NVIDIA's DLSS 4.5 unifies denoising and Super Resolution this August

NVIDIA's DLSS 4.5 Ray Reconstruction, slated for an August release, marks a major leap in real-time ray-traced gaming. The update introduces a 2nd-generation transformer model that consolidates denoising and Super Resolution into a single AI network, replacing traditional hand-tuned denoisers. With 35% more compute capability and 20% more parameters, the supercomputer-trained model delivers sharper, clearer motion and better temporal stability across all GeForce RTX GPUs. NVIDIA says it's trained on an expanded dataset for superior spatial awareness, getting closer to ground-truth visuals while offering developers finer temporal accumulation controls. This builds directly on DLSS 4.5's core innovations like 6X Multi-Frame Generation, improving lighting accuracy without a performance cost.

Other announcements from the same Computex 2026 keynote included the RTX Spark superchip for thin laptops and compact desktops, which pairs a 20-core Grace CPU with a Blackwell RTX GPU via NVLink-C2C for up to 1 Petaflop of AI compute and 128GB of unified memory. NVIDIA also unveiled the Vera CPU, a data center processor with 88 custom Olympus cores, 1.2 TB/s bandwidth, and 1.5TB SOCAMM LPDDR5X memory, cutting token-processing costs tenfold in the Vera Rubin NVL72 platform. But for gamers, DLSS 4.5 is the headline: a unified AI network that pushes visual fidelity and performance to new heights.