NVIDIA's new contest allows scientists to compete for a new supercomputing core suitable for a petascale supercomputer. This image shows a petascale Jaguar supercomputer at Oak Ridge National Laboratory..
Credit: Oak Ridge National Laboratory
Huge supercomputers can do everything from simulating the universe's Big Bang to modeling how the human brain works — if universities or labs can afford the expensive room-size computing clusters costing tens of millions of dollars. A new contest offers researchers the chance to win a supercomputer core by explaining what they would do with the computing power equivalent to 20,000 laptops.
The new supercomputer core is a Tesla Kepler graphics processing unit (GPU) — NVIDIA technology originally designed for running graphics-rich video games. NVIDIA hopes the $3,000 units can deliver petaflop supercomputing (1 thousand trillion calculations per second) at a relatively cheap price to boost science and innovation around the world.
"By the end of 2012, NVIDIA will launch a Tesla GPU based on our new Kepler architecture that will enable any university in the world to build a petaflop supercomputer that will fit into nearly any university’s data center and budget," said Sumit Gupta, senior manager in the Tesla GPU Computing HPC business unit.
Three lucky researchers will get "exclusive early access" to using the new Tesla GPU based on NVIDIA's Kepler architecture. Anyone wanting to enter the contest must submit a 100-200 word description of how they would use a GPU-equipped petaflop supercomputer by May 4 (full NVIDIA contest rules here).
Current generation NVIDIA GPUs already power three of the world's fastest supercomputers in the seemingly never-ending race for more computing power. A $100 million Titan supercomputer could become the world's fastest petascale supercomputer once it comes online at the U.S. Oak Ridge National Laboratory in Tennessee this year.
The cheaper price of petascale supercomputer comes at a time when NVIDIA is already looking forward to the next generation of supercomputing — exascale supercomputers capable of running 1,000 times faster than today's petascale supercomputers. But researchers must first figure out how to make exascale supercomputers that don't require the power equivalent to the Hoover Dam.