CHICAGO – Nvidia this week officially started up Cambridge-1, a massive supercomputer fully dedicated to healthcare, that the company believes will unleash innovation in what it called "digital biology" and accelerate the transfer of technology from the research laboratory to patient care.
Cambridge-1 is the fastest computer in the UK and the 41st fastest in the world, according to a recently released update to a compendium of the "top 500" most powerful computer systems.
Nvidia first announced Cambridge-1 last October at its annual GPU Technology Conference. The supercomputer is part of a center of excellence for artificial intelligence that the Silicon Valley-based firm is building in Cambridge, UK.
"I think we're all starting to be aware about how transformative AI is in terms of technology and how that's leading to advances in medical breakthroughs," David Hogan, the company's VP of enterprise for Europe, the Middle East, and Africa, said this week during an online press briefing. "But to date, nobody has had the scale of AI computing to really address some of the largest challenges around," including those brought on by the COVID-19 pandemic.
"Cambridge-1 is designed to address those challenges by providing a platform for wide-scale AI compute [power] to address some of the largest models that are out there," Hogan said.
"Cambridge-1 really is an AI blueprint of how to do this at scale, how to tackle some of the largest challenges in healthcare, but also how to run AI models at significant scale," Hogan explained. "Some of these challenges just cannot be done on other platforms."
Craig Rhodes, EMEA industry lead for healthcare and life sciences at Nvidia, said that genomics, molecular dynamics, and the COVID-19 pandemic have shaken up drug discovery and transformed high-performance computing. "Being able to go down to microscopic level, being able to look at how proteins react and change has been fundamental to drive science forward."
Rhodes said that a long-held maxim in computing — that computing power doubles and the cost falls by half every two years — no longer applies at the scale of data processing that bioinformatics now demands. "Moore's law just doesn't work in this world," he said.
"This change in the way that we're having to use data, not just the size and quantity of data, but using data at speed to understand what we're going to do, how compounds and targets react, is essential for the world that we live in today," Rhodes said.
Rhodes quoted Kim Branson, senior VP and head of AI and machine learning at GlaxoSmithKline, who said last year that the firm collected more data in 2020 than it had in its entire history going back 300 years.
"This exponential growth of data within these organizations, they have to change the way they compute this," Rhodes said. "This advancement of going quicker and quicker is just stretching and putting a strain on normal computation."
"The bottleneck is becoming the computational power to be able to make any sense of that data and really use it for value," Rhodes said. "This is where we're positioned in Cambridge-1."
The Cambridge-1 supercomputer is built on Nvidia DGX SuperPod supercomputing infrastructure, based on the company's A100 graphics processing unit and related DGX A100, a box that holds eight A100 processors and companion server hardware to increase computational speed.
Cambridge-1 features 80 DGX A100 units and has 400 petaflops of computing power. It is powered by 100 percent renewable energy, according to Nvidia.
The supercomputer also incorporates Clara Parabricks Pipelines, which Nvidia has offered since it purchased sequencing analysis software developer Parabricks in 2019. Parabricks, a University of Michigan spinout, had developed technology that leans on GPUs to accelerate the analysis of whole genomes to less than one hour.
Nvidia said this week that Cambridge-1 is a $100 million investment. When it announced Cambridge-1 last year, the Santa Clara, California-based company pegged initial costs at about $50 million.
Since Nvidia unveiled plans for Cambridge-1 nine months ago, the computing architecture has not changed much, but Nvidia's approach to data handling and global collaboration have evolved, according to Hogan. For example, feedback from the founding partners has prompted Nvidia to invest more money in data security, he said.
Those partners include GSK, Oxford Nanopore Technologies, Guy's and St Thomas' NHS Foundation Trust in London, King's College London, and AstraZeneca.
Oxford Nanopore uses Nvidia GPUs in its sequencing instruments. Access to Cambridge-1 will allow the fast-emerging company to accelerate and improve development of analytics algorithms.
"Harnessing the power of Cambridge-1 will help us further speed up our algorithm development to support powerful, accurate genomic analysis," Rosemary Sinclair Dokos, VP of product and program management at Oxford Nanopore, said in a statement provided by Nvidia.
Nvidia's Hogan said that Cambridge-1 actually is a two-part project, featuring both a powerful supercomputer and a "deep collaboration with industry, academia, research, [the] startup community, working very closely to really transform how we think about healthcare."
Nvidia has been working in healthcare and life sciences for more than a decade, originally starting in medical imaging. But with the growth of GPU computing and artificial intelligence, the firm has expanded into data-intensive fields including genomics, drug discovery, pathology, and the application of natural language processing to clinical research data.
Hogan said that Nvidia chose to build Cambridge-1 in the UK because that nation was among the first to make a national investment in bringing AI to the point of care. Plus, the company already had a history of collaboration with organizations there, including with the National Health Service, UK Biobank, and Genomics England.
"But fundamentally, the benefit of this will be seen not just in the UK but throughout the world," Hogan added.
For example, AstraZeneca is working with Nvidia to create transformer-based neural networks to analyze chemical structures, and Cambridge-1 will accelerate this work. This drug discovery model, called MegaMolBART, helps with reaction prediction, molecular optimization, and generation of de novo molecules, the companies said.
Rhodes said this week that AstraZeneca will make this work open-source, available to researchers worldwide in a field he called "digital biology."
That was a change in the original concept that Nvidia had envisioned for the supercomputer.
"We thought this was going to be a very closed, very private set of results, but we would get a bit of maybe a paper in Nature," Rhodes said. "What we've got is far more than we could ever have imagined. This will be going out to the communities."
Cambridge-1 partners cover the clinical, research, drug discovery, bioinformatics, and genomics realms. "Ultimately, we want to affect the patient pathway. We want to look at how can we improve patient outcomes based on the work that we're doing," Rhodes said.
"We hope this will be a blueprint for others around the world to learn from, to build upon, both at a clinical level, but also at a computing level," Hogan said.
"Even these organizations with enormous amounts of data — and I mean enormous amounts of data — they still struggle with really understanding how to take user cases and bring it to enormous scale," Rhodes said.
"There are a lot of big research supercomputers around the world," he added. "Rather than sharing time, how can we share some of the understanding of the activities that are actually going on?"
Hogan said that Cambridge-1 will be open to collaborating with other supercomputing centers, including Berzelius, a 300-petaflop system that Nvidia has built as part of the Wallenberg Artificial Intelligence, Autonomous Systems, and Software Program in Sweden.
"Yes, it's great to get a paper in Nature, but for us, the value is going to show how we can change patient pathways to improve patient care by what the work that will be done with Cambridge-1," Rhodes said.