The Rock Stars of HPC series is about the men and women who are changing the way the HPC community develops, deploys, and operates the supercomputers and social and economic impact of their discoveries.
As the lead developer of the VMD molecular visualization and analysis tool, John Stone’s code is used by more than 100,000 researchers around the world. He’s also a CUDA Fellow, helping to bring High Performance Computing to the masses with accelerated computing. In this way and many others, John Stone is certainly one of the Rock Stars of HPC.
insideHPC: What first sparked your passion for HPC?
John Stone: That’s quite a long story. I grew up in a very small town in the Midwest with about 7,000 people give or take so computers weren’t everywhere when I was a child, though I did have the good fortune of growing up when microcomputers first came onto the market. But I would say I first became fascinated by computers at a very young age through my exposure to the earliest arcade video games like Pong and Space Invaders. When I saw them I was just amazed – I thought to myself how could a machine do this? I wanted very much to understand what made them work and how it was possible for those graphical images to be drawn on the screen in real-time. I’d never seen anything like that before.
When I was 11 years old I attended a computer camp and I learned how to program in basic on Z-80-based microcomputers, and that really led me to a long-standing interest in computers and programming, which continued through my early years, through high school and into college when I became a computer scientist. One particular thing that got me interested in high-performance computing was a column in Scientific American called Computer Recreations. There was an analogous column called Mathematical Recreations that I used to read at the public library, and I remember one edition called Simulated Evolution: Wherein bugs learn to hunt bacteria. It was published in 1989 when I was a high school senior and it presented the fascinating idea that you could simulate bacteria on a computer.
I had done enough computer programming at that point to know how to do various things with computer graphics – simple computer graphics given the time period – but I wanted to try and write such a program, so I read that article and I devised my own little simulated evolution program that I let run and run, and run. That’s the first time I can remember really wishing that I had a much faster computer, and it caused me to then go bump around town looking for faster computers to run my program on so I could see if I could get something more interesting to occur without having to wait so long. Ever since, I’ve had this never-ending struggle wishing I had a faster computer. That in itself has given me a great deal of personal experience in understanding how increases in performance can enable a scientist to do something that they wouldn’t be able to do otherwise.
insideHPC: You’ve had an illustrious career since your days playing Pong. What would you say has been the main highlight?
Stone: My primary achievement has been the development of VMD over the last 19 years. Developed at the NIH Center for Macromolecular Modeling and Bioinformatics here at the University of Illinois, VMD is a molecular visualization and analysis tool, and it’s undoubtedly the single longest lived project I’ve ever worked on. Interestingly, it was begun before I arrived at the University by a group of my colleagues under Professor Klaus Schulten’s leadership. When I began leading the development in 1998 we had around 500 users and now we have over 100,000 users, and the program has been cited in more than 20,000 journal articles and research papers.
VMD is really coming into its own as a tool for HPC, and just in the last year we’ve received 2,800 citations by users of the program, which works out that once every three hours somebody somewhere in the world is writing a research paper and has taken the trouble to acknowledge the use of our software. That’s pretty amazing, and I would attribute it to a number of things; one of which is that the original developers of the software were scientists themselves. As biophysicists, they had a particular mind-set about what they wanted out of it, and that strongly flavored the original architecture of the program. They were also very gifted developers in their own right, so when I began taking over the development and ongoing leadership of VMD and evolving it into what it is today I began from a very solid, strong standing in terms of what the program could do for the community it was intended to serve.
insideHPC: That community will soon be gathering in Silicon valley for the GPU Technology Conference 2017 – what are you most looking forward to?
Stone: I think a main focus at GTC will be the recent advances in deep learning neural networks and Artificial Intelligence technologies, but I also think there will be a bunch of announcements from Nvidia, IBM and some of their other partners about upcoming hardware and software roadmaps. I don’t have any specific details, but if I were to predict I would say that several of the advanced GPU hardware and programming systems efforts that they’ve been quietly developing at Nvidia and elsewhere for several years may be at the point at which some of them can be announced to the public. And if that happens, some of those announcements will be very exciting.
Some of the hardware and software features that we originally asked for back in 2008 very soon after the introduction of CUDA and GPU programming are now on the horizon. It takes hard work – years, sometimes – to work through the sequence of challenges that have to be overcome to get something that can be implementable at a reasonable cost in commodity hardware. And what’s exciting is that a lot of the difficult problems that I’m aware of are being addressed. That’s going to be very exciting to see, and I’m hoping some of those things become public information at the GPU conference.
John Stone is Senior Research Programmer at the University of Illinois at Urbana-Champaign, the lead developer of VMD, and he has joined the Khronos Group Advisory Panel for the Vulkan Graphics API.
At GTC 2017, Stone will present a talk entitled ‘GPUs Unleashed: Analysis of Petascale Molecular Simulations with VMD’. He will showcase recent successes in the use of GPUs to accelerate challenging molecular simulation analysis tasks on the latest Nvidia Tesla P100 GPUs on both Intel and IBM/OpenPOWER hardware platforms, and large-scale runs on petascale computers such as Titan and Blue Waters.
See John Stone speak at the GPU Technology Conference in Santa Clara, CA on May 8-11 – Learn More)