Translations:OpenACC Tutorial - Optimizing loops/20/en

From Alliance Doc
Jump to navigation Jump to search

Once you have performed those steps, you should have the following information displayed. Fifth step of guided analysis This screenshot gives us a number of important information. First, the text tells us clearly that the performance is limited by the size of the blocks, which in OpenACC corresponds to the size of the gangs. Second, the "Active Threads" line tells us that the GPU is running 512 threads, while it could be running 2048. The occupancy line correspondingly states that the GPU is only used at 25% of its capacity. Occupancy is the ratio of how much the GPU is utilized over how much the GPU could be utilized. Note that 100% occupancy does not necessarily yield the best performance. However, 25% is quite low.