view article
Figure 5
(a) The time to complete 300 difference-map iterations for different thread counts on a single node. (b) The time for the maximum-likelihood refinement to reach a specific error value limit for various thread counts on a single node. (c) The I/O time to read the data and write the reconstruction result for various thread counts. Due to the use of a single-threaded HDF5 library, no performance gain was achieved by using more than one thread during I/O. For (a) and (b), the performance benefits of more threads are clearly visible. The increased synchronization time in the case of many threads, as well as the limited memory bandwidth, limits the performance gain in the case of more than 20 threads.

Journal logoJOURNAL OF
APPLIED
CRYSTALLOGRAPHY
ISSN: 1600-5767
Volume 53| Part 2| April 2020| Pages 574-586
https://doi.org/10.1107/S1600576720001776
Follow J. Appl. Cryst.
Sign up for e-alerts
E-alerts
Follow J. Appl. Cryst. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS