extremetech.com

Nvidia’s Project G-Assist Gives You an SLM Tool for Your GPU

Nvidia Project G-Assist app

Nvidia Project G-Assist AI tool. Credit: Nvidia

You can tell your graphics card to overclock itself. Well, you could always tell it to, but now, it will actually comply. That is, if you own an Nvidia RTX 30, 40, or 50 series GPU (with 12GB or more VRAM) and your PC meets the minimum requirements for the company’s new Project G-Assist feature. Nvidia tucked an experimental version of the AI system assistant into its latest update of the Nvidia App. Just press ALT+G and you can take the small language model (SLM) AI assistant for a test run.

The G-Assist feature lets you activate certain hardware and software settings by voice command. In most cases, you should be able to speak to it the way you’d talk to a human. You can ask your PC to run diagnostics, make adjustments (such as overclocking), or customize peripherals and software, so long as they support G-Assist. Corsair, Google Gemini, Logitech, MSI, Nanoleaf, and Spotify already do, according to Nvidia.

Nvidia App update screen

G-Assist comes in the latest Nvidia App update. Credit: Nvidia

G-Assist currently uses a Llama-based model with 8 billion parameters and was built with Nvidia ACE. Being an SLM AI tool means that you don’t need an internet connection to use it, but it also means that the tool is limited compared with massive, online AI language models. G-Assist is designed to be a very capable software assistant. And its capabilities will grow, Nvidia says. If you want to build a plug-in that uses G-Assist (or want to see what others have made), you can check out Nvidia’s GitHub repository.

When you use G-Assist, remember that the feature is a work in progress. It wasn’t all that long ago that another Nvidia App feature (the overlay) was involved in unexpected performance issues. In this case, a brief performance drag is to be expected if you’re pushing the GPU. “If you’re simultaneously gaming or running another GPU-heavy application, a short dip in render rate or inference completion speed may occur during those few seconds,” Nvidia stated in its announcement of G-Assist. “Once G-Assist finishes its task, the GPU returns to delivering full performance to the game or app.”

The minimum specs for G-Assist are forgiving. Beyond the RTX 30, 40, or 50-series GPU with 12GB VRAM, all your PC needs is Windows 10 or 11, a reasonably modern AMD or Intel processor, a modest amount of storage space, and any GeForce driver from 572.83 or later. (Get the detailed minimum requirements here.)

Read full news in source page