Cray, AMD to build 1.5 exaflops supercomputer for US government
AMD and Cray have announced that they are building "Frontier", a new supercomputer for the Department of Energy at the Oak Ridge National Laboratory. The goal is to deliver a system that can perform 1.5 exaflops: 1.5 × 10 18 floating point operations per second.
By comparison, a single Nvidia RTX 2080 GPU administers about 14 teraflops of computational performance with 32-bit numbers. Frontier will achieve 100,000 times more. The fastest supercomputer in the Top 500 list weighs in on 200 petaflops or 0.2 exaflops. As things stand, it will take the top 160 machines on the list to match Frontier's performance.
Frontier will use custom versions of AMD's Epyc processors (probably Zen 3 or Zen 4), matched with 4 GPUs, all connected to AMD Infinity Fabric. Between nodes, Crays Slingshot interconnect will be used, which has transmission rates up to 200 Gbps per port. The GPUs will have their own stacked HBM (High Bandwidth Memory). It is placed in 100 cabinets and takes about 7,300 square meters of floor space. The power consumption will be 30-40MW.
The plan is that Frontier will be delivered in 2021, at a cost of around $ 500 million for hardware and $ 100 million for research and development. It should be the fastest supercomputer in the world when delivered and it will be the US government's second exaflops capability. The first will be 1 exaflops Aurora, built with Intel Xeon SP processors and Intel Xe GPUs.
The supercomputer is made available to academics to run a wide range of simulations and experiments. Radeon Instinct GPUs include hardware dedicated to machine learning workloads, and Frontier is likely to be used for this type of task in addition to the more conventional weather and nuclear weapon simulations.
