Radeon ROCm 1.9.1 vs. NVIDIA OpenCL Linux Plus RTX 2080 TensorFlow Benchmarks
The OpenCL kernel latency with Radeon ROCm remains much higher than when using the NVIDIA driver stack.
The NVIDIA Turing graphics cards come out multiple times faster than Pascal (and the Radeon Fury/Vega cards) for the clpeak integer compute test.
For those curious about the performance-per-Watt for those Radeon vs. NVIDIA OpenCL benchmarks:
The Vega performance-per-Watt remains well behind Pascal, like with the Vulkan/OpenGL Linux gaming tests the power efficiency of Vega is usually aligned closer to the older Maxwell GTX 900 series.
Here's a look at the overall AC system power consumption during the duration of the compute benchmarks carried out.