0

I am constantly uploading two images from two cameras on the GPU. I am using GPU SURF to find corresponding features. However, after a while running the program, I am receiving this error:

opencv-2.4.9/modules/nonfree/src/cuda/surf.cu:250: error: (-217) the launch timed out and was terminated in function icvCalcLayerDetAndTrace_gpu Aborted (core dumped)

I am calling the GPU SURF from a thread. Any help please? Thanks!

afsaneh R
  • 51
  • 1
  • 3
  • 2
    Is this on windows or linux? A launch timeout means a kernel took longer than the watchdog timer would allow. It may indicate an application problem, but not necessarily. It might just mean that the GPU needs more time to run the tasks assigned to it. If things otherwise seem to be working correctly, you could try increasing the timeout period. The method will depend on whether you are on windows or linux. You might also indicate what GPU you are running on, and whether it is hosting a display. – Robert Crovella Sep 03 '15 at 17:30
  • @RobertCrovella Thanks, It's ubuntu 14and Its OpenCV 2.4.9 on CUDA 6.5 – afsaneh R Sep 03 '15 at 18:38
  • 1
    Is the gpu hosting a display? – Robert Crovella Sep 03 '15 at 19:07
  • I am not exactly sure what you mean by that. Do you mean that I am visualizing something? I am visualizing the point clouds being rendered on cloud viewer but I am not using the gpu for that.. – afsaneh R Sep 03 '15 at 19:47
  • another question is that can two threads perform gpu surf feature matching? – afsaneh R Sep 03 '15 at 19:48
  • 1. Are you using a gpu? 2. Which GPU is it? 3. Is a video monitor plugged into that GPU? 4. Is that video monitor displaying some portion of a graphical desktop or window? – Robert Crovella Sep 03 '15 at 19:54
  • I am asking simple questions I think. yes, it is NVIDIA GeForce GTX 570 and a monitor is connected to it. and I am working on that monitor. – afsaneh R Sep 03 '15 at 20:02
  • Then that GPU is hosting a display and anything you do on it is subject to the Linux display watchdog. You may want to read this article: http://nvidia.custhelp.com/app/answers/detail/a_id/3029/~/using-cuda-and-x – Robert Crovella Sep 03 '15 at 20:22
  • Thanks! Very useful article! Never thought the problem could be this, I was digging every where in my code checking locks and stuff! – afsaneh R Sep 03 '15 at 20:47

0 Answers0