Accelerating DFT calculations with GPU: A hybrid computing approach

COMP 199

Jing Kong, jkong@q-chem.com1, Zhengting Gan, zgan@q-chem.com1, Yihan Shao, yihan@q-chem.com1, Roberto Olivares-Amaya, olivares@fas.harvard.edu2, and Alán Aspuru-Guzik, aspuru@chemistry.harvard.edu2. (1) Q-Chem, Inc, 5001 Baum Blvd., Suite 690, Pittsburgh, PA 15213, (2) Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford St, Cambridge, MA 02138
Hybrid computing systems which incorporate hardware acceleration components (GPU,Cell, FPGAs) has become an emerging trend in HPC solutions. In this presentation we described our Fock XC matrix implementation accelerated by GPU using CUDA toolkit. A BLAS3 kernel based algorithm has been developed to achieve high performance on both CPU and GPUs. The benchmark calculation of Taxol employing aug-cc-pVTZ basis set (4025 basis functions) was speedup by more than 30 times using CPU (Quadcore Phenom) and GPU (C1060) together compared with original code running on single Phenom core.
 

Poster Session
6:00 PM-8:00 PM, Tuesday, August 18, 2009 Walter E. Washington Convention Center -- Ballroom A, Poster

Division of Computers in Chemistry

The 238th ACS National Meeting, Washington, DC, August 16-20, 2009