- Nvidia reveals supercharged mini PC with Blackwell architecture
- A single Project DIGITS supercomputer can handle 200b LLMs, two can handle 405b
- It’ll cost from $3,000 for access to the GB10 Superchip, with Blackwell GPU and Grace CPU
Nvidia has unveiled Project DIGITS, a brand-new mini PC designed to get AI into the hands of more users through the company’s flagship Blackwell hardware.
With the GB10 Grace Blackwell Superchip, Nvidia promises a petaflop of AI computing performance, making the device ideal for prototyping, fine-tuning and running large AI models as compute demands increase for businesses and consumers alike.
“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,” noted Nvidia CEO Jensen Huang.
Nvidia shoehorns Blackwell Superchip into mini PC
Based on the Grace Blackwell architecture, GB10 Superchip uses a Blackwell GPU with CUDA cores and fifth-generation Tensor Cores and a Grace CPU with 20 power-efficient cores built with the Arm architecture. Nvidia boasts of “best-in-class power efficiency, performance and connectivity.”
Project DIGITS also includes 128GB of unified DDR5X memory and up to 4TB of NVMe storage. A single supercomputer can handle up to 200-billion-parameter large language models, but two connected via NVIDIA ConnectX networking can handle up to 405-billion-parameter models.
Nvidia has set up a dedicated web page for prospective customers to sign up for more information regarding Project DIGITS, dubbed “a Grace Blackwell AI Supercomputer on your desk.” The supercomputer will be available from Nvidia and “top partners” from May, and prices will start at $3,000.
Further information regarding configurations has not yet been confirmed, and it’s unclear whether any further customization can be done beyond SSD storage upgrades.