Forward-looking: The Frontier supercomputing system will be built at Oak Ridge National Laboratory (ORNL) and will be more than 5 times faster than the current world's fastest computer. In a joint effort between AMD, Cray, the Department of Energy, and ORNL, the supercomputing world will be getting a new top dog. The system, called Frontier, is targeting over 1.5 exaflops of computing performance. That's 1.5 quintillion, or 1.5 billion-billion, floating point operations every second.
The exact specs for Frontier haven't been released yet, but it will be built using AMD EPYC CPUs, Radeon Instinct GPUs, and Cray Slingshot interconnects. Each node in the system will have one CPU and four GPUs. The supercomputer will be used for many applications in scientific fields including sub-atomic structures, weather, genomics, fusion, and more.
Frontier will cover the size of over 2 basketball courts and will have more performance than the top 160 current fastest supercomputers combined.
The world's current fastest supercomputer, Summit, is also a joint venture between ORNL and the Department of Energy. Summit is comprised of about 2.4 million cores and 2.8PB of memory and has a peak performance of 200 PetaFLOPS.
While traditional supercomputers have employed a large number of relatively low power nodes, Frontier will use fewer nodes but they will be much higher performance. This is designed to reduce the strain on the interconnect network and to make simulations easier to deploy on the supercomputer.
We haven't received any power consumption details, but the system is expected to be operational sometime in 2021 in the 9-figure price range. While this much compute may seem unfathomable, new advances in science will surely be able to make use of it. Each new generation of supercomputer is a chance to one-up the previous fastest computer. At least until the next supercomputer dethrones this one.