NVIDIA is revolutionizing personal computing with the unveiling of Project DIGITS, an innovative personal AI supercomputer designed for AI researchers, data scientists, and learners globally. This groundbreaking release was announced by NVIDIA at the recent CES event, where they introduced the GB10 Superchip, a cornerstone of the Project DIGITS platform.
The NVIDIA GB10 Grace Blackwell Superchip delivers an impressive petaflop of AI computing power, ideal for prototyping, refining, and executing large AI models. What distinguishes Project DIGITS is its capacity to run on a standard desktop setup, allowing model development and inference right on your desk, with effortless transitions to deployment on enhanced cloud or data center systems.
“AI’s role is expanding into every sphere of industry, and with Project DIGITS, we are placing the potent Grace Blackwell Superchip into the hands of millions of developers,” remarked Jensen Huang, CEO of NVIDIA. “By doing so, we are empowering the next wave of AI-driven innovation directly from the desktops of researchers, students, and data scientists.”
The GB10 Superchip is a marvel of engineering prowess, based on the NVIDIA Grace Blackwell architecture. It boasts up to 1 petaflop of AI performance at FP4 precision and includes a powerful NVIDIA Blackwell GPU equipped with advanced CUDA cores and Tensor Cores. Its connectivity is ensured through NVLink-C2C, pairing it seamlessly with a high-performance NVIDIA Grace CPU featuring 20 power-efficient cores built with Arm architecture. This exceptional design, achieved in collaboration with MediaTek, ensures superior power efficiency and connectivity.
With Project DIGITS, AI developers have access to 128GB of unified memory and up to 4TB of NVMe storage, empowering them to run models with up to 200 billion parameters. Via NVIDIA ConnectX networking, a system can double its capabilities by connecting two Project DIGITS AI supercomputers, handling models that demand up to 405 billion parameters.
The Grace Blackwell architecture is a game-changer, enabling enterprises and researchers to develop and test AI models locally on Project DIGITS before scaling them across cloud or data center setups. This consistent architecture, coupled with the NVIDIA AI Enterprise software platform, facilitates smooth transitions from experimentation to large-scale deployment.
NVIDIA’s Project DIGITS users can enrich their development experience by tapping into a vast array of AI software tools. The NVIDIA NGC catalog and developer portal offer an extensive selection of SDKs, frameworks, and models, bolstering experimentation capabilities. For developers eager to fine-tune their models, NVIDIA NeMo provides robust support, while NVIDIA RAPIDS accelerates data science work. Commonly-used frameworks such as PyTorch and Python, along with tools like Jupyter notebooks, are easily accessible.
For applications moving from the development stage to production, NVIDIA’s AI Enterprise license offers robust tools for enterprise-grade security and support, ensuring seamless integration and reliability.
Enthusiasts and professionals eager to jump on board can look forward to the availability of Project DIGITS in May, starting at a price point of $3,000. This launch promises to democratize access to high-level AI computing power, bringing the future of AI development to desktops worldwide.
