AMD EPYC and Radeon Instinct will form the basis of HPC-cluster with a record performance of 3.1 EFLOPS
But the performance of 3.1 Eflops will be available only for FP32 computations, and for “classic” FP64 this figure will not exceed 274.54 Pflops. But even with such numbers, the system could theoretically claim the second place in the TOP500 ranking, where Fugaku is now leading (1.07 Eflops FP32, 415.5 Pflops FP64), and Summit is on the second line (414 Pflops FP32, 148.6 Pflops FP64).
However, this solution is a distributed cluster with equipment located in Norway, Sweden and Canada. The core will be GPU servers manufactured by Gigabyte based on the second generation AMD EPYC processors and AMD Radeon Instinct accelerators. Specific characteristics have not yet been indicated. A Gigabyte spokesperson notes that the partnership with AMD will be able to design, manufacture, and deliver large volumes of products within a few months, as well as to adapt them to the needs of Northern Data. The tasks for these machines include rendering, machine learning, and AI, among others.
Northern Data, which specializes in the creation of specialized HPC systems, and Gigabyte announced a strategic partnership this summer, which precisely meant the creation of high-performance solutions based on high-density GPU servers. Northern Data operates what the company says is the largest data center focused on HPC applications.
