Performance of Coupled-Cluster Singles and Doubles on Modern Stream Processing Architectures

被引：20
|
作者：
Fales, B. Scott [1 ,2 ,3 ]
Curtis, Ethan R. [1 ,2 ,3 ]
Johnson, K. Grace [1 ,2 ,3 ]
Lahana, Dean [1 ,2 ,3 ]
Seritan, Stefan [1 ,2 ,3 ]
Wang, Yuanheng [1 ,2 ,3 ]
Weir, Hayley [1 ,2 ,3 ]
Martinez, Todd J. [1 ,2 ,3 ]
Hohenstein, Edward G. [1 ,2 ,3 ]
机构：
[1] Stanford Univ, Dept Chem, Stanford, CA 94305 USA
[2] Stanford Univ, PULSE Inst, Stanford, CA 94305 USA
[3] SLAC Natl Accelerator Lab, Menlo Pk, CA 94025 USA
来源：
JOURNAL OF CHEMICAL THEORY AND COMPUTATION | 2020年 / 16卷 / 07期
关键词：
BODY PERTURBATION-THEORY; ELECTRON CORRELATION; QUANTUM-CHEMISTRY; PARALLEL IMPLEMENTATION; APPROXIMATE INTEGRALS; NATURAL ORBITALS; EFFICIENT; CCSD; ALGORITHM; ENERGY;
D O I：
10.1021/acs.jctc.0c00336
中图分类号：
O64 [物理化学（理论化学）、化学物理学];
学科分类号：
070304 ; 081704 ;
摘要：
We develop a new implementation of coupled-cluster singles and doubles (CCSD) optimized for the most recent graphical processing unit (GPU) hardware. We find that a single node with 8 NVIDIA V100 GPUs is capable of performing CCSD computations on roughly 100 atoms and 1300 basis functions in less than 1 day. Comparisons against massively parallel implementations of CCSD suggest that more than 64 CPU-based nodes (each with 16 cores) are required to match this performance.
引用
收藏
页码：4021 / 4028
页数：8
相关论文
共 50 条
12345