Two simulations of one billion atoms, two new insights into how the SARS-CoV-2 virus works and a new AI model to accelerate drug discovery.
These are the results of the finalists in the Gordon Bell Awards, considered the Nobel Prize for High Performance Computing. They used AI accelerated computing or both to use NVIDIA’s technologies to advance science.
A finalist for the special award for COVID-19 research used AI to link multiple simulations to show how the virus replicates in a host on a new level.
The research – led by Arvind Ramanathan, a computational biologist at Argonne National Laboratory – offers an opportunity to improve the resolution of traditional tools for studying protein structure. That could provide new insights into how to stop a virus from spreading.
The team, made up of a dozen organizations in the US and UK, designed a workflow that ran across systems, including Mother-of-pearl, an NVIDIA A100-based system from Hewlett Packard Enterprise and Argonnes NVIDIA DGX A100 systems.
“The ability to conduct data analysis and simulations at multiple sites for integrative biology will be invaluable in harnessing large experimental data that are difficult to transfer,” the paper says.
As part of their work, the team developed a technique to accelerate molecular dynamics research using the popular NAMD program on GPUs. They also took advantage NVIDIA NVLink Accelerating data “well beyond what is currently possible with a traditional HPC network connection or … PCIe transfers.”
One billion atoms in high fidelity
Ivan Oleynik, professor of physics at the University of South Florida, led a team named a Standard Gordon Bell Prize finalist for their work on the first high-precision simulation of one billion atoms. It broke at 23x a record from a Gordon Bell winner last year.
“It’s a pleasure to uncover unprecedented phenomena, it’s a really great achievement that we’re proud of,” said Oleynik.
Simulating carbon atoms under extreme temperatures and pressures could open doors to new sources of energy and help describe the composition of distant planets. It’s especially impressive because the simulation has quantum-level accuracy and faithfully reflects the forces between atoms.
“We could only achieve this level of accuracy by applying machine learning techniques to a powerful GPU supercomputer – AI is revolutionizing the way science is done,” said Oleynik.
The team ran 4,608 IBM Power AC922 servers and 27,900 NVIDIA GPUs on the US Department of Energy summit Supercomputers built by IBM, one of the most powerful supercomputers in the world. It showed that their code can scale to simulations of 20 billion atoms or more with almost 100 percent efficiency.
This code is available to any researcher who wants to push the boundaries of materials science.
In a deadly droplet
In another billion atom simulation, a second finalist for the COVID-19 prize showed the delta variant in a droplet in the air (below). It shows biological forces spreading COVID and other diseases, and offers a first-hand look at aerosols at the atomic level.
The work has “far-reaching … implications for virus binding in the deep lungs and for the study of other airborne pathogens,” according to the paper by a team led by last year’s winner of the special award, researcher Rommie Amaro from the University of California San Diego.
Amaro’s team simulated the Delta SARS-CoV-2 virus in a breath droplet with more than a billion atoms.
“We show how AI, coupled to HPC at multiple levels, can lead to significantly improved effective performance and enables new ways to understand and study complex biological systems,” said Amaro.
Researchers used NVIDIA GPUs on Summit, the Longhorn supercomputer built by Dell Technologies for the Texas Advanced Computing Center, and commercial systems in the cloud from Oracle.
“HPC and cloud resources can be used to dramatically reduce time to resolution for large scientific endeavors, connect researchers, and enable highly complex collaborative interactions,” the team concluded.
The language of drug discovery
Finalists for the Oak Ridge National Laboratory (ORNL) COVID Prize have applied Natural language processing (NLP) on the problem of screening chemical compounds for new drugs.
They used a data set of 9.6 billion molecules – the largest data set used for this task to date – to train in two hours a BERT-NLP model that can accelerate drug discovery. Efforts to date have taken four days to train a model on a data set of 1.1 billion molecules.
The work put more than 24,000 NVIDIA GPUs on the Summit supercomputer to deliver a whopping 603 petaflops. Now that training is complete, the model can be run on a single GPU to help researchers find chemical compounds that could inhibit COVID and other diseases.
“We have employees here who want to apply the model to cancer signaling pathways,” says Jens Glaser, computer scientist at the ORNL.
“We’re just scratching the surface of training data sizes – we hope to be able to use a trillion molecules soon,” said Andrew Blanchard, a research scientist who led the team.
Rely on a full-stack solution
NVIDIA software libraries for AI and accelerated computing helped the team complete their work in a surprisingly short amount of time that one observer called.
“We didn’t have to fully optimize our work for the GPU’s tensor cores, because you don’t need any special code, you can simply use the standard stack,” said Glaser.
He summed up what many finalists felt: “Having the chance to be part of meaningful research that has a potential impact on people’s lives is very rewarding for a scientist.”
Tune in to ours special address at SC21 either live on Monday, November 15 at 3 p.m. PST or later on request. NVIDIA’s Marc Hamilton will provide an overview of our latest news, innovations, and technology, followed by a live Q&A panel with NVIDIA experts.
Disclaimer of liability
Nvidia Corporation published this content on November 15, 2021 and is solely responsible for the information contained therein. Distributed by Public, unedited and unchanged, on November 15, 2021 14:34:23 UTC.