TABLE III.

Processing parameters and time for each dataset. Wall clock time for analysis of three different datasets is provided as a benchmark for performance. All run times are presented in HH:MM:SS format (omitting zero values). For the multi-pass PDZ2 data, Laue-DIALS runtimes are presented for a single pass of 45 images, where 4 CPUs were used for spot finding and indexing, 8 CPUs were used for refinement and spot prediction, and 1 CPU was used for integration. Careless runtime includes all 1064 images. For simulated DHFR data (reference data only) 4 CPUs were used for all processes except for laue.integrate (2 CPUs) and Careless merging (1 GPU). Careless merging times are for training only, and not for running on half-dataset repeats. For the Anomalous HEWL data, 8 CPUs were used for indexing through refinement, 48 CPUs were used for spot prediction, and 64 CPUs were used for integration.

Dataset	Number of images	Number of CPUs	Indexing through refinement	Spot prediction	Integration	Careless merging
Simulated DHFR	180	4/2	12:52	8:10	1:06:67	39:00
Anomalous HEWL	3049	8/48/64	3:07:11	6:18:56	1:15:37	1:21:53
PDZ2	1064	4/8/1	2:35	0:17	15:23	20:00