Oct 22, 2023 23:50:00

Stable Diffusion is now 2x faster with NVIDIA driver update

It was announced that the 'Game Ready Driver' update distributed on October 17, 2023 will improve the startup performance of some games, and will also make the image generation AI Stable Diffusion up to 2 times faster.

New Game Ready Driver Released: DLSS 3 For NARAKA: BLADEPOINT and Warhammer: Vermintide 2, Plus RTX Video Super Resolution Enhancements & Stable Diffusion Is Now Up To 2X Faster | GeForce News | NVIDIA

https://www.nvidia.com/en-us/geforce/news/game-ready-driver-dlss-3-naraka-vermintide-rtx-vsr/

Nvidia Boosts AI Performance With TensorRT | Tom's Hardware

https://www.tomshardware.com/news/nvidia-boosts-ai-performance-with-tensorrt

With the February 2023 update , Game Ready drivers can now use NVIDIA's AI-based upscaling technology 'RTX Video Super Resolution (VSR).' This technology upscales video resolution by removing compression artifacts , which are distortions that occur when compression is applied, and is expected to improve clarity and edge sharpness.

This update enhances VSR to further improve quality, increase the level of detail, and preserve image detail during the upscaling process for an overall sharper and clearer image. About. Additionally, with expanded support, owners of GeForce RTX 20 series GPUs can now benefit from the same AI-enhanced video as owners of GeForce RTX 30 and 40 series GPUs.

Up until now, VSR has only been applied to upscaled video, but from now on it is possible to achieve high resolution even at native resolution.For example, even when viewing a 1080p stream on a 1080p display, VSR It is possible to improve the image.

Another feature of this update is that 'Stable Diffusion' has been sped up. NVIDIA announced that it has added

the TensorRT extension for Stable Diffusion Web UI powered by Automatic1111, successfully increasing GeForce RTX performance by up to 2x. Although TensorRT must be incorporated first, AI image generation is faster than ever, and users can easily iterate and save time.

When Tom's Hardware, a news site, conducted a performance test incorporating TensorRT, it was confirmed that the performance of Stable Diffusion was increased by up to 70%. Below are the results of the built-in tests.

Below are the results when using the Xformers library, which is often used to speed up Stable Diffusion.

Tom's Hardware pointed out, ``TensorRT is designed to perform better on NVIDIA GPUs, so this performance improvement is seen.''