NVIDIA and MITRE team up on an AI sandbox to support federal agency research

NVIDIA's headquarters in Santa Clara, Calif. The chipmaker announced a partnership with MITRE to support AI research for federal agencies.

NVIDIA's headquarters in Santa Clara, Calif. The chipmaker announced a partnership with MITRE to support AI research for federal agencies. Andrej Sokolow/picture alliance via Getty Images)

NVIDIA will bring its signature large-scale graphic processing systems to MITRE to help expedite advanced AI solutions for government missions.

NVIDIA is bringing advanced computing capabilities to federal partners with the installation of the company’s DGX SuperPOD, which leverages hardware and software specific to hosting artificial intelligence systems in a bid to advance federal research. 

In a collaboration between chip manufacturer NVIDIA and not-for-profit research corporation MITRE, the former’s DGX SuperPOD  — an AI-specific data center platform composed of up to 32 interconnected nodes and a total of 256 graphics processing units — will be installed physically at MITRE’s Mclean location, but the broader virtual sandbox the collaboration fosters will be hosted digitally in the cloud. 

Bringing federal agencies broad access to sophisticated AI technologies is the heart of the partnership, according to Anthony Robbins, vice president of North American public sector business at NVIDIA.

“The federal government generally lacks that AI infrastructure, which is required to do the work that needs to be done with respect to building out AI applications and getting AI into operations, they generally lack the infrastructure,” Robbins told Nextgov/FCW on Monday. “So by acquiring this DGX AI supercomputer, MITRE has a chance to build out a sandbox for exploration across their six federally-funded research centers.”

The GPUs embedded in the DGX SuperPOD have been designed by NVIDIA for three decades, with an earlier model  initially created to pixelate data for sophisticated visuals in video games. Updated versions of such GPUs work particularly well within AI systems, being able to support large volumes of data and complex mathematical computations, according to Robbins.

Processing power is key for any computational effort. Generally, the more data a given processor can handle, the better its outputs will be. Robbins said that connecting these GPUs with high-speed storage helps with scaling large AI tasks — yielding more sophisticated applications for federal researchers. 

Charles Clancy, senior vice president at MITRE, said that many federal government projects, such as the Department of Energy’s national labs, utilize GPU-powered systems. Rarely do they come together at this scale.

“There are lots of GPUs scattered across the federal government,” he told Nextgov/FCW. “They're not really accessible environments for sort of the big AI that we're trying to do.”

Thanks to the DGX SuperPOD’s large-scale GPU network, some immediate use cases on MITRE’s docket include training frontier models and prototyping for federal mission areas. 

“The federal government has unique missions, unique challenges and unique datasets, you need data holdings that are specific to that federal mission,” Clancy said. “So we're very focused on helping the federal government understand what the opportunity is for AI in support of their missions and being able to prototype it in this environment.”

He added that the work stemming from the DGX SuperPOD can inform future vendor opportunities and requirements, as well as the building of agent-based AI, which features a system that can take a high-level task and decompose it into subtasks and generate results. As with AI and machine learning in general, AI agents are intended to expedite problem-solving capabilities in niche areas.

“Think of it like a really smart summer intern that's fluent in the area that you’re working in,” Clancy said. “So if we can begin to incorporate trusted agent-based AI systems into these things — still managed and used by humans — then we'll be able to dramatically improve the pace and quality of public benefits administered by the U.S. government to the average person.”

Robbins agreed that the DGX SuperPOD’s application potential is vast, and anticipates that beyond MITRE’s starting point on cybersecurity applications leveraging AI technologies, advances in health care, climate solutions and building transportation sandboxes are close behind.

“The limit of what's possible with artificial intelligence is literally every application and every use case across government,” Robbins said. “So the applications are many. But what MITRE will do in the beginning is start in three or four areas. So they will create world class AI infrastructure for the researchers and engineers to focus in three or four areas.”

The delivery of NVIDIA’s DGX SuperPOD system is a large step in the Biden administration’s broader push for more public-private sector collaboration. Robbins said that this will help bridge the gap of simply studying problems and executing solutions, 

“We have rich expertise in building out systems, hardware and software platforms, that allow people to do this incredible work in the area of artificial intelligence and a bunch of use cases and workflows,” he said. “And MITRE brings to the equation really great depth in the business of government. So we can apply the commercial products that we have to their technical expertise and business of government.”