这是很久以前的说法了
In 2018, NVIDIA announced GeForce RTX series (Turing architecture).
What does RTX really do?
- It has particular hardware to trace rays
- 10 Giga rays per second
- → 1 sample per pixel (real-time applications)
There is no primary ray because we can use rasterization. Thus, we only need to send out 3 rays.
- If it can improve to 2 SPPs, it will be a great improvement.
State of the Art* Denoising Solution (2019年)
Relation to DLSS: orthogonal (can be run together or respectively)
Goals (1 SPP):
- Quality (no overblur, no artifacts, keep all details)
- Speed (< 2ms to denoise one frame)
Mission impossible
- Sheared filtering series (SF, AAF, FSF, MAAF, ...)
- Other offline filtering methods (IPP, BM3D, APR, ...)