HPE is authorized to build the Discovery supercomputer in the US. Discovery will succeed Frontier and is expected to achieve more than eight exaflops of computing power. Along with Lux, there will also be an HPC cluster accessible to a broader audience.
Hewlett Packard Enterprise (HPE) will build two new systems for the Oak Ridge National Laboratory (ORNL) for the US Department of Energy. These include Discovery, an exascale supercomputer succeeding the previously built Frontier by HPE, and Lux, an AI cluster providing researchers with cloud access to AI capabilities.
The US government announced in 2024 that it had a successor in mind for Frontier. The name Discovery was finalized, along with the ambition to achieve a computing power of 8.5 exaflops. This is significant, as Frontier, the world’s most powerful computer, was the first to surpass the one exaflops mark.
HPE Cray
Discovery will be based on the new HPE Cray Supercomputing GX5000 platform. This system combines computing power for both traditional high-performance computing (HPC) and AI workloads. The supercomputer will be complemented by the HPE Cray Supercomputing Storage Systems K3000, a storage solution utilizing DAOS (Distributed Asynchronous Object Storage), open-source software for faster data processing.
According to ORNL, Discovery should increase the productivity of some applications up to tenfold. This will enable the institute to make faster progress in fields such as precision medicine, cancer research, nuclear energy, and aerospace. The installation will also offer opportunities for experiments with quantum computing. The US hopes to have the system operational by early 2028 at the latest.
Lux: HPC for Everyone
In addition to Discovery, HPE will also build Lux, a cluster focused on artificial intelligence. This system is based on HPE ProLiant Compute XD685 servers with direct liquid cooling, AMD Epyc processors, and Instinct MI355X GPUs. Lux provides researchers in the US with a multi-tenant AI platform, offering access to infrastructure for training and inference of AI models.
Both systems are part of a broader strategy by the Department of Energy to strengthen the US position in AI and supercomputing. The construction of Discovery and Lux is being carried out in collaboration with AMD, which supplies processors and GPUs.
HPE states that Discovery is more compact compared to its predecessor Frontier, occupying 25 percent less space per rack and handling 300 percent more input/output operations per storage rack. The entire system is liquid-cooled, which not only benefits energy efficiency but is also necessary given the density of the components in their server cabinets and the heat they generate.