Nvidia DLSS Transformer model update lowers VRAM use by 20%

DLSS has greatly increased the memory efficiency of its DLSS Transformer model, reducing VRAM usage

Nvidia’s DLSS technology continues to evolve, and with version 310.3.0, Nvidia has dramatically lowered the VRAM usage of its DLSS Transformer model. This change was first spotted by Videocardz.

Nvidia’s DLSS Transformer model is a critical part of Nvidia’s DLSS 4 ecosystem. Compared to Nvidia’s older CNN model, Nvidia uses more parameters and higher levels of AI compute to deliver higher-quality upscaling results. This change has enabled higher levels of image quality in games using DLSS upscaling.

With its DLSS 310.3.0 update, Nvidia has lowered the memory usage of its newest DLSS Transformer model by approximately 20%. While memory usage remains higher than with Nvidia’s older CNN DLSS model, this update greatly increases the memory efficiency of Nvidia’s DLSS Transformer model upscaler.

(Nvidia DLSS Super Resolution VRAM use comparison – Image from Videocardz)

While a 20% reduction in VRAM usage is significant, it is worth noting that DLSS has a relatively small memory footprint. Regardless, this change should come as good news for gamers with older graphics cards. New PC games are demanding more VRAM, and lowering DLSS’ memory footprint will help make 8GB GPUs a lot more viable.

You can join the discussion on Nvidia lowering the VRAM usage of its DLSS Transformer model on the OC3D Forums.