. 2023 Jan 24;19(3):1080–1097. doi: 10.1021/acs.jctc.2c01078

Table 3. Average Elapsed Time t̅ for Local GPGG Geometry Optimization on N_cores Cores (or GPU) for Different Surrogate Methods Based on the GPGG Test Set^a.

Surrogate	t̅ (min)	N_cores	t̅_GA^b
GAFF AM1-BCC	0.024	1^b	3 min
GAFF HF RESP	0.028	1^b	3 min
AMOEBApro13	0.005	1 GPU^c	5 min
PM6	1.072	8^b	15 h
PM7	1.578	8^b	22 h
DFTB3-D3(BJ)	1.331	1^b	2.7 h
B3LYP/6-31G	19.508	2 GPUs^c	1 mth

B3LYP/6-31G(d,p)	150.235	8^b	2.7 mths

Inline graphic is an estimate of the average time spent on fitness evaluations for a 60-generation 40-individual GA run if executed on a 24-core 2-GPU workstation^b. The calculation details of each method are reported in Section 3.2.

24-core Intel Xeon E5-2650 v4 @ 2.20 GHz CPU, 2 Nvidia GeForce GTX 1060 GPUs.

16-core Intel Xeon E5-2630 v3 @ 2.40 GHz CPU, 2 Nvidia GeForce GTX 970 GPUs.

Table 3. Average Elapsed Time t̅ for Local GPGG Geometry Optimization on Ncores Cores (or GPU) for Different Surrogate Methods Based on the GPGG Test Seta.

Table 3. Average Elapsed Time t̅ for Local GPGG Geometry Optimization on N_cores Cores (or GPU) for Different Surrogate Methods Based on the GPGG Test Set^a.