Troops in the field may soon have supercomputers at their fingertips, thanks to a Pentagon project that’s set to move from prototype to product.
For 18 months, software companies Rescale and Parallel Works have worked to show that they can mix Defense Department supercomputers with cloud services to add capacity and the ability to remotely access high-performance computers. Now the Defense Innovation Unit has announced plans to make these services generally available through the Pentagon’s High Performance Computing Modernization Program.
“Researchers are 1719600042 able to access cloud resources when appropriate to augment their work at on-premises centers. This has given them access to a wider variety of hardware, and the ability to scale resources beyond what is currently possible, all within one secure, easy to use, environment,” said Benjamin Parsons, HPCMP’s chief technology officer, said in a news release on Thursday. “These capabilities are essential to meeting the needs of the DOD scientific and technology, test and evaluation, and acquisition communities who are key to supporting the warfighters across the globe.”
The HPCMP program, which is administered by the Army, spent about $300 million in fiscal year 2023, is spending about $255 million in 2024, and has requested about $240 million for 2025.
“Your laptop probably has four [CPU] cores on it. These people, what they do, is that when you solve the problem, they’ll run something across 10,000 cores,” Matt McKee, the chief operating officer of Rescale, told Defense One on the sidelines of the AWS Summit in Washington, D.C.
In an interview, McKee talked about the “success memos” issued by DIU for the Rescale and Parallel Works products.
D1: What, exactly, is a success memo?
McKee: When you are like a smaller tech company like we are…it’s almost a mountain to climb to get your technology in the hands of the DOD. It feels like an impossible task; you don’t even know where to start.
You need an organization like DIU to say, “You built something that we can really leverage. We like your technology; we think it solves a particular problem.”
And the success memo is the end of a two-year journey, where we have proven a number of milestones to say that our technology serves a very unique and vetted purpose for DOD…You’ve completed a successful prototype and now anybody in the DOD can go and rapidly procure your thing.
What exactly would defense customers be buying?
It’s a high-performance computing platform…Have you ever seen “The Martian”?
With Matt Damon?
So at one point, they were trying to save him and they said,“We’re gonna do this crazy idea where we’re going to use the Earth’s gravitational pull to slingshot [him] back around.” Do you remember the engineer that was in NASA like calculating and like doing all the computation to figure out if that would work? That guy was in there, he’s on his laptop, connected to the high-performance computer, like actually doing the calculation to see whether that slingshot maneuver would work.
NASA, DOE, DOD—they all use high-performance computing to solve really hard problems.
The more you want to use, like AI techniques, you need a ton of computing power.
And we’re helping to create an augmentation to the vast supercomputers they already have…through the cloud to be able to solve really hard problems in science and product development to, basically, address defense requirements.
So what’s an operational example of using your solution?
Let’s take a weather example. You have a number of entities that are deployed into far-flung locations and they’re planning a major operation. They need to do weather modeling to understand…what the impact the weather might have on an operation.
What about the pier in Gaza that was damaged by weather?
That can be a great example of that. What would be the structural integrity of this particular pier? How should we build it?
Potentially, you can call back into your supercomputing center that’s back in the U.S. somewhere. They’ve got a bunch of other stuff they’re already doing. But we need instantaneous resources—like, right now—to solve this particular problem.
[Cloud-based HPC] can provide you that agility to be able to solve mission-critical problems on the spot…We need the ability to spike up, spike down. We need to have that agility. It’s bringing the power of these hyperscalers, like AWS, like Microsoft Azure, to DOD end users.
It’s much more collaborative. It enables you to share across various entities and be able to be accessed from anywhere.
Furthermore, the cloud is constantly staying on the cutting edge of computing architectures….Let’s say you set up a two-year [request for proposal] process to procure a really big supercomputer. By the time you get that thing installed, tested, security-verified, put it on the floor, it’s already a year behind the cutting-edge architecture.
And you have sized it in a certain way that meets the requirements. What you do not anticipate is emerging requirements, emerging needs, right? And so you need an augmentation of that. And cloud is a way to augment what the DOD already does. It’s not a replacement for it. It’s an augmentation to help them stay on the cutting edge of all these diverse requirements that they may need.
Who are your competitors?
Systems integrators. You can try to go build it yourself, get a bunch of software engineers and try to build a secure platform yourself and try to get it approved and accredited by your own cybersecurity entities. What we’re building is a solution that is verified and the credit by the Department of Defense through DISA that essentially negates anybody having to go try and go build this yourself.
The success memo says the next phase is completing the IL5-certified version. We have FedRAMP Moderate and the IL5 platform is in the works as part of the next phase of this collaboration with DIU.
So you’re still working to get IL5 (unclassified) and IL6 (classified to secret)?
IL6 would be sometime in the future.
Do you know when you’ll have IL5?
Fall of this year.
What advice do you have for non-defense companies looking to work with DIU?
What’s key is finding something that DIU is interested in. Because they’ll match-make you…but you still have to be solving a problem. You still have to compete hard because they’ll put out a call. I think there were like 80 applicants to [this solicitation].
This interview was edited for clarity and length.
Read the full article here