In this poster, we present our latest work on acceleration standard DFT (LDA and GGA) as well as hybrid DFT in the Vienna Ab-initio Simulation Package (VASP) using NVIDIA graphics processing units (GPUs) with CUDA. Starting from our initial works[1,2] some new routines have been ported to GPU, and all the GPU parts have been tuned to exploit the K20 and K40 graphics cards of Nvidia at their maximum. Based on a careful profiling the VASP performances on different benchmarks (more than 20 cases), we will present the main modifications of the code. Then we will compare the GPU and original versions of VASP on chemical systems of up to 1100 atoms. These tests demonstrate that on average, the GPU version running on 3 IvyBridge CPU cores + 2 K40 GPU is as fast as the CPU version on 40 CPU cores, while producing the same results. [1] Hutchinson, M., Widom, M. VASP on a GPU: Application to exact-exchange calculations of the stability of elemental boron. Computer Physics Communications 2012, 183, 1422–1426. [2] Hacene, M., Anciaux-Sedrakian, A., Rozanska, X., Klahr, D., Guignon, T., Fleurat-Lessard, P., Accelerating VASP electronic structure calculations using graphic processing units. Journal of Computational Chemistry 2012, 33, 2581–2589

Accelerating DFT and hybrid DFT in VASP Using GPUs

249th National Meeting (2015)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.