CHICAGO – A new series of Nvidia graphics processing units has already found its way into some national-scale projects to understand the novel SARS-CoV-2 and accelerate research into potential therapies for COVID-19.
In May, the Santa Clara, California-based technology firm introduced the A100 GPU and related DGX A100, a box that holds eight A100 units and related server hardware to accelerate computational speed. As part of the announcement, Nvidia highlighted the COVID-19 work of Argonne National Laboratory in the US and the Ministry of Health and Prevention in the United Arab Emirates.
Since March, Nvidia has been offering a free 90-day license for Parabricks sequencing analysis software to any researcher sequencing the SARS-CoV-2 genomes or genomes of patients affected by COVID-19.
Nvidia acquired sequencing analysis software developer Parabricks in December. Parabricks, a University of Michigan spinout, had developed technology that leans on graphics processing units to accelerate the analysis of whole genomes to less than one hour. A partnership with DDN Storage yielded 100x acceleration of precision medicine workflows, Parabricks said in 2018.
The company said that Parabricks technology was able to analyze the DNA of a whole-genome sequence in less than 20 minutes on an array of A100 GPUs. Nvidia also created GPU-accelerated RNA-sequencing pipelines that can produce results in less than two hours.
While Argonne is not using Parabricks software, the national lab in Lemont, Illinois, ordered 24 DGX A100 boxes, each of which integrates eight A100 GPUs to support deep-learning research. The first three were installed in early May and the rest are going online as they arrive in this summer.
From a computational standpoint, the A100 servers are about three to five times faster than the V100s they are supplementing, according to Associate Laboratory Director Rick Stevens, who cited Argonne's own testing.
Among other applications, the research lab is using the new servers for modeling drug-protein interaction of COVID-19.
"We're using it to do machine-learning models for rapid drug screening in the SARS-CoV-2 proteome. We're using it to do molecular dynamics. We're using it to do drug docking," Stevens said.
Starting in July, Argonne will integrate the servers with its existing supercomputer to server as a new GPU-based resource to supplement the Intel infrastructure that is already in place. The goal is to complete this integration work by late summer or early fall, according to Stevens.
"Nvidia sells these things as a standalone system, each one of them. Think of this as 24 standalone systems. They don't sell it as a cluster. They don't sell it as a supercomputer. They sell it as boxes," Stevens explains. Argonne thus has to integrate with its existing infrastructure and layer software on top of it.
"This is just one little piece of what we have. We've got dozens and dozens of supercomputing projects that are ongoing," covering everything to cosmology and climate science to battery research, he said.
The new A100s support multi-instance GPU support, allowing for more efficient time-sharing. This will come in handy later this year, as Argonne will share resources with other participants in the COVID-19 High Performance Computing Consortium once the integration is complete and outsiders are able to book A100 server time through the lab's scheduling system, Stevens said.
"In order for it to be part of the HPC Consortium, it has to be part of our production infrastructure, and that takes weeks and weeks to get that integrated and debugged and everything. That's not going to happen for a few months," Stevens said.
Before Argonne opens up the new hardware to the consortium and other outsiders, researchers there are using the A100 servers internally to seek and model potential COVID-19 therapies. For now, the lab is just treating the new servers as standalone resources that can accelerate omics analysis.
The White House created the COVID-19 HPC Consortium in March, to bring government, industry, and academic partners together to apply computing resources to combat the pandemic. That group is led by the US Department of Energy and IBM in conjunction with the White House Office of Science and Technology Policy and Michael Kratsios, who serves as CTO of the United States.
Nvidia is a member, along with tech giants including Amazon Web Services, Google Cloud, Hewlett Packard, and Microsoft, plus academic researchers from Argonne, Massachusetts Institute of Technology, Rensselaer Polytechnic Institute, the University of Texas, Carnegie Mellon University, and others.
Consortium projects fall into three main categories: improving the understanding of the protein structure of the SARS-CoV-2 coronavirus; applying AI to pinpoint cellular binding sites or identify molecular candidates for drug discovery; and forecasting the spread of the pandemic.
Argonne earlier said that it is seeking apply artificial intelligence AI to improve understanding of fundamental biological mechanisms" of the novel coronavirus and the related respiratory disease in search of potential therapies.
In contrast to the work at Argonne National Laboratory, Parabricks is powering sequencing analysis for a new effort by the UAE's fledgling Population Genome Program to build a reference genome based on 10,000 sequences. The Population Genome Program launched in January, according to a spokesperson from G42 Healthcare, a unit of Abu Dhabi-based artificial intelligence and cloud computing firm Group 42.
G42 Healthcare is a technology contractor to the UAE Ministry of Health. Through the vendor, the Ministry of Health now is sampling every COVID-19 patient in the emirate of Abu Dhabi. G42 Healthcare also sequenced the strain of SARS-CoV-2 prevalent in the UAE.
The UAE was the first country in the Middle East to confirm a case of COVID-19. As of Tuesday, the emirates had reported more than 48,000 cases and 315 deaths, according to the Khaleej Times newspaper.
The Ministry of Health is running whole-genome sequencing for both the Population Genome Program and the COVID-19 sequencing project, using Oxford Nanopore PromethIon and BGI DNBSEQ instruments.
Group 42 has built a supercomputer called Artemis on an infrastructure of 81 Nvidia DGX-2 systems, the predecessor to the DGX A100. Nvidia said that Artemis has a maximum performance of 7.2 petaflops, making it 120 times faster than the system it replaced at G42. An unspecified order of DGX A100 hardware will augment that.
In May, G42 Healthcare introduced the G42 Health AI Platform, a set of analytics tools meant in part to help researchers improve detection of COVID-19 and accelerate the search for effective treatments for the novel disease.
The spokesperson said that G42 had found local isolates of SARS-CoV-2 in samples collected in May, but did not elaborate.
Emirati health authorities adopted RT-PCR testing for COVID-19 in March, but G42 Healthcare and the Ministry of Health has recently developed a new diagnostic test that combines reverse transcription-loop-mediated isothermal amplification (RT-LAMP) with DNA sequencing. "This diagnostics platform will enable more rapid and efficient population-scale testing," the spokesperson said via email.
The effort is intended to "understand genetic and epigentic landscapes that contribute to the varying degree of morbidity," according to the spokesperson.