ORNL seeks proposals for Frontier supercomputer successor • The Register

Oak Ridge National Laboratory (ORNL) has announced a contract for a successor to its Frontier supercomputer, just years after launching the world’s first exascale system.

In a letter sent to interested parties, UT-Battelle, the nonprofit that manages ORNL for the U.S. Department of Energy (DoE), invited proposals for its next-generation high-performance computing (HPC) system, OLCF-6 — to be known as Discovery.

The letter said OLCF-6 is expected to be delivered in late 2027 or early 2028 and accepted in late 2028, with an expected budget of $500 million. Suppliers have until August 30 to submit their bids.

The Frontier is still officially the fastest supercomputer in the world, at least according to the High-Performance Linpack (HPL) benchmark, with a score of 1,206 exaFLOPS. It is configured with 9,472 third-generation AMD Epyc processors and 37,888 Instinct MI250X GPUs, as well as a Cray HPE Slingshot interconnect.

ORNL believes the Frontier supercomputer will reach the end of its life cycle in calendar year 2028, so a replacement will need to be found 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 did say it expects Discovery to deliver “three to five times more compute 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 boundaries of what is possible,” Discovery’s OLCF project director Matt Sieger said in a statement.

The call for proposals outlined a number of goals for this next-generation system, including the inevitable advanced AI and machine learning performance, but also more traditional modeling and simulation capabilities, as well as increased 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 needed to meet future mission needs of the DoE’s Office of Science Advanced Scientific Computing Research (ASCR) program. This supports research at hundreds of U.S. institutions by creating open supercomputing and networking facilities at DoE national laboratories.

There is an unspoken recognition that the U.S. is in something of an arms race with China to have the most capable supercomputers to simulate and develop technologies such as nuclear weapons or fighter jets. Beijing is suspected of having several supercomputer systems capable of 1 exaFLOPS or more, but the country is notoriously secretive about its exact capabilities in this regard, leaving American scientists eager to gain an advantage.

Energy-efficient computing will continue to be a priority at OLCF, especially in light of recent reports of HPC systems escalating power consumption to meet the demands of AI workloads. OLCF said it has increased the facility’s computing power by a factor of 500 over the past decade, while power consumption has only quadrupled.

However, the question arises as to who is likely to bid for the OLCF-6/Discovery build, other than HPE and its subsidiary Cray, as discussed by our colleagues from Next platformThe 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 suggests a cloud-based solution is possible, meaning Microsoft or AWS could be in the running.

“If the beneficiary is a traditional systems integrator, (UT-Battelle) intends to pay for its machine using financing provided by the beneficiary or through the option of using Lease-To-Own (LTO) financing from a third party,” the letter reads, but adds, “The company expects that cloud solution proposals will include financing as part of the price of the service.” ®