DLSS 4.5 Ray Reconstruction update arrives in August with transformer model and 27 game launch lineup

Nvidia used Computex to finally announce a proper update for DLSS Ray Reconstruction. The new DLSS 4.5 Ray Reconstruction arrives in August, bringing the AI denoiser in line with the rest of the DLSS 4.5 suite.

The second-generation RR algorithm swaps out the old convolutional neural network for a transformer model — the same shift Super Resolution made back in January. The AI denoiser also draws on a larger training dataset, processing 20% more parameters, which Nvidia says delivers improved lighting accuracy, better temporal stability, and clearer motion in ray-traced and path-traced content, all at similar performance to the old model. At launch it'll be supported in at least 27 games, including Cyberpunk 2077, Doom: The Dark Ages, Alan Wake 2, Pragmata, Resident Evil Requiem, Indiana Jones and the Great Circle, Hogwarts Legacy, and Crimson Desert.

Until now, players using ray tracing with DLSS had to pick one or the other: the DLSS 4.5 transformer upscaler or Ray Reconstruction — not both. That's because RR ran its own internal upscaling stage using the older convolutional model, meaning anyone enabling RR was locked out of the newer algorithm's visual improvements. Nvidia launched DLSS 4.5 in January, Dynamic Multi Frame Generation also took a while to ship, and RR is arriving later still — a timeline that's drawn some criticism given how central ray tracing is to Nvidia's pitch.

AMD also has an AI denoiser in the works as part of its FSR Redstone project, but it's currently only active in one game — Black Ops 7 — with no word yet on a broader rollout. August should be a meaningful moment for Nvidia's ray tracing stack; for the first time, the full DLSS 4.5 package will actually work together without compromise.