Kernel Analysis

For analysis of the kernel side of the application, the following reports are generated:

Occupancy

The Occupancy report shows, for each kernel, the occupancy of each execution unit in the GPU.

You can also see the number of GPU threads launched, and the min, max and average thread execution time.

Ocupancy of execution units during kernel execution

Ticks per Thread

The Ticks per Thread report shows, for each number of active threads, the amount of time this number of threads was active.
Tick per number of threads graph

Threads per Time

The Threads per Time report shows this number of threads that were active at each point in time during the execution.

Latency

The Latency pane shows, for each kernel file, the overall latency of the memory commands.
Click the kernel name to see the latency of each memory command in the source code of this kernel.

latency view