Title: Performance issues in GPGPU programming and parallelizing LAMMPS with CUDA

Speaker: Dr. Hyesoon Kim, Georgia Institute of Technology

Date/Time: Tuesday, March 31, 2009 – 11:00 AM

Location: IPB/1154 (NM)

Brief Abstract: As the computing power of the GPU increases dramatically, more and more programmers want to parallelize their applications for the GPU.  However, optimizing programs for the GPU is still a big challenge to programmers.  To help programmers, we have developed several techniques.  In this talk, first, I’ll describe our GPU architecture performance analytical model.  Second, I’ll present our dynamic compilation system that dynamically distributes computation work between the GPU and CPU.  And finally, I’ll show our on-going effort on parallelizing LAMMPS using CUDA.

CSRI POC: Paul Crozier, (505) 845-9714