Associate
- Joined
- 29 Jun 2016
- Posts
- 541
I welcome the DLSS/FSR upscaling revolution. It is analogous to the magic that has been worked to make games look as good as they do without supercomputers. Like how things are rendered is mathematical magic, so why not upscale. It makes perfect sense.
Where I'm not super keen is frame interpolation. There were some fantastic strides at frame generation in the VR space (ASW, etc), i.e. limited frame prediction based on motion vectors and head pose changes, however DLSS3.0 is interpolation. It inserts a frame between two known frames, which in a real-time context is just meh. Maybe DLSS3.0 will improve dramatically and I'll be an adopter, but I kinda doubt it...
I'll be very much onboard when DLSS 4.0 attempts frame generation, similar to ASW, but using ML techniques. Have a stall in your pipeline? No worries, just generate a frame and away you go.
Where I'm not super keen is frame interpolation. There were some fantastic strides at frame generation in the VR space (ASW, etc), i.e. limited frame prediction based on motion vectors and head pose changes, however DLSS3.0 is interpolation. It inserts a frame between two known frames, which in a real-time context is just meh. Maybe DLSS3.0 will improve dramatically and I'll be an adopter, but I kinda doubt it...
I'll be very much onboard when DLSS 4.0 attempts frame generation, similar to ASW, but using ML techniques. Have a stall in your pipeline? No worries, just generate a frame and away you go.