NVIDIA is gearing up for a major update to its DLSS technology, aiming to switch to a “Transformer Model” that promises several exciting improvements, including reduced VRAM usage. This upgrade is particularly appealing for gamers with GPUs offering 8GB or less VRAM, addressing a common concern in the gaming community.
Currently, DLSS relies on Convolutional Neural Networks (CNN), but NVIDIA is shifting to a more advanced transformer model. This new approach is expected to enhance image quality and improve ray reconstruction, addressing a broader range of gamer needs. Just a few days ago, the DLSS Transformer model emerged from its beta phase, and it’s reported that NVIDIA has successfully optimized VRAM requirements, as detailed in the DLSS Programming Guide.
On average, this model reduces VRAM usage by about 20%, a significant improvement that remains consistent across all resolutions. This change is a vital benefit for gamers who want to leverage DLSS’s capabilities but are limited by their current hardware.
The transformer model uses a vision transformer to analyze every pixel within a frame, determining their importance at an individual level. By doing this across multiple frames, it generates highly detailed pixels, dramatically enhancing visuals. The model boosts parameters by two times and rendering compute by four times, indicating substantial enhancements to the upscaling process.
With the beta phase complete, we can expect NVIDIA to officially roll out this transformative technology in the coming months, bringing these benefits to gamers everywhere.
