Nvidia’s Game-Changer in AI-Driven Drug Discovery
Nvidia’s new Hopper 100 GPU represents a massive, industry-changing leap forward for drug discovery powered by machine learning.
Written by: Noah Parekh
Last month, I had the chance to offer our readers an overview of how artificial intelligence and machine learning are changing drug discovery. Bringing a new drug to market using traditional methods often takes up to ten years and more than $2.5 billion. Recent innovations in data science technology promise cost savings of 70% and a tenfold increase in development speed.
In Q3 of this year, the field is going to take another huge step forward. That’s when NVIDIA’s next generation of processors will start shipping to data centers worldwide. The Hopper 100 (H100) GPU leverages a new data architecture alongside a dozen other minor revolutions in data processing and storage and was designed specifically for the kind of large-scale, mind-bogglingly complex search and optimization problems that allow for the identification of new candidate molecules. It is a breakthrough technological advance. For instance, used in concert with the companion software and hardware NVIDIA is offering, the H100 will accelerate the dynamic programming used for genomics, protein structure prediction, and other problems by up to 4,000% compared to top CPUs. (The increase is roughly 700% over leading GPUs.) The chip contains 80 billion transistors, almost a 70% increase from the company’s previous industry-leading product. The company’s CEO described it as “the biggest generational leap ever” in GPU performance, and emphasized its scalability and the massive increase it represents in the commercial accessibility of ultra-high-powered machine learning.
Just as importantly, the H100 is arriving on the biotech scene at a fortuitous moment. The last several years have seen an unprecedented increase in the availability of cheap, efficient data storage at the scales needed to power machine learning. The big names in data warehousing, like Snowflake, Amazon Redshift, and Google BigQuery, have built the infrastructure required to make these technologies not cost-prohibitive for startups. Last year was an inflection point for data storage and processing—VentureBeat called it “the big unlock.” That made the H100’s announcement in March all the more exciting: even new players in the life sciences will be able to afford both big data and AI tools to manage it.
While we can’t predict the exact changes that we’ll see once the H100 hits shelves, we can highlight some of the most exciting possibilities. The first is screening compounds. A supercomputer running the chip’s predecessor—which allowed for only 1/9th the processing speed—was able to simulate almost 1.5 billion protein-ligand binding possibilities in less than 12 hours, identifying thousands of potential target molecules for wet-lab testing.
Another area of rapid progress will be generative chemistry, which I mentioned last month. Advanced AI tools build models of a given chemical space, and then, working within those parameters, produce lists of molecules with tailored pharmacokinetic properties. With the new technology, it is likely that we will soon be able to generate tens of thousands of candidate molecules per second.
One more example. The Schrödinger platform for physics-based simulations of molecular properties has been widely used for years, and the company had a successful IPO in 2020. Several of the biggest pharma companies in the world have standardized its use in preclinical research. With new tech updates and powered by the H100, the technology may be able to synthesize molecules and identify their binding affinities for “a few dollars of computing costs” on NVIDIA GPUs, according to the company’s CTO. The equivalent cost in a wet lab ranges from several thousand dollars to several hundred thousand.
These are just a few of the fronts on which the H100 and its companion technologies are set to change the face of the industry. Most experts hope that advances like these will help to reverse the long-term trend of drug discovery becoming slower and more expensive. I’ll be keeping readers updated as the field adjusts to that new reality.
To discuss this further, we invite you to reach out to Noah Parekh.
Ready to realize your future?