SC Conference - Activity Details

Peter Messmer  (Tech-X Corporation)

GPU Software Tools

Wednesday,  11:00AM - 11:30AM

Room 19A/19B

Bigger detector sizes, larger sensor counts and more challenging data reduction algorithms require an increasing amount of processing power for data analysis, image processing and simulation post-processing. The tremendous processing power offered by modern Graphics Processing Units (GPUs) is a cost effective alternative to clusters for accelerating these tasks. However, data analysis tools like IDL, MATLAB or Python do not support GPUs natively and users have to write low-level C code to take advantage of these devices. GPULib, developed by Tech-X Corporation, enables data analysts to take advantage of GPUs from within their favorite high-level language, offering a large collection of algorithms, ranging from basic vector operations, to kernels like array interpolation and reshaping. Real-world applications can achieve orders of magnitude speedup without requiring any special GPU code.