October eighteenth (mathematically written as 10^18) is the day we celebrate the advancement of science and discovery with the fastest supercomputers in the world – such as the Frontier supercomputer at Oak Ridge National Laboratory. National Exascale Day celebrates the scientists and researchers who make breakthrough discoveries in medicine, material sciences, energy, AI and beyond with the use of the fastest supercomputers in the world. Together AMD, HPE, and ORNL transported the world into the era of exascale! The Frontier supercomputer—at the Department of Energy’s Oak Ridge National Laboratory—is the world’s first exascale-class supercomputer, with 1.1 exaflops of performance. Thanks to the close collaboration between AMD, HPE and ORNL, Frontier’s performance is more than double the number two system and greater than the sum of the next seven systems on the latest Top500 list. Not only does this incredible design hold the top spot for computing horsepower as of June 2022, but it is also the world’s most efficient supercomputer system as well – by taking the #1 spot on the Green500 list.
To understand the importance of this advancement in computing horsepower, let’s break down what kind of potential an exascale computer has. For a computing system to achieve exascale performance means that it can perform at least one exaflop ― or one quintillion (a billion billion) calculations per second. The human brain contains 100 billion neurons – to reach one quintillion would require firing all the neurons of 10,000 people! This incredible achievement in computational technology means we can accelerate discoveries – researchers are able to tackle grand challenges that were once thought beyond reach They are able to investigate next-generation sequencing and molecular modeling that leads to new medicine and new medical discoveries. They can better understand how to protect our planet with more accurate climate simulations. This type of exciting scientific promise will make a profound difference in our everyday life.
Looking under the hood of Frontier
Each node inside Frontier contains an optimized 3rd Gen AMD EPYC™ processor and four AMD Instinct™ accelerators – for a total of more than 9,400 CPUs and more than 37,000 GPUs in the entire system. Those AMD Instinct MI200 Series GPUs provide the flexible, high-performance compute engines, high bandwidth memory, and scalable fabric and communications technologies to address the computational challenges faced by exascale supercomputers. Ensuring the cache coherency between thousands of CPUs and GPUs is AMD Infinity Architecture – which provides a high-bandwidth, low latency fabric connection between AMD Instinct GPUs and AMD EPYC CPUs – enabling a unified platform that delivers next-gen performance.
Researchers and scientists can get access to all of Frontier’s computing power via the AMD ROCm™ software platform. AMD ROCm is an open-source platform for HPC-class heterogeneous GPU compute and data center system designs. It incorporates optimized drivers, compilers, libraries, support for OpenMP®, and tools needed for large HPC systems – including the ability to code down to the metal, if desired. Developers can further accelerate their time to science with the AMD Infinity Hub – a library of advanced GPU software containers and deployment guides across HPC, Machine Learning and AI applications. AMD supports developers working on Frontier with access to this open, accelerated platform to keep driving exascale supercomputing forward.
Before Frontier, the challenges ORNL was asked to address had become increasingly difficult to answer with high confidence without increased compute power. They needed more predictive codes that run with higher fidelity models. AMD Instinct accelerators are designed to do just that; exceling at AI and ML-based data analysis, enabling models that are fast, accurate, and capable of finding valuable insights.
Together we advance the future of exascale
AMD and HPE, in conjunction with the U.S. Department of Energy’s Lawrence Livermore National Laboratory, are looking to continue pushing the boundaries of supercomputing with the design of El Capitan. This system, which will target over 2 exaflops of double-precision processing power, is expected in early 2023. El Capitan will use next generation AMD Instinct™ MI300 accelerators for high-bandwidth memory and fast I/O sub systems. The MI300 accelerator will leverage a 3D chiplet design combining AMD CDNA™3 GPU cores, “Zen 4” CPU cores cache memory and HBM chiplets all on the same silicon. This design will help the El Capitan achieve the advanced scale of 3D simulations that are essential to meet the demands of the NNSA Life Extension Programs (LEPs) and national security.
Together, AMD and HPE help solve the world’s toughest challenges through the use of compute power – enabling unprecedented time-to-insight for crucial simulations. We celebrate Exascale Day for its focus to advance science and move the future forward. Welcome to Exascale Day!