Accelerating a cloud resolving method with GPUs [poster]
Bollig, E. F., Kelly, R., Mayer, B., Garcia, J. H., & Erlebacher, G. (2009). Accelerating a cloud resolving method with GPUs [poster]. In AGU Fall Meeting 2009. American Geophysical Union: San Francisco, CA, US.
Clouds play an essential role in in Earth's climate, but to this day remain a major source of uncertainty in global simulations due to resolution restrictions incurred by computational cost. Because of the promising nature of GPU architecture and numerous reported successes of utilizing graphics ... Show moreClouds play an essential role in in Earth's climate, but to this day remain a major source of uncertainty in global simulations due to resolution restrictions incurred by computational cost. Because of the promising nature of GPU architecture and numerous reported successes of utilizing graphics processing units (GPUs) for scientific computing, we attempted to port a Cloud Resolving Model from NCAR's CCSM package onto the GPU. Utilizing shared memory, coalesced memory access patterns and other techniques for GPU optimization, we present a code implemented according to restrictions commonly demanded by today's large scientific applications. Although benchmarks suggest the acceleration was unsuccessful, our effort reveals rarely discussed limitations of GPU computing for scientific problems. Show less
