Nvidia’s $3,000 miniature computer for AI developers turns heads at CES

Nvidia CEO Jensen Huang was hailed as a rock star this week at CES in Las Vegas, as the artificial intelligence boom has turned the chipmaker into the world’s second-most valuable company.

In a nearly two-hour keynote speech Monday to kick off the annual technology conference, Hwang packed a 12,000-seat arena and drew comparisons to the way the late Steve Jobs introduced products in the United States. apple event.

Huang concluded with a typical Apple trick: a surprise product announcement. He introduced one of Nvidia’s server racks and used stage magic to hold up a much smaller version that looked like a small cube of a computer.

“This is an AI supercomputer,” said Huang, wearing a crocodile skin jacket. “It runs the entire Nvidia AI stack. All of Nvidia’s software runs on it.”

Huang said the computer, called Project Digits, will run on a relative of the Grace Blackwell graphics processing units (GPUs) that currently power state-of-the-art AI server clusters. GPU is armBased Grace central processing unit, or CPU. Nvidia has teamed up with Chinese semiconductor company MediaTek to develop a system-on-chip called GB10.

CES, formerly known as the Consumer Electronics Show, is typically the place to unveil flashy, futuristic consumer gadgets. At this year’s show, which begins Tuesday and concludes Friday, several companies announced AI integration with home appliances, laptops and even grills. Other major announcements include Lenovo’s laptops with rollable screens that can expand vertically. New robots have also appeared, including a Roomba competitor with a robotic arm.

Unlike Nvidia’s traditional gaming GPUs, Project Digits is not targeted at consumers. Instead, machine learning researchers who want to develop advanced AI but don’t have billions of dollars to build large data centers or buy enough cloud credits; Targeted at small businesses and universities.

“There’s a big hole for data scientists and ML researchers and people who are actively working and actively building things,” Huang said. “You probably don’t need a huge cluster. You’re just developing an initial version of the model and iterating all the time. You could run it in the cloud, but it would just cost a lot more.”

Nvidia says the supercomputer will cost about $3,000 when it goes on sale in May and will be available from Nvidia itself as well as some manufacturing partners. Huang said Project Digits is a placeholder name, indicating that it may change by the time the computer is released.

“If you have a good name, please contact us,” Huang said.

This is a dramatically different kind of product than the GPUs that have driven Nvidia’s historic boom over the past two years. OpenAI, which launched ChatGPT in late 2022, and other AI modelers such as Anthropic are collaborating with leading cloud providers to use Nvidia for its ability to power their most demanding models and compute workloads. of data center GPUs.

Data center sales accounted for 88% of Nvidia’s $35 billion in revenue in the most recent quarter.

Wall Street is focused on Nvidia’s ability to diversify its business to reduce its dependence on a small number of customers buying large AI systems.

“It was a little scary to see NVIDIA launch such a great product at such a low price,” Melius Research analyst Ben Reitzes wrote in a note this week. He said NVIDIA may have “snatched the conversation” with Project Digits, as well as other announcements such as gaming graphics cards, new robotic chips, and a deal with Microsoft. toyota.

Project Digits runs Linux and the same Nvidia software used in the company’s GPU server clusters, and will significantly increase the power of researchers and universities, said David Bader, director of the New Jersey Institute of Technology’s Data Science Institute. states that it has improved.

Bader, who has worked with Nvidia on research projects in the past, said the computer appears capable of processing enough data and information to train the largest and most advanced models. He told CNBC: google, Amazon To get a system with this kind of functionality, some say, “We’ll pay $100 million to build a training supercomputer.”

For $3,000, users can immediately get a product that plugs into a standard electrical outlet in their home or office, Bader said. This is especially interesting for academics, who often move to private companies to gain access to larger and more powerful computers, he said.

“Students who can own one of these systems, which cost about the same as a high-end laptop or gaming laptop, will be able to do the same research and build the same models,” Bader said.

Reitzes said the computer could be Nvidia’s first move into the $50 billion PC and laptop chip market.

“It’s not hard to imagine that it would be easy to do everything ourselves and one day be able to run Windows on our systems,” Reitzes wrote. “But I think they don’t want to step into it too much.”

Hwang did not rule out the possibility when asked about the matter by Wall Street analysts on Tuesday.

He said MediaTek could potentially sell the GB10 chip to other computer makers in the market. He tried to leave a mystery in the air.

“Of course we have a plan,” Huang said.

Spotlight: Nvidia pulls out due to CES expectations