At the kickoff of the SC14 supercomputing conference, the crowd – some 250 packed deep in the NVIDIA booth theater – kept listening, even as a New Orleans marching band festooned with pink and blue feathers stormed by.
They heard senior execs from Oak Ridge and Lawrence Livermore national labs discuss why they’re using GPU accelerators to help power the U.S.’s next-generation of flagship supercomputers – systems in the CORAL program that will run 3-6X faster than today’s most powerful systems.
They heard Oak Ridge’s associate lab director Jeff Nichols describe how NVIDIA’s Tesla Accelerated Computing Platform will provide “10-times more performance, letting us do 10-times more science” than the lab’s current system – the fastest in the U.S. – while sticking to the same power envelope.
And they heard about NVIDIA’s just-launched Tesla K80 dual-GPU accelerators. They’ll provide unprecedented power to a wide range of applications for machine learning, data analytics and hard science.
But more than the sounds, it was the sights of the booth that kept them lingering well after the talk concluded.
Decked out in green and black, the booth – smack in the middle of the New Orleans Morial Convention Center hall – shows off in dazzling 4K detail how NVIDIA GPUs can both simulate complex scientific experiments while simultaneously visually rendering the result. Researchers, by being able to to see their computational work in real time, can work far faster and even change simulations on the fly – in contrast to following the older method of simulating their work and then, once it’s completed, taking more cycles to render it.
One interactive visual demo, streamed halfway around the world from Switzerland’s Piz Daint system, shows an in situ visualization of the formation of millions of stars in the Milky Way. Another, streamed halfway across the country from Silicon Valley, depicts the simulation of the full 100 million atoms of the chromatophore, a subunit of bacteria responsible for photosynthesis, to help researchers understand how biological processes create energy.
Right around the corner, a machine-learning demo shows how researchers at Switzerland’s IDSIA’s AI Lab were able to train a computer to annotate 3D representations of neurons and dendrites from a brain scan dramatically faster and more accurately than human beings.
But the NVIDIA booth isn’t all about the far edges of science.
For those just getting interested in how to accelerate their scientific work, there’s a small classroom where they can learn to program GPUs. NVIDIA’s Will Ramey, who’s helping to staff the sessions, said some 300 individuals will likely spend an hour or two there getting up to speed. While intended primarily for a sophisticated audience, the sessions last year included a talented 12-year-old middle-school girl whose dad was a researcher at a nearby university.
Perhaps the booth’s most popular feature: the theater. Every half hour, one of three dozen GPU experts will talk about their work – ranging from using machine learning to help detect breast cancer to work GE is doing to make turbines run more efficiently. (For a complete schedule of talks – and, starting Tuesday, video replays – click here).
Another popular contender: Monday night’s NVIDIA-logoed ice sculpture. It housed an internal luge-course, into which bartenders poured the makings of a swamp-green bayou alligator cocktail, which sloshed its way into a waiting glass.