NVIDIA Announces New GPU Architecture, AI Compute System, Genomics Analysis Record, Smart Hospital Platform, More
By Allison Proffitt
May 14, 2020 | NVIDIA moved its Graphics Technology Conference keynote online today. CEO Jensen Huang hosted the company’s first “kitchen keynote” with hardware laid out on the kitchen counter among spatulas and pepper mills.
Huang made a host of announcements--the six part keynote is available on YouTube--including unveiling the NVIDIA DGX A100 AI system, delivering 5 petaflops of AI performance. Originally scheduled for March 23, the GTC 2020 keynote was postponed due to the coronavirus pandemic.
For life sciences applications, Kimberly Powell, NVIDIA’s vice president of healthcare, shared details about the first NVIDIA DGX A100 installation, a speed record for genomics analysis from NVIDIA Clara Parabricks computational genomics software, the new NVIDIA Clara Guardian healthcare platform, and AI models for COVID-19.
The company also launched the NVIDIA DGXpert program, connecting customers with AI experts, and the NVIDIA DGX-Ready Software program, which helps customers take advantage of certified, enterprise-grade software for AI workflows.
COVID-19 colored all of the announcements, and Powell emphasized the new technology’s relevance to the challenge at hand. Our data generation and artificial intelligence capabilities today have positioned us to better deal with a global health crisis than ever before. “No other time in history has the combination of healthcare and technology been more important in the fight against COVID and the fight against future infectious diseases,” Powell said. “Because this is a phenomenon we’re going to have to live with for the rest of our lives.”
Twenty-Minute Genomics Analysis
Thanks to the Ampere architecture and the DGX architecture, NVIDIA Clara Parabricks has set a speed record, Powell said, completed DNA sequencing analysis of the whole human genome in just under 20 minutes. Parabricks, an Ann Arbor, Michigan startup, developed a GPU-based solution that speeds up the process of analyzing whole genomes, and was acquired by NVIDIA in December 2019.
Parabricks is built using NVIDIA CUDA-X and benefits from CUDA, cuDNN and TensorRT inference software and runs on NVIDIA entire computing platform from NVIDIA T4 to DGX to cloud GPU instances. At the time of the acquisition, Parabricks was working on solutions for single cell and RNA pipelines. Today, NVIDIA introduced GPU-accelerated RNA-sequencing pipelines as part of Clara Parabricks.
To make it easier for users to benefit from NVIDIA Clara Parabricks acceleration on the cloud, NVIDIA is partnering with the DNAnexus enterprise genomics cloud platform to seamlessly provision, run, and return results.
Clara Parabricks will be used for accelerating the sequencing analysis in the United Arab Emirates population genome program, Powell said. “They’re going to be using our NVIDIA Clara Parabricks so they can build up the reference genome specific to UAE citizens, then quickly go from 10,000 genomes to their entire population.”
COVID-19 News: NVIDIA is making Clara Parabricks computational genomics software available via a free, 90-day license to COVID-19 researchers.
Clara Guardian For Smart Hospitals
Launched today, NVIDIA Clara Guardian is a smart hospital platform linking AI edge computing with sensors of all types: cameras, microphones, and speakers. “Edge computing is enabling the real time, secure deployment of these applications and AI is what’s enabling the processing and sensor fusion that’s going to really help us automate the insights and actions within the walls of the hospital,” Powell said. “Deploying AI for voice and video—these are pretty complicated deployment frameworks. We take a lot of that difficulty out, and really build a rip-and-replace in terms of the AI applications to deploy at the edge in these very small form factor compute devices.”
Clara Guardian is built on top of NVIDIA’s Edge AI Stack which includes the Jetson product line, T4 edge inferencing servers, and the new EGX A100 series.
More than 18 global partners are creating applications for Clara Guardian, Powell said, and the platform is already deployed at more than 50 hospitals worldwide.
COVID-19 News: Hospitals—"and really any building around the world”—need to be able to track exposure to infection, Powell said. “Body temperature screening is a huge area exploding with new solutions coming to market,” she said. From a public safety perspective, multiple and varied sensors, like the ones Clara Guardian links, will be crucial in coming days, Powell argued.
Fastest AI Supercomputer
Argonne National Laboratory is currently installing the first DGX A100 supercomputer, with a deep learning emphasis. “This supercomputer is going to combine both accelerated computing and artificial intelligence,” Powell said. “Deep learning is a huge part of this workload.” Critical applications: molecular dynamics simulations, virtual drug screening, protein docking, and more.
The system is made of 24 of the new DGX A100 supercomputers—a total of 192 NVIDIA A100 GPUs—and uses high-speed Mellanox HDR 200Gbps interconnects, thanks to the $7 billion NVIDIA acquisition that closed on April 27. Powell highlighted the Mellanox “incredible datacenter fabric that makes this unprecedented amount of compute all connected together.” The new system will achieve 120 petaFLOPS of AI computing power, she said.
“This is a datacenter-level supercomputer in a node,” Powell said. And the beauty of the DGX architecture along with the Mellanox technology is the ease with which they can be set up. Several nodes are already up and running at Argonne, Powell said. Within NVIDIA, 50 nodes are operational, and it took three weeks to set up.
A single rack of five DGX A100 systems replaces a data center of AI training and inference infrastructure, with 1/20th the power, 1/25th the space and 1/10th the cost, the company said.
Other institutions that will be installing DGX A100 systems include the University of Florida; the Center for Biomedical AI at the University Medical Center Hamburg-Eppendorf, Germany; Chulalongkorn University in Thailand; Element AI in Montreal; German Research Center for Artificial Intelligence (DFKI); Harrison.ai in Sydney; the UAE Artificial Intelligence Office; and VinAI Research based in Hanoi and Ho Chi Minh City.
COVID-19 News: The first application of the new system at Argonne National Laboratory will be for drug and vaccine discovery. “We’re delighted to work with Argonne National Labs,” Powell said. “We’re working very closely with them as part of the COVID-19 HPC Consortium that has come out of the White House Office of Science and Technology Program.”
New AI Models for COVID-19
Finally, through the Clara Imaging framework, NVIDIA is making two pre-trained AI models of COVID-19 lung images widely available. One model is for lung segmentation; the other is for 2D COVID classification in CT images. The models are immediately available in the latest release of Clara Imaging on the NGC software hub.
The models were co-developed with the National Institutes of Health Clinical Center based on images of COVID-19 patients worldwide. NIH radiologists labeled lung images and the NVIDIA Clara Imaging Framework developed two models in less than three weeks.
While CT scans are not currently used for diagnosing COVID-19 in the United States, they are used for diagnosis in China and other countries. Powell predicts that lung imaging models will nevertheless enable significant advances in COVID-19 research. “You can take these pre-trained models, and you can adapt them locally to your environment… and build on top of them for other predictive models,” she said, suggesting predictions of prognosis or which patients will need the ICU.