Journal of Computer Chemistry, Japan [No.2015-0041] Published online by J-STAGE
<Title:> OpenFMOにおける4中心クーロン相互作用項計算のGPGPU化の試み
<Author(s):> 梅田 宏明, 塙 敏博, 庄司 光男, 朴 泰祐, 重田 育照
<Corresponding author E-Mill:> umeda(at)ccs.tsukuba.ac.jp<Abstract:> GPU acceleration of four-center (4C) inter-fragment Coulomb interaction term (IFC) for OpenFMO, a fragment molecular orbital calculation program, has been implemented and its performance was examined. FMO calculation has two time-consuming steps: Fock matrix construction and IFC calculation, and in our previous letter, it was reported that the former is successfully accelerated with our GPU-enable code. The 4C-IFC calculation is the core part of the latter and its code is similar to that of Fock matrix construction. In this letter, we briefly describe the GPU-accelerated 4C-IFC calculation routine, and report a performance benchmark for GPU-accelerated FMO calculation. The GPU-accelerated program shows 3.3× speedups from CPU only FMO-HF/6-31G (d) calculation for 642 atomic protein on 8 nodes of HA-PACS base cluster.
<Keywords:> GPGPU, FMO, Inter-fragment Coulomb interaction, Molecular orbital calculation, Performance benchmark
<URL:> https://www.jstage.jst.go.jp/article/jccj/14/3/14_2015-0041/_article/-char/ja/
<Title:> OpenFMOにおける4中心クーロン相互作用項計算のGPGPU化の試み
<Author(s):> 梅田 宏明, 塙 敏博, 庄司 光男, 朴 泰祐, 重田 育照
<Corresponding author E-Mill:> umeda(at)ccs.tsukuba.ac.jp<Abstract:> GPU acceleration of four-center (4C) inter-fragment Coulomb interaction term (IFC) for OpenFMO, a fragment molecular orbital calculation program, has been implemented and its performance was examined. FMO calculation has two time-consuming steps: Fock matrix construction and IFC calculation, and in our previous letter, it was reported that the former is successfully accelerated with our GPU-enable code. The 4C-IFC calculation is the core part of the latter and its code is similar to that of Fock matrix construction. In this letter, we briefly describe the GPU-accelerated 4C-IFC calculation routine, and report a performance benchmark for GPU-accelerated FMO calculation. The GPU-accelerated program shows 3.3× speedups from CPU only FMO-HF/6-31G (d) calculation for 642 atomic protein on 8 nodes of HA-PACS base cluster.
<Keywords:> GPGPU, FMO, Inter-fragment Coulomb interaction, Molecular orbital calculation, Performance benchmark
<URL:> https://www.jstage.jst.go.jp/article/jccj/14/3/14_2015-0041/_article/-char/ja/