Accelerating High-Throughput Computing through OpenCL
High-Performance Computing Research Group, University of Huddersfield, Huddersfield, United Kingdom
EMerging Technology Conference (EMiT), 2016
As the computational trend diverges from standard CPU computing, to encompass GPUs and other accelerators, the need to integrate these unused resources within existing systems becomes apparent. This paper presents the implementation of a HTCondor pool with GPU execution capabilities through OpenCL. Implementation is discussed from both the system setup and the software design standpoint. The GPU landscape is investigated and the efficiency of the system is evaluated via FastFourier Transform computations. Experimental results show that HTCondor GPU performance matches a dedicated GPU cluster.
June 28, 2016 by hgpu
