NVIDIA updates G-Assist AI assistant: more features, faster responses, and 40% less memory consumption

At Gamescom, NVIDIA has introduced updates for Project G-Assist — an experimental built-in AI assistant that allows you to control a system with an RTX graphics card by voice or text. It works on the basis of a local small language model and now acts as a central command center and provides quick access to functions that are usually hidden in the depths of the menu.

With this update, the user can perform the following actions using G-Assist voice or text commands:

• run diagnostics to optimize game performance,
• display or build graphs with frame rate, latency, and GPU temperature,
• change the settings of the graphics processor or even peripherals, such as the keyboard backlight.

The G-Assist update also brings a new, more efficient AI model that is faster, consumes 40% less memory, and maintains the same accuracy. As a result, G-Assist can now run on more devices equipped with RTX graphics cards with 6 GB or more of memory, including laptops. Previously, NVIDIA RTX 30, 40, or 50 series graphics cards with a minimum of 12 GB of memory were required, and Even the RTX 5080 was put to shame by the AI assistant.

To get started, just install the NVIDIA app and the Game Ready driver from August 19, download the G-Assist update, and activate it with Alt+G.

Another G-Assist update is scheduled for September. It will add support for special commands for laptops — including NVIDIA BatteryBoost and Battery OPS.

NVIDIA also added the G-Assist Plug-In Hub, created in partnership with mod.io. It helps to find and download plug-ins that extend the functionality of the assistant. Thanks to the integration with mod.io, users can simply ask G-Assist about available add-ons and install them immediately.

In addition, the G-Assist Hackathon was reported to have been held, where users submitted their own plugins. To develop such custom plugins, the Plug-In Builder is available, which allows you to write JSON and Python-based code, even using natural language.

The following projects reached the finals of the G-Assist Hackathon:

• Omniplay — search for game lore in the wiki or create notes in real time.
• Launchpad — Launch and switch groups of apps to increase productivity.
• Flux NIM Microservice — AI image generation directly from G-Assist using local NVIDIA NIMs.

The winners of the hackathon will be announced on August 20.

Source: NVIDIA