I feel like I'm going to stir up the hive with this one. But oh well.

So on some adventures of looking up to see if AMD cards ever ran a DXR fallback layer compatible app, I came across this article (https://www.pc-better.com/dxr-on-radeon-vii/) of someone who ran a Radeon VII on the DXR samples that Microsoft has on GitHub. Then something interesting caught my eye

The two statistics I'm looking at are the FPS (at 10) and the Million Primary Rays per second, at 7.69.

So why do I bring this up? Well there was a thread on Reddit of people who ran the same demo on non-RTX NVIDIA hardware back in October 2018. The particular demo in the screen cap is the "reflection test". I don't think there's a GPU that was tested that was below 10 million primary rays per second.

It makes me wonder what's going on here. My best guess at the moment is DXR is optimized to run on the graphics pipeline, which is a weaker point in GCN's architecture. I'm sure if you managed to convert this to compute functionality (which AMD's Radeon Rays ray tracing API leverages using OpenCL), the results would be much better.