Nvidia debuts Project DIGITS, a palm-sized AI supercomputer that can sit on any desk

Nvidia will make its most powerful GPUs available to everyone with its upcoming launch Project DIGITS – “personal supercomputer with artificial intelligence”, running on the basis of a ready-to-launch Nvidia GB10 superchip Grace Blackwell.

The artificial intelligence supercomputer was announced late today at the CES consumer electronics show in Las Vegas. When the DIGITS project launches in May, it will provide users with access to 1 petaflops of AI compute performance, bringing researchers, data scientists and students the performance typically limited to large enterprises, right to their desktops. It will support prototyping, training and fine-tuning of advanced large language models, as well as inference, allowing anyone to develop artificial intelligence systems that can match the power of applications like ChatGPT.

Nvidia’s upcoming GB10 super chip is a system-on-chip based on most advanced graphics processor, NVIDIA Grace Blackwell. It has all the supporting components needed to run large-scale AI projects: dozens of CUDA cores and fifth-generation Tensor Cores connected to equally high-performance Nvidia Grace CPUs using NVLink’s inter-chip interconnect technology. .

According to Nvidia, the GB10 superchip enables Project DIGITS supercomputers to deliver massive artificial intelligence computing power to anyone through a standard laptop or PC. Each machine will provide access to 128 gigabytes of single coherent memory and up to 4 terabytes of NVMe storage. This is more than enough to run an LLM with 200 billion parameters, meaning users will be able to create and experiment with LLMs that exceed the capabilities of the OpenAI GPT-3 model, which has 175 billion parameters.

For those who need even more power, two Project DIGITS AI supercomputers will be able to connect together using Nvidia ConnectX networking technology to support models with up to 405 billion parameters, the company says.

Nvidia co-founder and CEO Jensen Huang (pictured with the GB10 superchip and AI supercomputer) said during a keynote at CES that the goal of the DIGITS project is to make the tools for advanced AI development more accessible so that everyone can participate in building the future. He explained that the goal is to provide affordable access to Grace Blackwell’s superchip to millions of developers.

However, millions of developers could still be left behind as Project DIGITS machines aren’t exactly what you’d call “cheap” with prices starting at $3,000. But Huang seems to think there are plenty of AI developers willing to pay that kind of money.

“Putting an AI supercomputer on the desks of every scientist, AI researcher and student gives them the opportunity to participate and shape the AI ​​era,” Huang said.

These developers will be able to build, customize and test their models in Project DIGITS before deploying them to the Nvidia DGX cloud platform, accelerated cloud instances, or their own on-premises data center infrastructure, the company said.

Holger Mueller from Constellation Research Inc. stated that Project DIGITS represents Nvidia’s attempt to increase its influence in the artificial intelligence hardware market, targeting entry-level users as well as the largest enterprises. While this is not unique, what is new is that Nvidia is trying to do this at the same time.

“Blackwell’s enterprise GPUs are just now starting to ship, and the company has already shifted its focus to the lower end of the market with Project DIGITS,” the analyst said. “But this isn’t all that surprising since Nvidia is known for breaking away from semiconductor industry norms, as we saw when it previously introduced new GPUs annually instead of following a two-year update cycle like its competitors.”

Mueller said that if Nvidia can succeed with DIGITS, it will potentially change the artificial intelligence industry.

“Traditionally in AI, bigger is better, leading to questions about whether the technology is cost-effective and can be easily used in remote locations,” he noted. “But if DIGITS becomes popular, it will essentially end such questions, and as a result, Nvidia GPUs will be everywhere.”

To help developers using Project DIGITS, Nvidia said it will provide them with access to an extensive library of AI development tools, including software development kits, frameworks and pre-built AI models, all of which can be found in the Nvidia NGC catalog on the Nvidia Developer site. portal.

In addition, the company produces a number of its “AI Blueprints” is available, along with free access to its NIM Microserviceswhich can be found on the same developer portal.

Photo: Nvidia/live broadcast