At the end of October AMD released the long-awaited AMDGPU-PRO 16.40 update. For some birthday benchmarking fun today, I finished up a comparison of the AMDGPU-PRO 16.40 stack with its proprietary OpenGL and Vulkan components on various AMD GPUs compared to NVIDIA results using the 375.10 binary driver.

It's a binary vs. binary AMD and NVIDIA comparison today with both OpenGL and Vulkan via Valve's Dota 2 game. The RADV / RadeonSI Git stack wasn't tested for that article but will have that out in its own results later this week due to the mass amount of data already queued up for today's posting.

Built off the 13-way NVIDIA comparison I did recently, the AMD cards I had available for testing on AMDGPU-PRO were the Radeon R9 285, RX 460, RX 480, and R9 Fury.

While at 1080p the OpenGL results appear CPU bound on NVIDIA and the Vulkan results didn't fluctuate too much, unfortunately, the AMD results remained much lower. With this latest 16.40 public driver, the OpenGL driver was faster than the official AMD Vulkan driver (not RADV) and for both the Vulkan and OpenGL usage the performance was significantly lower than NVIDIA with their proprietary Linux driver.

Pushing the GPUs much harder by running at 3840 x 2160, the AMD results continued to remain much lower and the AMDGPU-PRO Vulkan results were trailing the OpenGL numbers. It looks like the 16.40 driver might be in a bit of a regressed state for Vulkan given the minimal difference between the RX 480 and R9 Fury.

Those are the latest numbers to share today. RADV and RadeonSI Git metrics coming up shortly.

Update (10 November 2016): It appears the slowdowns with 16.40 may be due to running it on an unsupported (4.8) kernel. Additional tests happening now.