Nvidia's new device, Project Digits, is based on the GB10 Grace Blackwell Superchip and is expected to be priced starting from $US 3,000.

“AI will be mainstream in every application for every industry. With Project Digits, the Grace Blackwell Superchip comes to millions of developers,” Huang said. “Placing an AI supercomputer on the desks of every data scientist, AI researcher and student empowers them to engage and shape the age of AI.”

Each Project Digits model will include a whopping 128GB of RAM while disk space appears to be a configurable option, with capacities up to 4TB. However, two Project Digits systems can be linked together to combine their computing power for even more demanding purposes.

The GB10 Grace Blackwell Superchip can handle AI models with up to 200 billion parameters, and delivers up to one petaflop of AI performance, a 'flop' being floating-point operations per second. This combines with Nvidia's CUDA cores and Tensor Cores.

The system will run a Linux-based operating system, the Nvidia DGX OS. Users will get access to Nvidia's AI software library including development kits, pre-trained models, and orchestration tools. Popular frameworks like Python, Jupyter, PyTorch, and more, and models can be fine-tuned using the Nvidia NeMo framework.

AI applications developed on Project Digits systems can run there, or deployed to more powerful cloud or data centre infrastructure using the same Grace Blackwell hardware and Nvidia AI enterprise software platforms.

This news follows Nvidia's recent announcement of its $US 249 Jetson computer for AI applications, that affordably handles AI models up to eight billion parameters.