Nvidia deploys desktop supercomputers at universities and the South Pole

Nvidia, the US chipmaker, says its DGX Spark desktop supercomputers are now in use at universities and research sites around the world, including the IceCube Neutrino Observatory at the South Pole.

There, scientists are running artificial intelligence models on local data in one of the most remote environments on Earth.

The company said the compact systems deliver petaflop-class performance and support AI models of up to 200 billion parameters, with each unit powered by the Nvidia GB10 superchip and the company's DGX operating system.

At the University of Wisconsin-Madison's IceCube observatory, researchers are applying DGX Spark to neutrino experiments that aim to observe parts of the universe invisible to traditional astronomy.

Benedikt Riedel, computing director at the Wisconsin IceCube Particle Astrophysics Centre, said the South Pole's extreme conditions make conventional computing infrastructure difficult to maintain, with relative humidity below 5%, an elevation of 10,000 feet and severely limited power.

"DGX Spark allows us to deploy AI in a compartmentalised and easy fashion, at low cost and in such an extremely remote environment, to run AI analyses locally on our neutrino observation data," Riedel said.

At New York University's Global AI Frontier Lab, the ICARE project runs entirely on a DGX Spark to evaluate AI-generated radiology reports without sending sensitive imaging data to the cloud.

"Being able to run powerful LLMs locally on the DGX Spark has completely changed my workflow," said Lucius Bynum, data science assistant professor and faculty fellow at the NYU Centre for Data Science.

Researchers at Harvard's Kempner Institute are using the systems to study approximately 6,000 genetic mutations linked to epilepsy and to validate workflows before scaling to larger GPU clusters.

Arizona State University has deployed multiple units for campus-wide AI initiatives including robotics perception and search-and-rescue research, while Mississippi State is using DGX Spark as a hands-on teaching platform.

At the University of Delaware, Sunita Chandrasekaran, professor of computer and information sciences and director of the First State AI Institute, described an Ascent GX10 unit powered by DGX Spark as "transformative for research."

The Institute of Science and Technology Austria is using an HP ZGX Nano AI Station based on DGX Spark to train and fine-tune large language models (LLMs) of up to seven billion parameters, aided by 128GB of unified memory.

A Stanford research team reported that DGX Spark delivers performance comparable to large cloud GPU instances, achieving roughly 80 tokens per second on a 120 billion-parameter model from a desktop, while keeping the entire workload local.

Related reading

• NVIDIA and CoreWeave to accelerate AI factories
• Dassault Systèmes and NVIDIA form industrial AI partnership
• Human API reveals man-bites-dog approach to agentic AI

Stanford's Treehacks hackathon, running from 13 to 15 February, will feature DGX Spark units, with a livestream scheduled for 13 February.

Nvidia said the systems integrate with its NeMo, Metropolis, Holoscan and Isaac platforms, and that purchase options are available on the DGX Spark product page.

The Recap

• DGX Spark deployed at universities and the IceCube observatory.
• Each unit supports models of up to 200 billion parameters.
• Treehacks runs 13–15 February and features a DGX Spark livestream.