PGI User Forum
 SearchSearch   MemberlistMemberlist     RegisterRegister   ProfileProfile    Log inLog in 

Free OpenACC Webinar

launch timed out and was terminated....
Goto page Previous  1, 2
 
Post new topic   Reply to topic    PGI User Forum Forum Index -> Accelerator Programming
View previous topic :: View next topic  
Author Message
mkcolg



Joined: 30 Jun 2004
Posts: 6218
Location: The Portland Group Inc.

PostPosted: Tue Jun 21, 2011 3:12 pm    Post subject: Reply with quote

Hi cyFeng,

I suspect it's the same issue where the OS is killing your long running job. This can occur if your GPU is attached to a monitor.

Note the following code in your kernel:
Code:
     if ( I <= N )then
       do j=1, N
         C(i)= A(i) + B(i)
       end do
     end if

You have every thread execute the vector add N times. Granted you may be doing this on purpose for benchmarking, but it is causing your code to take a lot longer then it should. To fix, remove the do loop.

Hope this helps,
Mat
Back to top
View user's profile
cyFeng



Joined: 05 Mar 2010
Posts: 17

PostPosted: Sun Jun 26, 2011 5:23 am    Post subject: Reply with quote

Hi Mat,

I did the benchmark on my notebook characterized by the NVIDIA OPTIMUS technique. So, the GPU should be attached to monitor. Thank you for the response.
Back to top
View user's profile
cyFeng



Joined: 05 Mar 2010
Posts: 17

PostPosted: Sun Jul 03, 2011 1:31 am    Post subject: Reply with quote

Is there any solution to prevent the running job from being killed by the OS ?
I have the only one laptop to test Cuda Fortran.. :(
Back to top
View user's profile
mkcolg



Joined: 30 Jun 2004
Posts: 6218
Location: The Portland Group Inc.

PostPosted: Wed Jul 06, 2011 12:47 pm    Post subject: Reply with quote

Hi cyFeng,

Do a web search for "CUDA Windows Watchdog Timer" and you'll find a work around. However, the work around requires you to edit your registry and disable the GPU watchdog, leaving your systems susceptible to freeze-ups. You can try this but it's not recommended. Instead, you should consider breaking your long running kernels into smaller, shorter, ones, using smaller data sets, or getting a dedicated compute GPU.

- Mat
Back to top
View user's profile
cyFeng



Joined: 05 Mar 2010
Posts: 17

PostPosted: Thu Jul 07, 2011 7:04 pm    Post subject: Reply with quote

Hi, Mat

Actually.. I'm breaking the long running kernels into smaller ones.
That is the spirit of CUDA.
Thanks for your kind response :)

Feng
Back to top
View user's profile
Display posts from previous:   
Post new topic   Reply to topic    PGI User Forum Forum Index -> Accelerator Programming All times are GMT - 7 Hours
Goto page Previous  1, 2
Page 2 of 2

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © phpBB Group