view article
Figure 2
NVIDIA visual profiler snapshot of launch times. Implementing standard launches from the host upon completion (left) causes driver overhead in the range of tens of microseconds per operation. Using CUDA stream memory operation (right) decreases latency to a few microseconds.

Journal logoJOURNAL OF
SYNCHROTRON
RADIATION
ISSN: 1600-5775
Volume 27| Part 5| September 2020| Pages 1297-1306
https://doi.org/10.1107/S1600577520008140
Follow J. Synchrotron Rad.
Sign up for e-alerts
E-alerts
Follow J. Synchrotron Rad. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS