Nvidia Announces $3,000 Personal Ai Supercomputer Called Digits

If you were looking for your ain individual AI supercomputer, Nvidia has you covered.

The chipmaker announced astatine CES it’s launching a individual AI supercomputer called Project Digits successful May. The bosom of Project Digits is nan caller GB10 Grace Blackwell Superchip, which packs tin processing powerfulness to tally blase AI models while being compact tin to caller connected a array and tally from a modular powerfulness outlet (this benignant of processing powerfulness utilized to require overmuch larger, overmuch power-hungry systems). This desktop-sized strategy tin grip AI models pinch up to 200 cardinal parameters, and has a starting worth of $3,000. The merchandise itself looks a batch for illustration a Mac Mini.

“AI will beryllium mainstream successful each exertion for each industry. With Project Digits, nan Grace Blackwell Superchip comes to millions of developers,” Nvidia CEO Jensen Huang said successful a spot release. “Placing an AI supercomputer connected nan desks of each accusation scientist, AI interrogator and student empowers them to prosecute and style nan spot of AI.”

Each Project Digits strategy comes equipped pinch 128GB of unified, coherent practice (by comparison, a bully laptop mightiness personification 16GB aliases 32GB of RAM) and up to 4TB of NVMe storage. For moreover overmuch demanding applications, 2 Project Digits systems tin beryllium linked together to grip models pinch up to 405 cardinal parameters (Meta’s champion model, Llama 3.1, has 405 cardinal parameters).

The GB10 spot delivers up to 1 petaflop of AI capacity (which intends it tin execute 1 quadrillion AI calculations per second) astatine FP4 precision (which helps make nan calculations faster by making approximations), and nan strategy features Nvidia’s latest-generation CUDA cores and fifth-generation Tensor Cores, connected via NVLink-C2C to a Grace CPU containing 20 power-efficient Arm-based cores. MediaTek, known for their Arm-based spot designs, collaborated connected nan GB10’s betterment to optimize its powerfulness ratio and performance.

Users will too get entree to Nvidia’s AI package library, including betterment kits, orchestration tools, and pre-trained models disposable done nan Nvidia NGC catalog. The strategy runs connected Linux-based Nvidia DGX OS and supports celebrated frameworks for illustration PyTorch, Python, and Jupyter notebooks. Developers tin fine-tune models utilizing nan Nvidia NeMo exemplary and accelerate accusation taxable workflows pinch Nvidia RAPIDS libraries.

Users tin create and proceedings their AI models locally connected Project Digits, past deploy them to unreality services aliases accusation halfway infrastructure utilizing nan aforesaid Grace Blackwell architecture and Nvidia AI Enterprise package platform.

Nvidia offers a scope of akin devices successful nan aforesaid accessibility style — successful December, it announced a $249 type of its Jetson instrumentality for AI applications, targeting hobbyists and startups, called nan Jetson Orin Nano Super (it handles models up to 8 cardinal parameters).