Over at the Nvidia Blog, Roy Kim writes that SDSC researchers are using NVIDIA Tesla K20X GPU accelerators to help improve earthquake forecasts, enabling engineers to design safer buildings and save lives. The power of GPUs enabled SDSC to run high-frequency, compute-intensive wave propagation simulations to better understand how a broad range of structures will respond in a major quake.
To meet the needs of the CyberShake 3.0 project, Cui realized they would need 750 million CPU hours on a traditional CPU-based supercomputer, costing over $800,000 just in power cost to support his simulations. That’s when they turned to GPUs for help. AWP-ODC, the research team’s primary seismic application, is more than 5x faster when run with GPUs, allowing researchers to discover insights they would not have been able to before. At the same time, they would save over $600,000 in power costs for their simulations. Less than a month ago, Cui’s team achieved over one petaflop of performance running on over 8,000 GPUs on the Titan supercomputer, shattering their previous record of 220 teraflops of sustained performance on Oak Ridge’s Jaguar supercomputer.
This video from Amit Chourasia at SDSC depicts a magnitude-8 earthquake in the northern San Andreas Fault.
Read the Full Story.