By Dylan Martin
‘The COVID-19 HPC Consortium is the Apollo Program of our time. Not a race to the moon, this is a race for humanity. The rocket ships are GPU supercomputers, and their fuel is scientific knowledge,’ Nvidia exec Ian Buck says of the White House-led effort, which is lending supercomputing resources to accelerate coronavirus research.
Rivals Nvidia and AMD are teaming up with some of the world’s largest cloud service providers to lend supercomputing resources to accelerate research of the novel COVID-19 coronavirus.
The two Santa Clara, Calif.-based companies announced on Monday that they’re joining the COVID-19 High Performance Computing Consortium, a project led by the White House Office of Science and Technology Policy, the U.S. Department of Energy and IBM that is directing compute resources from the world’s fastest supercomputers to support research proposals.
The goal of the consortium, which was formed two weeks ago, is to expedite research on ways to detect, treat and contain COVID-19, which has taken more than 70,000 lives across the world and drastically disrupted the global economy in a matter of months. Other members of the consortium include Amazon Web Services, Microsoft, Google Cloud, Hewlett Packard Enterprise as well as several federal and academic research labs.
Nvidia said the team it has assigned to the consortium, which is being led by data center executive Ian Buck, consists of experts across several critical areas, including artificial intelligence, drug discovery, molecular dynamics, genomics, medical imaging and data analytics.
“The COVID-19 HPC Consortium is the Apollo Program of our time,” Buck said in a statement. “Not a race to the moon, this is a race for humanity. The rocket ships are GPU supercomputers, and their fuel is scientific knowledge. NVIDIA is going to help by making these rockets travel as fast as they can.”
In addition to domain expertise, Nvidia said it will help the consortium with accelerating the rate at which data is ingested and processed and with optimizing performance on supercomputers for coronavirus research. The company is also contributing AI and life-sciences applications it has developed as part of its Nvidia NGC hub for software that can be accelerated using the chipmaker’s GPUs.
Nvidia said many of the supercomputers that are lending compute resources to coronavirus research efforts are using its GPUs. This includes the Summit supercomputer at the DOE’s Oak Ridge National Laboratory, which running on more than 27,000 Volta GPUs and has already identified 77 drug compounds that could potential treat COVID-19.
AMD is also contributing compute resources to the consortium, most notably through cloud instances being offered by Microsoft Azure and Google Cloud that are running on AMD’s EPYC Rome processors. In addition, the company said its Radeon Instinct MI50 data center GPUs are being used for the “Corona” system at DOE’s Lawrence Livermore National Labs to provide an extra two petaflops, or two quadrillion calculations per second, of peak compute power for molecular modeling work.
“We are honored to join the @WhiteHouse supercomputing partnership alongside other industry leaders working together to fight #COVID19,” AMD CEO Lisa Su tweeted Monday. “@AMD is dedicated to [bringing] high performance compute to the brightest minds in the scientific community to fight this pandemic.”
Combined, the supercomputers in the consortium are providing 402 petaflops, or 402 quadrillion calculations per second, to COVID-19 research from more than 105,000 compute nodes running on over 3.5 million CPU cores and 41,000 GPUs, according to the consortium’s website.
Dominic Daninger, vice president of engineering at Nor-Tech, a Burnsville, Minn.-based HPC system integrator that partners with Nvidia and AMD, said his company has seen increased interest from pharmaceutical companies that are looking at HPC applications to accelerate COVID-19 research efforts.
HPC is particularly important to the field of molecular modeling, according to Daninger, because of the high volume and complexity of calculations required to complete models for drug research.
“That takes just an unbelievable amount of horsepower to simulate combining molecules of various compounds together,” he said, adding that HPC clusters offer compute power that is unrivaled by traditional systems. “It would take years to do something like that on a standard workstation.”
Processors like AMD’s EPYC CPUs and Nvidia’s Volta and Tesla GPUs are especially suited for HPC work because their high core counts can allow for more calculations to be done in parallel. It can sometimes mean taking a process that would traditionally take a week to complete and reduce it to a day or less.
“Whenever you can break up these tasks up in parallel you can get to the solution or get higher fidelity sooner,” Daninger said. “Fidelity makes a difference because it increases the accuracy of the model.”
The channel executive said Nor-Tech has been doing consulting and services work with pharmaceutical companies to help them get molecular modeling programs like Gromacs installed and running. The company is also helping customers adjust their systems in reaction to software vendors reverting to free licenses, which has implications on things like job schedulers and resource managers in HPC workloads.
“There’s an awful lot of work that’s been done in that area,” said Daninger.