consistent 500 delay on every Nth kernel launch

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am running on Windows 7 on a GeForce GTX 280. In my app (a custom ray tracing app), I notice a regular, periodic delay when launching the kernel. On one of my simple test scenes, for example, it normally completes in about 15ms. But on every 164th call to launch the kernel, it takes about 556 ms to complete.

I noticed the problem because the app re-renders the scene when the size of its window changes, and its easy to rack up 164-size changes as you test the rendering speed by resizing the window with the mouse.

I don’t have any further data at this point (since I haven’t set up Parallel Nsight yet), but I wanted to see if this was a known behavior or issue.

Thanks,

Roland

2

Hi,

I am running on Windows 7 on a GeForce GTX 280. In my app (a custom ray tracing app), I notice a regular, periodic delay when launching the kernel. On one of my simple test scenes, for example, it normally completes in about 15ms. But on every 164th call to launch the kernel, it takes about 556 ms to complete.

I noticed the problem because the app re-renders the scene when the size of its window changes, and its easy to rack up 164-size changes as you test the rendering speed by resizing the window with the mouse.

I don’t have any further data at this point (since I haven’t set up Parallel Nsight yet), but I wanted to see if this was a known behavior or issue.

Thanks,

Roland

3

can you post the code that launches the kernel and then synchronizes on the result?

4

can you post the code that launches the kernel and then synchronizes on the result?