I just finished up some testing of the following:
- VR in forward with DLSS (quality mode)
- VR in forward with MSAA
- VR in deferred with DLSS (quality mode)
- VR in deferred with TAA
Unfortunately I was a little disappointed but I need to do far more testing before I draw broad conclusions. In the one scene I tested across all three scenarios, I found:
- DLSS vs MSAA: DLSS was still pretty blurry, even with the forward renderer. This seems to be because DLSS is still using temporal data so you get a lot of the same downsides of TAA. If TAA in VR was too blurry, DLSS in VR will be too blurry for you, too.
- DLSS vs TAA: This one surprised me. Overall blurriness was about the same, except for particle motion. Particles seemed *much *blurrier with DLSS on.
- DLSS didn’t save much, if anything, on performance. This one definitely needs more testing to verify, but I was surprised to find at times DLSS was even more expensive. It added about 2ms to the frame time, which was about as much as it reduced.