Performance of Coupled-Cluster Singles and Doubles on Modern Stream Processing Architectures.
Performance of Coupled-Cluster Singles and Doubles on Modern Stream Processing Architectures.
复制标题
DOI：
10.1021/acs.jctc.0c00336
复制
发表时间：
2020-06
影响因子：
5.5
通讯作者：
B. S. Fales;Ethan R. Curtis;K. G. Johnson;Dean Lahana;Stefan Seritan;Yuanheng Wang;Hayley Weir;T. Martínez;E. Hohenstein
中科院分区：
文献类型：
--
作者：
B. S. Fales;Ethan R. Curtis;K. G. Johnson;Dean Lahana;Stefan Seritan;Yuanheng Wang;Hayley Weir;T. Martínez;E. Hohenstein
我们开发了一种新的耦合集群单打和双打 (CCSD) 实现，针对最新的图形处理单元 (GPU) 硬件进行了优化。我们发现，配备 8 个 NVIDIA V100 GPU 的单个节点能够在不到一天的时间内对大约 100 个原子和 1300 个基函数执行 CCSD 计算。与 CCSD 大规模并行实现的比较表明，需要超过 64 个基于 CPU 的节点（每个节点有 16 个内核）才能匹配此性能。
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 one 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.