Problem in global memory access when multiple threads trying to access same location

Accelerated Computing CUDA CUDA Programming and Performance
21

Hi,

@Dittoaway: yes, thts correct.

I am afraid I forgot to mention this before: if TERMS was less than 5000, I get correct result. But for value higher than that, am getting all 0s.

I just tried checking cuda error after the kernel call and am getting this error:
cuda error!!! the launch timed out and was terminated.

Anyone has any idea what should this indicate…? and is there any workaround…? I read in some of the posts that this might be due to watchdog timer… Am running on Linux machine and dont know if there is any work around…

Thanks!

22

Hi,

@Dittoaway: yes, thts correct.

I am afraid I forgot to mention this before: if TERMS was less than 5000, I get correct result. But for value higher than that, am getting all 0s.

I just tried checking cuda error after the kernel call and am getting this error:
cuda error!!! the launch timed out and was terminated.

Anyone has any idea what should this indicate…? and is there any workaround…? I read in some of the posts that this might be due to watchdog timer… Am running on Linux machine and dont know if there is any work around…

Thanks!

23

Ahhh. Yes, that’s the watchdog timer.
Any possibility of running your monitor on a 2nd graphics card?

24

Ahhh. Yes, that’s the watchdog timer.
Any possibility of running your monitor on a 2nd graphics card?

25

There is only 1 graphics card on that machine. :| Um… but its a remote server that am using… It need not have a monitor as such… I dont understand the concept of this watchdog timer really :P

26

There is only 1 graphics card on that machine. :| Um… but its a remote server that am using… It need not have a monitor as such… I dont understand the concept of this watchdog timer really :P