Performance of the ANI-2X neural network in Deep-HP in terms of molecular dynamics simulation production (ns per day) for selected water boxes of increasing sizes using Nvidia V100 and A100 GPU cards^a.

Systems (number of atoms)/number of GPU devices	1	4	8	16	28	44	68	84	100	124
GPU V100
Puddle (96 000)	0.11	0.27	0.44	0.67	0.70	0.78	0.91	1.05	1.05	1.05
Pond (288 000)	n/a	0.11	0.19	0.31	0.46	0.57	0.66	0.67	0.71	0.71
Lake (864 000)	n/a	n/a	0.07	0.10	0.19	0.26	0.33	0.40	0.48	0.40
Bay (2 592 000)	n/a	—	n/a	0.04	0.06	0.09	0.14	n/a	n/a	n/a
Sea (7 776 000)	n/a		—	—		n/a	0.04	0.05	0.06	0.06
GPU A100
Puddle (96 000)	0.16	0.41	0.63	n/a			—			n/a
Pond (288 000)	n/a	0.16	0.26	n/a			—			n/a
Lake (864 000)	n/a	n/a	0.11	n/a			—			n/a
Theoretical performance (V100)
Puddle (96 000)	0.11	0.27	0.46	0.75	0.79	0.90	1.14	1.39	1.40	1.40
Pond (288 000)	0.03	0.11	0.20	0.33	0.49	0.65	0.77	0.89	0.88	0.89
Lake (864 000)	0.01	0.02	0.07	0.14	0.21	0.30	0.38	0.49	0.59	0.49
Bay (2 592 000)	0.004	0.007	0.02	0.06	0.08	0.12	0.16	n/a	n/a	n/a
Sea (7 776 000)	0.001	0.003	0.005	0.009	0.02	0.03	0.06	0.08	0.10	0.10

^an/a: not available.