Racking Up a Super Celebration: Tech Marks Exascale Day

by Rick Merritt

If you ever need an excuse to unleash your inner geek, try this — October 18 is now National Exascale Day.

It’s not a mega-match for a new weight class of outsized wrestlers. It’s a celebration of a new generation of supercomputers being born.

Exascale supercomputers can calculate more than a quintillion floating-point operations per second or 1018 FLOPS. So, the eighteenth day of the tenth month seemed a logical day to mark this milestone.

You need reach no further than your smartphone to feel the benefits of top supercomputers. They handle complex simulations used to improve materials like lithium-ion batteries so they hold more energy and charge faster.

The beefy systems model how cancer cells work, opening up new ways to attack them. They’re pioneering pollution-free ways to generate energy. They’re even at work to make the morning weather forecast more accurate.

For such reasons, getting to exascale, like breaking the petascale barrier a few years ago, is a milestone in supercomputing that has recently galvanized the industry.

The host of the new holiday is Cray, a legendary company in the history of supercomputing, now part of Hewlett Packard Enterprise. In the last year, Cray won awards to build what are expected to be three of the first exascale-class supercomputers in the world.

The U.S. Department of Energy aims to switch on the massive number crunchers in its national labs starting sometime in 2021. In anticipation, Cray christened the new holiday and organized an online event that includes some of the happy parents-to-be at the expectant labs.

All are invited to this tech baby shower at 11 a.m. ET on Exascale Day. (Watch for the replay.) No need to buy gifts because, after all, what would you buy an exaflopper that has everything

Arguably, the exascale era has already begun. Today’s most powerful supercomputer, the Summit system at Oak Ridge National Laboratory, has racked up a handful of exascale milestones. The 27,648 NVIDIA V100 Tensor Core GPUs in Summit can drive 3.3 exaflops of mixed-precision horsepower on AI tasks

Harnessing some of that oomph, government and academic researchers shared the 2018 Gordon Bell Prize for using AI to determine the genetic roots of being susceptible to opioid addiction and chronic pain. Their work on one of America’s most pressing epidemics pushed the GPUs on Summit to 2.36 exaflops.

In addition, a team of a dozen researchers from U.S. national labs and NVIDIA used neural networks to track with high accuracy extreme weather patterns. One part of their paper released last year describes work on a half-precision version of the DeepLabv3+ neural network running on NVIDIA Tensor Core GPUs that hit a peak performance of 1.13 exaflops

NVIDIA GPUs are now used in 125 of the Top 500 systems worldwide. Beyond Summit, they include the world’s second, sixth, eight and 10th most muscular systems. Over the last several years, designers have increasingly relied on GPU accelerators to propel these big-iron beasts to new performance heights

Traditionally supercomputers are measured by High-Performance Linpack (HPL), a test of how fast a computer can solve a dense system of linear equations. In June, specialists tested on Summit a new metric, HPL-AI, designed for the mixed-precision work that enables efficient work in deep learning. It gave one AI job that involved more than 10 million equations a 3x speedup

So, a new era in computing has arrived. Happy Exascale Day!