Highlights:

  • Nvidia’s upcoming GB10 Superchip is a system-on-a-chip featuring its most advanced GPU, the Nvidia Grace Blackwell.
  • Nvidia claims that the GB10 Superchip allows Project DIGITS supercomputers to provide powerful AI computing resources accessible through a standard laptop or PC.

Nvidia Corp. is set to make its most powerful graphics processing units available to everyone with the upcoming launch of Project DIGITS, a “personal AI supercomputer” powered by the soon-to-be-released Nvidia GB10 Grace Blackwell Superchip.

Nvidia unveiled Project DIGITS recently at the CES consumer electronics show in Las Vegas. Scheduled for release in May, Project DIGITS will deliver 1 petaflop of AI computing power, bringing enterprise-level performance to researchers, data scientists, and students directly on their desks. It will enable the prototyping, training, and fine-tuning of advanced large language models, as well as inference, empowering users to develop AI systems with capabilities comparable to applications like ChatGPT.

Nvidia’s forthcoming GB10 Superchip is a system-on-a-chip featuring its most advanced GPU, the Nvidia Grace Blackwell. Equipped with all the essential components for running large-scale AI projects, it boasts dozens of CUDA cores and fifth-generation Tensor Cores, seamlessly integrated with high-performance Nvidia Grace CPUs via the company’s NVLink chip-to-chip interconnect technology.

Nvidia states that the GB10 Superchip empowers Project DIGITS supercomputers to provide substantial AI computing capabilities accessible from a standard laptop or PC. Each system features 128 gigabytes of unified, coherent memory and up to 4 terabytes of NVMe storage—ample resources to run a 200-billion-parameter large language model (LLM). This enables users to develop and experiment with LLMs surpassing the capabilities of OpenAI’s GPT-3 model, which has 175 billion parameters.

The company announced that for users requiring greater computational power, two Project DIGITS AI supercomputers can be connected using Nvidia’s ConnectX networking technology, enabling support for models with up to 405 billion parameters.

Nvidia co-founder and CEO Jensen Huang emphasized during his CES keynote that Project DIGITS is designed to make advanced AI development tools more accessible, enabling more people to contribute to shaping the future. He highlighted the goal of offering affordable access to the Grace Blackwell Superchip for millions of developers.

However, many developers may still be excluded, as Project DIGITS machines come with a starting price of USD 3,000, making them far from “affordable.” Nonetheless, Huang appears confident that many AI developers will find the investment worthwhile.

Huang said, “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.”

Developers will have the capability to create, refine, and test their models on Project DIGITS before deploying them on Nvidia’s DGX Cloud platform, accelerated cloud instances, or their own on-premises data center setups, as stated by the company.

To support developers working with Project DIGITS, Nvidia is offering access to a comprehensive library of AI development tools, including software development kits, frameworks, and prebuilt AI models, all available through the Nvidia NGC Catalog on the Nvidia Developer portal.

Additionally, the company is offering several of its “AI blueprints” and providing free access to its NIM microservices, both of which are available on the same developer portal.