I profiled my application which use OpenGL code and therefore uses nvoglnt.dll
At first I used Parallel Amplifier for the profiling and later also with the VTune.
I discovered that calls to DrvCopyContext take about 20 percents of the whole program execution. The DrvCopyContext in his turn call to WaitForSingleObject.
I have the latest version of the driver (which is 195.6 I think) using Windows XP SP3 and NVidia 8800 graphics card.
Does anyone aware of this problem?
Is it possible to improve performace (like enforching the driver to run on different thread)?
Does anyone knows if there are certain OpenGL functions that eventually call this function and otherOpenGL function that can do the same (even if it requires more sophisticated code) but bypass these calls?
Could it be that the interaction between the OS and the driver cause the problem and with other OSs it doesn't happen?