SAN MARCOS – Martin Burtscher, a professor in the Department of Computer Science at Texas State University, has never attempted to model an inductively coupled plasma torch using predictive science before. That’s what makes the current research project he’s involved in so exciting.
In Oct. 2020, the Oden Institute for Computational Engineering and Sciences at the University of Texas at Austin was selected by the United States Department of Energy to tackle the ambitious, multi-disciplinary project. Burtscher is the only faculty member from outside of the UT system participating.
“I’m a tiny speck in this gigantic project. It’s huge, and I haven’t even met everyone yet. I’m mostly interacting with the computer science team,” Burtscher said. “My expertise is making complex programs run really fast on graphics processing units—GPUs—but used for non-graphics computations.
“I’m helping the team speed up some of the codes and algorithms they come up with and write, so they can run as efficiently as possible on these GPUs,” he said. “Many high-end computers and supercomputers now have GPUs because they’re so powerful and energy efficient.”
The challenges of modeling the plasma torch are manyfold. There are heat issues to consider, plasma factors, the behavior of different gases involved, all of which have to represented numerically.
Once the teams have numerical methods and formulae to simulate those factors accurately, with the margin of error within acceptable limits, that’s where the computer science team comes in.
“We’re trying to make their simulations as efficient as possible, both in terms of how quickly we can run them on these supercomputers but also how quickly and easily we can implement them in the first place,” Burtscher said. “This is at the forefront of what is humanly possible right now in terms of the science behind it, otherwise it wouldn’t be research.
“The supercomputers are consuming huge amounts of data and producing huge amounts of data. It used to be that computers had a CPU (central processing unit) and that was it, so you wrote a program for the CPU and you were fine. Now, computers have CPUs and GPUs and possibly other accelerators,” he said. “Now, you can’t just write one program and call it a day. You have to write one program for the CPU that’s optimized for the CPU, and you have to write a substantially different program that’s optimized for the GPU. That’s a problem, right? You have to do twice as much work or three times as much work as you used to on the older machines to get good performance out of the new machines.”
Participating in the project affords Burtscher the opportunity to work toward a tangible goal, as opposed to the projects he’s normally involved with, which can often be theoretical or far removed from the end results generated by the user.
“A lot of times computer scientists working in high-performance computing enable things for other people, but we often don’t have cool stuff on our own,” he said. “We make it possible for the geologists, the chemists, you name it, to run their simulations and get their results, but our work itself is not nearly as eye-catching because it’s mostly numbers and performance. They get the pretty pictures that everyone wants to see.
“It’s nice to be affiliated with this, where I actually have something to show. I couldn’t build this torch if you put a gun to my head, but I’m still part of the project and there’s something real behind it,” he said. “This project is just cool. Just being able to work on it is awesome.”