All profiler counters are 0. Why?

Hi, all.

I began to optimize my program with CUDA profiler but encountered unexpected problem.

All counters are reported to be equal to 0. OS - Vista64, CUDA 2.3, the profiled program platform - Win32.
cudaThreadExit is called in the end of the program. Kernel calls are reported. They, of cause, access global data.
I requested all counters so that the program was launched 5 times.

Any ideas??


I have 2 CUDA devices - Tesla C1060 and GeForce 9400. The program runs on Tesla.

Profiling the demo programs from the CUDA package gives the same result.

May be, it is coz of Vista… Read up the profiler documentation to see if they have any exception for Vista…

After all, Vista is an exceptional OS. lol…