The US government is going to get a new supercomputer built by AMD and Cray. They have announced that they’re working on their new project called Frontier. This new supercomputer is made for the Department of Energy at Oak Ridge National Laboratory. The goal of this supercomputer is to perform 1.5×1018 floating-point operations per second – 1.5 exaflops
- More Powerful than Nvidia RTX 2080 GPU?
Can it run Crysis, though? Probably five of them at the same time. For comparison, the Nvidia RTX 2080 GPU can perform 14 teraflops with 32-bit numbers. The new supercomputer will have 100,000 times more! The strongest supercomputer has 0.2 exaflops and compared to Frontier, it isn’t anywhere close.
- Taking Advantage of ZEN Platform
The new Frontier will use some of the Epyc processors from Ryzen. The speculations are that this might be ZEN 3 or 4, combined with 4 GPUs, connected with Infinity Fabric. This is where Cray’s Slingshot interconnect comes into play. It has up to 200Gb/s transfer rate per single port. The GPUs will take advantage of the fastest HBM, a type of memory that AMD used in their older Fury GPUs.
To grasp the idea of its size, it’s going to take up to 7,300 square feet of floor space or around 100 cabinets. The power consumption will be at 30-40 MW.
- The Purpose of Frontier
As with every supercomputer, Frontier will be able to run a variety of experiments and simulations. The dedicated Instinct GPUs will be able to take advantage of machine-learning workloads, meaning that the supercomputer could be used for more advanced tasks, such as weapon simulations.
Frontier will be delivered in 2021 and its cost will be $500 million just for the hardware. The whole research and development process will cost $100 million. When it’s released, it should be the fastest supercomputer in the world and second exaflops system in the US government. Source: arstechnica