Accelerating Common R Commands Using the Gputools Package

• Activity Number: 354
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Mark Seligman*+ and Josh Buckner
• Companies: Rapid Biologics LLC and University of Michigan Molecular and Behavioral Neuroscience Institute
• Keywords: Parallel ; R ; performance ; GPU ; CUDA ; hardware

Abstract

The graphics processing unit, or GPU, is standard equipment on every new computer having a graphics display. Until recently the role of the GPU had been limited to control of the visual display. In the last few years, however, the parallel-processing capabilities of these devices have been made accessible to the user. This has resulted in an explosion in the number and variety of high-performance numerical applications exploiting their advanced hardware features. The Gputools toolkit is the first R package to offer statistical utilities specifically tuned for high performance on the GPU. Recent additions to the package include high-speed implementations of R's commands for generalized linear modeling, matrix cross product and QR decomposition. We discuss features of the package and comment on the performance of recent upgrades. We also examine future directions for GPU support in R.