clpeak put it in line with the RTX 2080 Ti / TITAN RTX.
The single-precision float performance via clpeak put the RX 6800 XT at 86% the speed of the RTX 3080. The RX 6800 meanwhile was 23% higher performance for single-precision floats. Over the Radeon RX 5700 series the gains are extremely compelling.
It's with the double precision OpenCL performance where the Radeon Vega and Navi graphics cards dominate over the consumer NVIDIA GeForce graphics cards.
The cl-mem test profile with stressing the video memory capabilities in the OpenCL context is where the HBM2-backed Radeon VII continued to shine.