Discovery Supercomputer Set to Redefine Exascale Computing for AI, Quantum, and Fusion Research

At Oak Ridge National Laboratory, the upcoming “Discovery” supercomputer is set to succeed Frontier as a next-generation exascale system, powering breakthroughs in AI, quantum research, and high-performance scientific computing. Designed as a cornerstone of the DOE’s Genesis initiative, Discovery will leverage extreme-scale FP64 precision computing and advanced electronic architectures to tackle problems ranging from fusion energy modeling to large-scale AI training that can’t be solved experimentally.

Transcript

00:00:00 Hey guys, I'm here with Matt Sieger. He's project director for the new discovery supercomputer that's getting built right here at Oak Ridge National Laboratory. So, why don't we start out talking about what is discovery? Sure, discovery is the nation's next evolution in exascale supercomputing capability. It is the designated successor to Frontier. So, Frontier when it debuted

00:00:21 in 2021 was the world's fastest supercomputer, and discovery is going to be the replacement for successor to Frontier. So, this is part of the big Genesis mission that's across all 17 national labs. So, what role will this play? So, discovery is a real foundational element of the Genesis mission. Genesis is not just about AI, it's about tying together AI, HPC, and

00:00:45 quantum, and experimental laboratories all across the DOE complex to make discoveries that we could not do otherwise. And discovery is going to be the cornerstone of this effort. It'll be uh when it's deployed exascale class supercomputer with computational capabilities above and beyond anything else in the in the DOE complex. So, it's going to be a tool which is going to be

00:01:08 used by Genesis to solve all those challenging problems that they want to solve. And what breakthroughs do you hope come from this a system of this scale? Oh, there's so many things that we do with a supercomputer like Frontier or like discovery. Uh so, just for an example, if you look at fusion, right? Some of the work that we want to do in fusion to enable uh a commercially

00:01:28 viable fusion reactor is to optimize the cooling salts around a fusion reactor to produce enough tritium to be able to sustain the fusion reaction. This is something that you can't do experimentally, right? You can't build a fusion reactor and then try different chemistries in the uh salt blanket around it. Um this is something you have to do computationally. Discovery has the

00:01:52 capabilities to do high-precision mathematics, FP64 uh computation on discovery. That's a really fundamental capability. It's something that you have to have if you're going to be doing high-precision scientific calculations like are needed to support the Genesis program, either to generate data for training AI models or to be the agent that does

00:02:15 computations in support of experimental efforts that are tied into Genesis.