cutil Timer and visual profiler

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

i’ve tested running-time of my program both with cutil Timer and visual profiler, but the results are very different!

in Profiler shows, my kernel needs GPU Time 289187 (in usec??). Using cutilCheckError(cutStartTimer(timer)); and cutilCheckError(cutStopTimer(timer));
showes just 0,143321 ms.

who has any ideal?

regards!

2

The host timer is probably just measuring the kernel launch time and not the run time. You will need to add a cudaThreadSynchronize() call before stopping the timer to get an reasonable estimate of total kernel run time.

3

problem solved! thx