TECH NEWS – Nvidia’s new transformer-based upscaling model could be a game-changer for mid-range graphics cards.

The next big DLSS update from Nvidia appears to be shifting toward a transformer model that is expected to bring several benefits, including reduced VRAM consumption. Currently, Nvidia uses CNNs (convolutional neural networks) for DLSS, but the green team is now transitioning to a more advanced approach that could significantly enhance image quality and ray reconstruction.

The DLSS Transformer recently exited beta testing, and according to reports, Nvidia has implemented VRAM optimizations. These improvements are outlined in detail in the updated DLSS programming guide.

On average, the DLSS Transformer model within the 310.3.0 SDK requires about 20% less VRAM, and this reduction holds consistently across all resolutions. This suggests the update could have a broad impact, especially for gamers using cards with 8 GB or less of VRAM.

Given that Nvidia’s recent strategies haven’t done much to benefit budget-conscious gamers, this progress in DLSS technology is a welcome development. It might finally give some relevance to the GeForce RTX 5050 (which uses GDDR6 in its desktop version) and the RTX 5060—both of which feature 8 GB of VRAM.

The Transformer model leverages a vision transformer that evaluates every pixel in a given frame and determines their individual significance. This is repeated across multiple frames to reconstruct detailed pixels and enhance overall visual output. It’s reported to double the number of parameters and quadruple rendering calculations, making the upscaling experience significantly better.

Now that the DLSS Transformer has exited beta, an official announcement could be just around the corner. Could Nvidia launch it alongside the upcoming GeForce RTX 5000 Super cards? It wouldn’t be surprising if that’s the path the Jensen Huang-led company takes.

Source: WCCFTech, Videocardz