Argonne Gets New Supercomputing Cluster to Power Further COVID-19 Vaccine and Drug Research
![](https://cdn.nextgov.com/media/img/cd/2020/05/15/NGsuper20200515/860x394.jpg?1627425914)
Dmitriy Rybin/Shutterstock.com
The national laboratory received NVIDIA nodes to boost performance as researchers work on new methods of testing, treating and monitoring coronavirus patients.
Supercomputers across the nation are powering up research in the fight against the novel coronavirus—and Argonne National Laboratory is adding a new machine to the mix to deliberately zero in on COVID-19.
During a phone briefing with reporters Wednesday, NVIDIA’s Vice President of Health Care Kimberly Powell unveiled the delivery of “the fastest” artificial intelligence-focused supercomputer to the Illinois-based lab, which will be used in scientists’ search for new drugs and vaccines as they race to halt the pandemic. Powell also detailed other news, including that the tech company collaboratively developed with the National Institutes of Health pre-trained AI models to improve the detection of COVID-19 through medical imaging.
“Everyone is really coming together to try to understand and discover new ways of testing, treating, and monitoring what's happening in the COVID-19 pandemic,” she explained at the top of the call.
The latest ‘AI supercomputer’ to launch at Argonne will eventually be composed of a cluster of 24 of NVIDIA’s DGX A100 nodes, and though not all have been delivered yet, parts of the system are already operating. Each individual DGX A100 provides 5 petaflops of performance, meaning it will have the capability to run an additional 120 petaflops of computing power once its fully turned over to the lab. The system will be used to boost Argonne’s existing efforts to better understand the virus—and in support of the COVID-19 High Performance Computing Consortium that was recently established by the administration, IBM and Energy Department—which, for perspective, encompassed 16 supercomputing systems that represented around 330 petaflops of power at first launch. During a virtual keynote for the company’s 2020 GTC event this week, NVIDIA’s CEO Jensen Huang unveiled the powerful A100 chips that make up the machine.
NVIDIA delivered the first three nodes of the system on May 6, and installation was initiated on the following day, an Argonne spokesperson told Nextgov Friday. Those initial pieces were operational for the lab’s COVID-19 team by May 9 and the rest of its nodes and management servers “are due to arrive in June,” the official said. The new system is currently being integrated into Argonne’s Theta supercomputer, so it’s been aptly named “Theta-GPU,” according to the official.
The DGX A100s are built to specifically support AI-focused workloads and, in the announcement of its delivery, Argonne’s Associate Laboratory Director for Computing, Environment and Life Sciences Rick Stevens said the systems’ compute power “will help researchers explore treatments and vaccines and study the spread of the virus, enabling scientists to do years’ worth of AI-accelerated work in months or days.” And the already-in-use nodes are presently running AI models and simulations, the lab’s spokesperson said. To help visualize the computational power during the call with reporters, Powell explained that the system is capable of screening 1 billion small molecule drugs in the “unprecedented” timescale of 24 hours. Through “critical applications called molecular dynamics simulations,” she said researchers will seek to find specific molecules that they can use to interfere with the virus protein that binds with cells—and ultimately block them from binding. Once the right molecules are identified, officials can quickly move the drug candidates into experimentation and clinical trials.
In a matter of three weeks, NVIDIA recently built its own in-house supercomputer that incorporates 50 of the DGX A100 nodes and it, too, will support pandemic-related research. The company is also selling the systems, and its press release details that other “early adopters” include the University of Florida, the German Research Center for Artificial Intelligence, Thailand’s "top research-intensive" institution Chulalongkorn University—and more. NVIDIA’s announcement indicates its first system was delivered to Argonne, and that the price of each DGX A100 starts around $199,000. The lab’s spokesperson confirmed that Argonne “paid for the machine,” and added that it “will be transitioned to general open scientific research post-pandemic, with continued focus on the integration of AI and science.”
Powell also revealed other announcements in the briefing, including that the powerful system has helped the company and a partner “create a speed record” in genomic analysis. NVIDIA-built GPUs were applied to Oxford Nanopore Technologies handheld sequencers and enabled researchers to sequence the viral genome in “less than seven hours.” Such work can help officials better understand how the virus is migrating, and mutating, Powell said.
Further, the company also worked directly with the NIH Clinical Center to co-develop two pre-trained AI models to spot and forecast COVID-19—one focuses on lung segmentation and the other uses CT-scanned imagery to help spot the virus. Data to support the work came from “thousands of diverse images from very hard hit areas of the infection,” including the U.S., China, Japan and Italy, Powell said. An NIH radiologist expertly labeled the information and then officials were able to co-create the pre-trained models in less than a month.
Currently the U.S. and many others do not endorse using detection models or CT-scanned imagery as the standard of care. But Powell noted that China has approved medical imaging for prognosis and the world “is still undecided” in regards to the proper guidelines. With that in mind, she said it’s important to continue developing and improving the AI-based models for COVID-19 detection.
“These models can be used as building blocks to future, you know, things, like predicting whether a patient is going to go into an ICU,” Powell said. “So these are building blocks that will accelerate future research—that’s what these models are intended to do.”