May 7th, 2019 by Michael Rink
AMD & Cray Secure 9-digit DOE Contract
Today AMD (Advanced Micro Devices) and Cray announced that they had a contract to deliver what will be the world’s fastest supercomputer to Oak Ridge National Laboratory for the DOE (Department of Energy). While rarely taking top place, AMD has been one of the biggest names in processing and graphics chips. Cray used to be a household name, and while its successes rarely make it into mainstream media anymore, it remains one of the premier providers of supercomputers as it has been since 1976.
Oak Ridge National Laboratory (ORNL) is the largest US Department of Energy science and energy laboratory. While most people who are familiar with Oak Ridge National Laboratory likely associate it with its nuclear research, it also works hard to sustain US leadership in high-performance computing. It is currently home to the world’s fastest supercomputer known as Summit. Summit has a peak performance of 200 petaFLOPS (one thousand million million FLoating-point Operations Per Second), which is almost twice that of the second place supercomputer, which is also owned and operated by the USA. The third-place supercomputer, Sunway TaihuLight, has virtually identical performance and is operated by China. The new supercomputer, named Frontier, that AMD & Cray have been contracted to build will blow them all out of the water with a targeted performance of 1.5 exaFLOPS (one billion billion FLOPS) which would be more than 7 times faster than the current fastest supercomputer and an entire order of magnitude faster than the second and third place supercomputers.
China is expected to deploy its own exascale supercomputer in 2020, the Tianhe-3. Current speculation anticipates Tianhe-3 to be capable of only 1.29 exaFLOPS, which means Frontier's 1.5 exaFLOPS will edge it out to reclaim the top spot for the USA when it is completed in 2021. Researchers at Oak Ridge National Laboratory will use the Frontier system's unprecedented computing power, and next-generation AI techniques to simulate, model and advance understanding of the interactions underlying the science of weather, sub-atomic structures, genomics, physics, and other important scientific fields.
Frontier will be built using both AMD EPYC CPUs and AMD Radeon Instinct GPUs. Each node of the supercomputer will be composed of four AMD Radeon Instinct GPUs connected to a single AMD EPYC CPU by custom high-bandwidth, low-latency coherent Infinity Fabric. To organize and manage all of the nodes, Cray is developing an enhanced version of the open source ROCm programming environment.