NVIDIA has officially launched its DLSS Transformer model as part of the 310.3.0 SDK, significantly advancing upscaling while reducing video memory requirements. This new approach directly tackles a major concern for gamers with 8 GB or lower graphics cards by trimming VRAM usage by 20%.
DLSS Transformer
Unlike previous DLSS versions that used convolutional neural networks, the Transformer model analyses the relationships among all pixels across multiple frames to infer missing data. Despite this increased complexity, NVIDIA’s engineers have refined their memory management to maintain lean resource demands and deliver sharper, more consistent visuals… I don’t really understand how this works, but if it’s using less VRAM, I don’t care.
Resolution Matters
Reports from Videocards show that DLSS at 1080p now uses just 87.8MB of VRAM, a drop from 106.9MB. That’s roughly a 20% reduction! This impressive improvement also applies to 1440p, 4K, and even 8K resolutions.
Saving VRAM
This VRAM saving is a huge win, especially for people using graphics cards with less memory. It means older hardware can now run games with the best graphics without compromise. Game developers can look forward to integrating the DLSS Transformer model into their titles and tools in the coming months.
I expect we’ll see further improvements, although I’m sure many gamers would also simply like more VRAM on their lower cost models too.
