uBench: Performance Impact of CUDA Block Geometry
Computer Science Department, University of Valladolid, Spain
University of Valladolid, Tech. report IT-DI-2012-0001, 2012
Nowadays, there is a lack of performance models for the execution of programs implemented using the CUDA model for GPU (Graphics Processing Units) devices. We have designed and implemented a suite of micro-benchmarks, called uBench. The purpose of uBench is to identify the effects on performance derived from the combination of: (1) the hardware details of each GPU device, and (2) the configuration of the thread-block, an important runtime configuration parameter in the CUDA programming model.
January 3, 2013 by hgpu
