Nvidia has announced Project Digits at CES 2025, a compact AI-focused desktop supercomputer designed to make high-performance computing accessible to individual developers, researchers, and students.
Powered by the newly developed GB10 Grace Blackwell Superchip, Project Digits delivers up to one petaflop of AI performance, enough to process models with up to 200 billion parameters. Priced at $3,000, the system is set to launch in May 2025, marking a significant step in Nvidia’s mission to democratize AI technology.
“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.” said Jensen Huang, CEO of Nvidia, during the company’s keynote at CES 2025.
Related: Nvidia Reveals RTX 50-Series Blackwell GPUs with DLSS 4 and Twice the Power of the RTX 4090
The Hardware Behind Project Digits
At the core of Project Digits is the GB10 Grace Blackwell Superchip, developed in collaboration with MediaTek. The chip combines Nvidia’s Blackwell GPU and a 20-core Grace CPU using NVLink-C2C technology, a chip-to-chip interconnect that ensures high-speed communication between the two components.
The Superchip is engineered to deliver one petaflop of AI performance at FP4 precision, an optimized floating-point format that accelerates calculations for machine learning tasks without sacrificing essential accuracy.
In addition to the GB10 chip, Project Digits includes 128GB of LPDDR5X memory and up to 4TB of NVMe SSD storage, providing the speed and capacity required for modern AI workloads. The system also features Nvidia’s ConnectX network adapter, enabling advanced features like RDMA, GPUDirect, and NCCL, which are essential for scaling AI tasks across multiple systems.
The compact design of Project Digits is comparable to a Mac Mini, allowing it to fit seamlessly into a desktop setup. Despite its small size, the device’s capabilities rival those of much larger, power-hungry systems traditionally found in enterprise data centers.
Software Ecosystem and Usability
Nvidia has pre-installed its full AI Enterprise software stack on Project Digits, making it a turnkey solution for AI development. This includes frameworks like PyTorch, Python, and Nvidia’s NeMo framework, which simplifies the fine-tuning of large language models. The system runs on Nvidia’s DGX OS, a Linux-based platform tailored for high-performance computing tasks.
According to Nvidia, Project Digits offers the flexibility to prototype, train, and fine-tune models locally before scaling them to cloud or data center environments. Project Digits allows users to develop and test AI models locally and then deploy them on cloud or local infrastructure.
The system supports a range of applications, from academic research to commercial AI development. A single unit can process models with up to 200 billion parameters, and for larger workloads, two units can be linked to handle models with up to 405 billion parameters, putting it in the same league as enterprise-grade solutions like Meta’s LLaMA 3.1.
Collaboration with MediaTek and the GB10 Superchip
The GB10 chip represents a milestone in Nvidia’s collaboration with MediaTek. MediaTek contributed its expertise in Arm-based system-on-chip (SoC) designs to optimize the chip’s performance and power efficiency. The Grace CPU within the GB10 features 20 cores, split evenly between high-performance Cortex-X925 cores and power-efficient Cortex-A725 cores.
Ashish Karandikar, Nvidia’s Vice President of SoC Products, highlighted the importance of this collaboration: “Our collaboration with Arm on the GB10 Superchip will fuel the next generation of innovation in AI.” This partnership underscores Nvidia’s commitment to innovation in the AI hardware space.
Market Position
Project Digits is part of Nvidia’s broader strategy to make AI development tools accessible to a wide range of users. It follows the release of the Jetson Orin Nano Super, a $249 entry-level AI device targeting hobbyists and startups. Together, these products demonstrate Nvidia’s commitment to serving both professional developers and the growing community of AI enthusiasts.
The launch of Project Digits also aligns with industry trends toward decentralized AI development. By enabling users to prototype and train models locally, the device reduces reliance on cloud-based solutions, which can be expensive and require stable internet connectivity.
