Time limit on Linux ? Is there one ?


I have a small test program that fails on large datasets, but works on smaller ones.

Actually I have 2 versions : one fast version, that “works” on all sizes (~takes 1.4 seconds on big case)
And a slow version, which uses only global memory. Works on smaller cases, but on the big one, after ~11s the program ends (no error message), but the output array is all zeros, looks like all threads where ‘killed’ before the end.

I am aware of a 5s time limit on Windows, but what about Linux ? Nothing about that in the FAQ.

I’m using Red Hat 4 Update 4, cuda 1.0.



Linux also has a watchdog timeout if you’re using X on the same GPU where CUDA is being used. There is no timeout if X is not in use on the CUDA GPU.

OK, thank you.

Hi again,

I have disabled X and yet I’m still getting “unspecified launch failure” when my program takes more than a few seconds.

Could this be another issue ?

Btw I killed X but I still have the console on this card though.


I have installed two GTX 8800 cards and the system now rusn Fedora Core 6. I am wondering whether I can make X run on one GPU card and run my CUDA program on the second card, so that I can avoid the “watch dog”? Thank you!!!

I have the same doubt. Relases notes are not very explicative:

"Individual GPU program launches are limited to a run time

of less than 5 seconds on a GPU with a display attached.

Exceeding this time limit causes a launch failure reported

through the CUDA driver or the CUDA runtime. GPUs without

a display attached are not subject to the 5 second run time

restriction. For this reason it is recommeded that CUDA is

run on a GPU that is NOT attached to an X display."

"While X does not need to be running in order to use CUDA,

X must have been initialized at least once after booting

in order to properly load the NVIDIA kernel module. The

NVIDIA kernel module remains loaded even after X shuts down,

allowing CUDA to continue to function."

Basically a X display is no more no less that the X server.

So does it remove that limit if you shutdown X server and run cuda under full screen console?

Someone has tryed that?

The issue here is purely that if the GPU is being used by X, the watchdog timeout will be in force. If the GPU is not being used by X, then there shouldn’t be any watchdog timeouts.

I still have not downloaded cuda and like to know if is it possible to use this trick for the 5s limit.

Did you use “init 3” to shut down X? Check also “ps -e” if X is still running.

Ok. So killing X will work! As i will use it for launching time intensive simulations I do not need a X disply.