Oak Ridge National Laboratory (ORNL) has issued a request for proposals (RFP) for a successor to the Frontier supercomputer, just a couple of years after the world's first exascale system came online.

In a letter sent to relevant interested parties, UT-Battelle, the not-for-profit org that manages ORNL for the US Department of Energy (DoE), invited proposals for its next generation of high performance computing (HPC) system, OLCF-6 - which is to be known as Discovery.

The letter states that OLCF-6 is expected to be delivered in late 2027 to early 2028 and accepted in late 2028, with an anticipated budget of $500 million. Vendors have until August 30 to submit their proposals.

Frontier is still officially the fastest supercomputer in the world, at least when measured by the High-Performance Linpack (HPL) benchmark, at 1.206 exaFLOPS. It is configured with 9,472 of AMD's third-gen Epyc CPUs and 37,888 of its Instinct MI250X GPUs, plus the Cray HPE Slingshot interconnect fabric.

ORNL believes that by calendar 2028, the Frontier supercomputer will be nearing end of life, so a successor is needed soon.

Unlike previous RFPs issued by the Oak Ridge Leadership Computing Facility (OLCF), this one does not include a specific performance goal, but the facility said it expects Discovery to deliver "three to five times more computational throughput for benchmarks and scientific applications than Frontier."

"This project is exciting because we will be building something even more capable than Frontier, with technologies that will push the edge of what's possible," OLCF project director for Discovery Matt Sieger said in a statement.

The RFP lays out a range of goals for this next-generation system, including the inevitable advanced artificial intelligence and machine learning performance, but also more traditional modeling and simulation capabilities, plus improved energy efficiency and interoperability with other DoE facilities through the Integrated Research Infrastructure (IRI) initiative.

According to ORNL, all of Discovery's cutting-edge capabilities will be required to meet the future mission needs of the Advanced Scientific Computing Research (ASCR) program within the DoE's Office of Science. This supports research at hundreds of US institutions, establishing open-access supercomputing and network facilities at DoE national laboratories.

Unspoken is the recognition that the US is in an arms race of sorts with China to have the most capable supercomputers for simulating and developing technologies such as nuclear weapons or fighter jets. It is suspected that Beijing has several supercomputer systems capable of 1 exaFLOPS or more, but the country is notoriously secretive about its exact capabilities in this respect, leaving US researchers keen to ensure they have the edge.

Energy efficient computing will also continue to be a priority at OLCF, especially in light of recent stories about the burgeoning power consumption of HPC systems to meet the demands of AI workloads. OLCF said it has improved the facility's computational power by a factor of 500 over the past decade, while energy consumption has increased only by a factor of four.

But a question mark hangs over who is likely to bid to build OLCF-6/Discovery, other than HPE and its Cray subsidiary, as was discussed last year by our colleagues over at The Next Platform. The company was a partner in building not only Frontier, but also the Intel-based Aurora supercomputer and the upcoming El Capitan system at Lawrence Livermore National Laboratory.

The RFP letter hints that a cloud-based solution is possible, which means either Microsoft or AWS might be in the running.

"If the awardee is a traditional system integrator, the company (UT-Battelle) intends to pay for its machine by using awardee-provided financing or exercising an option to use third-party Lease-To-Own (LTO) financing," the letter states, but adds: "The company expects cloud proposals to include financing in their service price." ®

https://www.theregister.com//2024/07/24/oak_ridge_discovery/