NVIDIA Unveils Grace Blackwell Supercomputer
NVIDIA made a significant announcement at CES, unveiling Project DIGITS, a personal AI supercomputer designed to empower AI researchers, data scientists, and students with access to the transformative technology of the NVIDIA Grace Blackwell platform. This innovation features the new NVIDIA GB10 Grace Blackwell Superchip, promising a remarkable petaflop of AI computing performance tailored for prototyping, fine-tuning, and running extensive AI models.
Project DIGITS: Accessibility Meets Performance
With Project DIGITS, users can harness the power of AI directly from their desktop systems, facilitating the development and inference of models that can subsequently be deployed on more robust cloud or data center infrastructures. “AI will be mainstream in every application for every industry. With Project DIGITS, the Grace Blackwell Superchip comes to millions of developers,” proclaimed Jensen Huang, NVIDIA’s founder and CEO. This statement underscores the aim of making advanced AI capabilities universally accessible.
The Powerful GB10 Superchip: A Technical Breakdown
The GB10 Superchip is a sophisticated system-on-a-chip (SoC) that integrates cutting-edge NVIDIA technology, capable of delivering up to 1 petaflop of AI performance at FP4 precision. This superchip combines an NVIDIA Blackwell GPU featuring state-of-the-art CUDA® cores and fifth-generation Tensor Cores with a high-performance NVIDIA Grace™ CPU composed of 20 power-efficient cores designed on Arm architecture. Collaborative efforts with MediaTek have led to GB10’s design breakthroughs in power efficiency and performance.
Environmentally Friendly and Accessible AI
Project DIGITS is designed to operate efficiently, drawing power from a standard electrical outlet. Each unit is equipped with 128GB of unified memory and up to 4TB of NVMe storage, enabling users to execute large-scale models of up to 200 billion parameters, which can accelerate AI innovation significantly. Moreover, by leveraging NVIDIA ConnectX® networking technology, two Project DIGITS systems can be interconnected to run models of up to 405 billion parameters.
Seamless Integration into Workflows
With the Grace Blackwell architecture, both enterprises and researchers can prototype, fine-tune, and test their models on local Project DIGITS systems that run the Linux-based NVIDIA DGX OS. This integration facilitates a smooth transition from local development to deployment on NVIDIA DGX Cloud™ or data center infrastructures, utilizing the same powerful architecture.
A Comprehensive Suite of Tools for Developers
Project DIGITS users will gain access to an extensive library of NVIDIA AI software, supporting experimentation and prototyping. From software development kits to orchestration tools and frameworks available on the NVIDIA NGC catalog, developers can enhance their workflows with resources like NVIDIA RAIPS™ libraries, NVIDIA Nemoframework, and various common frameworks including PyTorch and Jupyter notebooks.
Launching Soon
For those looking to dive into AI development, Project DIGITS is set to be available starting in May, offered at a competitive price of $3,000. Interested individuals can sign up for notifications to stay updated on this groundbreaking technology.