Photo credit: www.theverge.com

Nvidia Introduces Personal AI Supercomputer: Project Digits

If you’ve been in search of a highly efficient personal AI supercomputer, Nvidia has unveiled an exciting option for you. The tech giant recently announced the launch of Project Digits, scheduled for release in May. This innovative solution is powered by the new GB10 Grace Blackwell Superchip, designed to offer significant processing capabilities typically reserved for larger, more energy-consuming systems, now in a compact desktop form.

Project Digits can efficiently manage complex AI models containing up to 200 billion parameters, all while fitting neatly on a desk and operating from a standard electrical outlet. Priced starting at $3,000, its design bears similarities to a Mac Mini, emphasizing both functionality and modern aesthetics.

Nvidia’s CEO, Jensen Huang, stated, “AI will be mainstream in every application for every industry. With Project Digits, the Grace Blackwell Superchip comes to millions of developers. 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.” This sentiment underscores Nvidia’s commitment to democratizing access to advanced AI technologies.

Project Digits resembles a compact PC.

Each unit of Project Digits comes outfitted with an impressive 128GB of unified, coherent memory, far surpassing the typical 16GB or 32GB found in average laptops. Additionally, it boasts up to 4TB of NVMe storage. For those with more demanding needs, dual units can be interconnected to accommodate models with an astonishing 405 billion parameters, like Meta’s Llama 3.1.

The GB10 chip delivers remarkable performance, clocking in at up to 1 petaflop of AI processing power — equating to 1 quadrillion calculations per second at FP4 precision. This design enhances calculation speeds by using approximations. Moreover, the system is built with Nvidia’s cutting-edge CUDA cores and fifth-generation Tensor Cores, all linked through NVLink-C2C to a Grace CPU embedded with 20 efficient Arm-based cores. Collaboration with MediaTek aimed to maximize both power efficiency and performance in the GB10 design.

Specifications of the Digits supercomputer.

Users gain access to Nvidia’s comprehensive AI software ecosystem, which includes development kits, orchestration tools, and an array of pre-trained models from the Nvidia NGC catalog. The platform operates on Linux-based Nvidia DGX OS and is compatible with widely utilized frameworks such as PyTorch, Python, and Jupyter notebooks. Developers can refine their models through the Nvidia NeMo framework and streamline data science processes using Nvidia RAPIDS libraries.

With Project Digits, users can not only develop and test their AI models locally, but they also have the flexibility to deploy them across cloud services or data center infrastructures that utilize the same Grace Blackwell architecture alongside the Nvidia AI Enterprise software platform.

Nvidia has begun to diversify its offerings in the AI sector with various accessible devices, including a $249 version of its Jetson computer aimed at hobbyists and emerging startups. This model, known as the Jetson Orin Nano Super, can support applications up to 8 billion parameters, further representing Nvidia’s initiative to broaden the reach of AI capabilities.

Source
www.theverge.com