Parallel: General Purpose Graphics Processing Units, Part 2
SESSION: Parallel: GPGPU, Part 2
EVENT TYPE: Communities, Education
TIME: 3:30PM - 5:00PM
Speaker(s):Charlie Peck, Henry Neeman, Tom Murphy, Dan Ernst
ABSTRACT: This session builds upon the previous session on GPU acceleration, with a focus on performance considerations in CUDA programs. The prerequisite is the GPGPU Part I session, or equivalent. Topics include: organizing data into effective block arrangements (matrix multiplication); basic performance tuning; more in-depth GPU architecture features such as block-shared memory; hands-on examples of using fast memories to increase performance (tiled matrix multiplication and n-body); an overview of other performance factors to consider in optimization.