After an initial preview at Computex 2024, the tool is now available for desktop RTX GPU users through the Nvidia app, with laptop support expected to follow later.
Nvidia has released Project G-Assist, an experimental AI assistant that runs locally on GeForce RTX graphics cards. The release is part of a broader trend where major tech companies are developing AI assistants specifically for gamers. Microsoft, for example, is working on a "Copilot for Gaming" that will initially serve as a chat assistant for game tips and later analyze gameplay scenes in real-time. At Computex in Taiwan, Nvidia had already presented various AI products, including a version of Project G-Assist for contextual help in games, Nvidia ACE for digital humans, and the RTX AI Toolkit for developers.
Unlike cloud-based AI assistants that require internet connections and subscriptions, G-Assist runs entirely on the user's GeForce RTX GPU and uses a Llama-based model with 8 billion parameters. When activated via Alt+G, the GPU temporarily allocates resources for AI processing, which may briefly affect the performance of other applications. The system requires RTX graphics cards from the 30, 40, or 50 series with at least 12 GB of VRAM.
Community improvements expected for G-Assist
G-Assist performs functions such as system diagnostics, game optimization, GPU overclocking, and performance monitoring. The assistant can also control compatible peripherals from Logitech, Corsair, MSI, and Nanoleaf. The tool specifically focuses on PC-related functions rather than general conversation.
Nvidia has designed G-Assist for community extensions and provides a GitHub repository where developers can create plugins using JSON formats. Example plugins include Spotify integration and Google Gemini connectivity. The company is actively requesting user feedback to guide the future development of this experimental feature.
Nvidia develops its own language model for NPCs
G-Assist isn't Nvidia's first venture into local AI assistants. In May 2024, the company updated its experimental ChatRTX chatbot for RTX GPU owners with additional AI models, including Google's Gemma, ChatGLM3, and OpenAI's CLIP model for photo searches. Additionally, Nvidia recently introduced Nemotron-4 4B Instruct, a compact language model designed to improve the conversational capabilities of game characters.
This development aligns with Nvidia's long-term vision of increasingly integrating AI systems into games and other applications. Back in late 2018, Nvidia first demonstrated a system that could generate an interactive 3D city in real-time based on training videos.
