cudaErrorLaunchTimeout

Accelerated Computing CUDA CUDA Programming and Performance
1

Does anyone have an idea about what this error means and how it can be resolved?

cudaErrorLaunchTimeout

2

Does this appears after 5 seconds of computation? When monitor is connected to videocard, there is 5 seconds limit for computations.

3

No. This is the return value of cudaMemcpy, when I try to copy a 400 size array of floats from device memory to host memory. Do you have any idea of what this return value means?

4

Does no one have any idea? I am really stuck just because of this. If anyone has any information, I would be very grateful if he could share it with me.

5

Just extract minimal code which reproduces the problem and post it here.

6

If you do not check for errors after your kernel call, the next things is going to give an error, so if this is the cudaMemCpy after your kernel call that is throwing the error, it is the kernel that caused the error.

So yes, your kernel was running for more than 5 seconds and got killed by the watchdog.

7

I am using Tesla, without the monitor being connected to the GPU, so there is no question of a 5 second limit. I guess the problem was due to latency on part of the CPU, because after restarting the machine, I am no longer getting that error. Anyways, thank you so much for your help.

8

I have got the same problem.But I am using 9500GT with the monitor being connected to the GPU.So ,does my kernel run for more than 5 seconds?