Not all AIs work in the cloud, which is why with some, GPU performance is important.
NVIDIAwith the support of Microsoft, has announced that the latest version of its drivers for your GeForce graphics cards, you will double the performance of Stable Diffusion generative artificial intelligence.
Stable Diffusion was one of the first generative AIs which became famous even before ChatGPT. It allows create a realistic drawing or photo, from a text.
Most generative artificial intelligence models, such as DALL-E and Midjourney, work in the cloud and they do all the calculations on remote servers. Stable Diffusion, on the other hand, can work in local modewith the sole condition of using a graphics card with at least 8 GB of RAM.
That is the reason that its performance, that is, the time it takes to create an image from text, depends on the power of the graphics card, if you use stable diffusion in local mode.
New NVIDIA drivers and Stable Diffusion
NVIDIA today releases version 532.03 of its drivers for GeForce cards. It includes the usual bug fixes and support for the new RTX 4060 cards, which are going on sale these days.
But the big news is a new technology that allows double the performance of Stable Diffusion.
As Tom’s Hardware explains, NVIDIA has integrated a Microsoft tool called Olive. This tool allows you to adjust the base model of artificial intelligence, depending on the power of the hardware.
NVDIA claims that Olive is capable of improving the performance of AI operations that take advantage of the RTX Tensor cores present in Nvidia GPUs. Using an Olive-optimized version of the Stable Diffusion text-to-image generator, the yield is multiplied by more than 2.
It is good news for those who use this generative AI. and an announcement of what is to come. The new NVIDIA GeForce drivers are usually dedicated to games, but from now on they are also going to have to make room for artificial intelligence. You can download this latest version optimized for Stable Diffusion, 523.23, on the official NVIDIA website.