Recently, Nvidia announced their RTX 4000 series at GTC 2022. At a high level, Nvidia took advantage of a process node shrink to evolve their architecture while also scaling it up and clocking it much higher. The top end AD102 die found in the 4090 packs 76.3 billion transistors, which represents a whopping near 2.7x increase over the 28.3 billion transistors found in the RTX 3090’s GA102 die. SM count jumps from 82 to 128 with a hypothetical fully enabled AD102 die going up to 144 SMs. Even with the massive increase in parallelism, Nvidia has managed to increase boost clocks to 2.52 GHz. The result is a massive 450W card that requires a triple 8-pin power connector setup or one of the new 16-pin PCIe 5 power connectors.
Raytracing is the star of the show yet again. Nvidia has spent the last couple GPU generations betting hard on raytracing and the architecture behind RTX 4000, codenamed Ada Lovelace, continues this trend by evolving both Nvidia’s raytracing acceleration hardware along with the upscaling technologies designed to counter the framerate loss typically associated with enabling raytracing.
Ampere doubled the triangle intersection rate compared to Turing, and Ada Lovelace doubles it again. Box intersection rate was not mentioned. Triangle tests happen at the bottom level of a BVH, so Nvidia may be looking to avoid paying memory latency penalties by generating fatter BVH structures.