AMD is once again showing its incredible prowess in supporting the computing needs of the world as they power everything from Microsoft to Google and also the new Frontier supercomputer.
Sitting in the front seat of exascale computing with Frontier
For one, true exascale computing is finally delivered through the new US-based supercomputer Frontier. Winning the crown of the Top500 list for computing power and Green500 lists for best energy-efficient computing, the entire system located at Oak Ridge National Laboratory delivers a crazy 62.68 gigaflops/watt performance which results in a mixed-precision number of 6.86 exaflops measured in High-Performance Linpack-Accelerator Introspection which is literally never been seen before, although the numbers for general performance are at 1.1 exaflops. As the system readies for more testing and validation, it will start to contribute to scientific research starting in 2022 by partially and going full force starting early 2023.
In case you’re intrigued by the sheer number of cabinets, there are a total of 74 of them comprising 9400 CPUs and 37,000 GPUs to handle all the 3D simulation and ultra-large data set processing and to tame it down to operational temperature, 4 high-powered pumps pushing 25,000 liters of water each and every minute is all it takes.
Microsoft and Google bolstering the power of cloud
Moving on to the partners, Microsoft is continuing its effort in boosting the power of the Azure cloud platform by announcing the installation of AMD’s Instinct MI200 accelerators to power future large scale AI workloads in addition to further optimization worked together by the PyTorch Core Team and the folks over at AMD data centers. As a result, they will join the just announced Azure HBv3 virtual machines that utilize 3rd Gen AMD EPYC processors with 3D V-Cache for all sorts of workloads.
As for the same deployment of the strongest EPYC chips to date, Google is also following suit by leveraging the power of Confidential Computing on the latest Google Compute Engine (GCE) N2D and C2D virtual machines (VMs) that fulfills the requirements of high performance, general computing, and highly optimized. They are available in regions across the globe, including us-central1 (Iowa), asia-southeast1 (Singapore), us-east1 (South Carolina), us-east4 (North Virginia), asia-east1 (Taiwan), europe-west4 (Netherlands) and more while Confidential VMs are available anywhere N2D and C2D machines are available.
