So it's been some time and I haven't touched DXR that much yet due to one main reason - its support.

The official GitHub states: 

• NV 397.31+ drivers do not properly support compute Fallback Layer on Nvidia Volta. Samples have artifacts and/or scenes miss altogether. Use the recommended DXR / driver based raytracing mode of samples on this configuration instead.
• AMD: current/v1.5 revision of the Fallback Layer is not supported on AMD cards and will fail to run. Temporarily, you can use previous v1.2 source code snapshot with v1.1 SDK overlay binary snapshot which work on AMD:

Which is as of now quite disappointing, and I remember this note being present for quite some time.

This literally requires one to use v1.2 with v1.1 overlay binary - otherwise I'm cutting off support from quite large amount of devices (additionally when you're running on AMDs this is a problem). Shouldn't the Fallback layer be actually part of the driver?

7 hours ago, Vilem Otte said:

Shouldn't the Fallback layer be actually part of the driver?

It kind of could be, AFAIK DXR doesn't demand any dedicated hardware acceleration, so if a given vendor chooses, they can implement support for it in driver. It's probably what nVidia did with the 16xx non RT Turing and some older Pascal cards, where they enabled DXR not too long ago. The fallback layer by Microsoft seems only like a temporary solution for developers, so they can start integrating raytracing while the hardware and the driver support catches up.