Star Power: World’s Largest GPU In-situ Visualization System Models Formation of Milky Way

by Bhushan Desam

Simulating the formation of an entire galaxy is no easy task. It takes precise calculations and massive amounts of computational horsepower.

It’s even harder to run a simulation of this size while simultaneously rendering the results interactively  on the same system.

Yet researchers at the Swiss National Supercomputing Center (CSCS) have done just that. For the first time, they used the ultra-powerful Tesla GPU-accelerated Piz Daint supercomputer to run a massive in-situ visualization – simultaneously  computing and generating a visualization of the formation of the Milky Way galaxy (for more on in-situ visualization see “Interactive Supercomputing with In-Situ Visualization on Tesla GPUs“).

CSCSembed
Researchers used the Tesla GPU-accelerated Piz Daint supercomputer to run a massive in-situ visualization – simultaneously computing and generating a visualization of the formation of the Milky Way galaxy.

Piz Daint is the world’s largest supercomputer capable of doing this. It allowed researchers to tap into a whopping 2,048 GPU nodes for simulating 256 million stars, planets and other celestial bodies.

This has never been done on GPUs at this scale … until now.

This feat highlights a growing trend in computational research – tapping the power of today’s computers to give researchers new tools to advance their work.

And it wouldn’t have been possible without Tesla GPU accelerators.

What In-situ Visualization Is, and Why It Matters

Numbers and formulas are great. But the ability to visualize their computational work often gives researchers new insight and perspectives.

Rendering visualizations of scientific simulations has been done for years. They’ve been separate jobs, though, requiring two different systems: one for computation (running the science application), another for rendering the results visually.

Days or weeks were required for simulation work to be completed before it could be visualized. If, during the process, an error was discovered or a parameter changed, it would require re-running the entire computation on the simulation system. Then it would have to be rendered again. This wash-rinse-repeat cycle could be long and cumbersome.

In-situ visualization changes all this. It allows researchers to simultaneously run computations and visualize the results on a single system. It lets them see things in a way that formulas on a page can’t convey, bringing new perspectives and faster results. They can even change simulations on the fly to optimize their work.

Our Tesla GPUs provide the processing muscle for all this work to be done efficiently and in real time, without taxing the performance of the system.

The result is fewer steps, faster time to discovery – a single system to get the job done.

To see the GPU-accelerated in-situ Milky Way demo first hand, swing by our booth 1727 this week at SC14. It shows a simulation and visualization of the galaxy formation on Piz Daint, in Switzerland, and on Titan, another GPU-powered supercomputer, at Oak Ridge National Labs.

If you’d like to learn about the latest developments to leading HPC visualization software, Kitware’s Robert Maynard will be presenting on Wednesday, Nov. 19, at 3pm CT.

Information about these and other presentations are available on our SC14 website:  http://www.nvidia.com/object/sc14-technology-theater.html