CUDA Spotlight: Michael Bussmann

By Andrea Hutchison and Calisa Cole, posted August 24, 2012
Read more CUDA in Action Spotlights

GPUs and Computational Radiation Physics

This week's Spotlight is on Dr. Michael Bussmann of HZDR (Helmholtz-Zentrum Dresden-Rossendorf), a large-scale research facility based in Dresden, Germany.

Dr. Bussmann leads a "Junior Research Group" in the field of Computational Radiation Physics.

This interview is part of the CUDA Spotlight Series.

---

Q & A with Michael Bussman

NVIDIA: Michael, why is computational radiation physics important?
Michael: Computational radiation physics helps us to build new, compact accelerators and light sources. These sources of radiation could in the future be used in medical applications such as radiation therapy for tumors and for fundamental research such as understanding the dynamic behavior of matter on the atomic scale. Making these sources small and cheap enough would allow more people to get access to them.

NVIDIA: How does GPU computing play a role in your work?
Michael: With GPU computing we have seen dramatic speed up of our code, which for the first time allows us to have live simulations of realistic scenarios. We are just learning how to use this power. We can make large parameter surveys to optimize acceleration, discuss new ideas and test them on the fly - we can even use these simulations to teach students the complex dynamics of laser-matter interaction. For us, the notion of "frames per second" has become the new "CPU hours".

Laser-particle accelerated electrons – simulation on 16 NVIDIA GPUs

But we just have started. We can now add physical phenomena to our code that we never thought would be possible to compute. One example is taking 100 billion electrons, following each individual particle and calculating its contribution to the overall radiation emitted from laser-generated plasma.

In experiments, this radiation can then be used to identify the particle dynamics which otherwise can hardly be resolved. Here, our GPU simulation really opens up the way to a new understanding of laser-matter interaction.

Laser WakeField acceleration of ions

NVIDIA: What are some advantages of working with the CUDA programming model?
Michael: CUDA is both easy to learn and mature and powerful enough to get the most out of modern GPU hardware. For us, it gives us the best control of the computational power while minimizing the time for coding.

NVIDIA: Tell us about the PIConGPU (“Particle in Cell” on GPU) project.
Michael: PIConGPU started as a school project. Heiko Burau, the inventor of PIConGPU, had won a prize in the German "Jugend Forscht" junior research program. This program allows school pupils to work in research labs.

After six weeks working at HZDR he presented a first single-GPU version of PIConGPU that produced a live video of the particles and field dynamics on screen. We were absolutely amazed by this, but did not realize what the future would hold for us.

We teamed up with ZIH Dresden, the high-performance computing center at Technical University Dresden (a CUDA Center of Excellence). Again, we found some bright students and programmers that helped us with putting PIConGPU on a cluster of GPUs.

After more than three years PIConGPU has become one of the fastest particle-in-cell codes around. And it has helped us find a lot of new partners in Dresden and around the world to collaborate in such diverse fields as plasma physics, laser physics, astrophysics, medicine, visualization, database technologies and high-performance computing (some of them have joined forces with the Dresden CUDA Center of Excellence).

What I like most about PIConGPU is that as of yet none of the developers holds a university degree. We work with a team of young, enthusiastic students and programmers who often have totally crazy ideas that in the end give PIConGPU the extra push that makes it so fast. I'm really grateful for being part of one of the best teams I've ever worked with.

NVIDIA: What are you most excited about in terms of future advances?
Michael: Clearly, techniques like DirectGPU will make our life much easier when using large GPU clusters. On current GPU clusters we mainly use CPUs to transfer data between nodes and to the file system. This could be done by much more basic hardware. Putting a simple processor like an ARM together with a powerful GPU would allow us to get the most out of existing hardware while reducing our coding effort.

For future systems I am really excited by the possibilities to simulate, analyze and visualize data on the fly with GPUs. Frames-per-second simulations exist, but we do not yet have the tools to leverage the full power of these simulations.

Interactive feedback, visual analytics and complex, on-the-fly data analysis would change the way we think about HPC today. I want to use a supercomputer as an explorative tool which enables me to play with even the most complex physics. I think this future is not that far away, we just have to use the tools that are already there in a more intelligent way.

---

Bio for Michael Bussmann

After not finding the Higgs particle, Michael Bussmann followed his studies on laser cooling of relativistic ion beams at Ludwig-Maximilians University Munich and GSI Darmstadt. After receiving his PHD in experimental physics he switched subjects and became a theoretical laser plasma physicist at Helmholtz-Zentrum Dresden-Rossendorf.

Since 2008 he has been head of the Junior Research Group "Computational Radiation Physics" at the HZDR Institute for Radiation Physics. He is part of the Laser-Particle Acceleration Group and likes to work together with all the experimental physicists there. In his free time Michael likes hiking, diving and sailing and enjoys life in Dresden.

Relevant Links

Computational Radiation Physics Institute at HZDR http://www.hzdr.de/crp
PIConGPU http://picongpu.hzdr.de
Dresden CUDA Center of Excellence http://ccoe-dresden.de/

Contact Info

m.bussmann (at) hzdr (dot) de
Helmholtz-Zentrum Dresden-Rossendorf
Bautzner Landstrasse 400, D-01328 Dresden, Germany