I’m currently running CUDA just with one 8800gtx in one machine, and it works just fine.
Reading the forum though, seems like some people uses 8800gtx only as a computation device, and have another video card for display.
Is there any advantage in doing this? more available memory i guess?
The main reason people do this is to work around the watchdog timer, which limits maximum kernel execution time to around 5 seconds.
See Question 23 in the FAQ: "What is the maximum kernel execution time? "
I have an NVIDIA graphics card as the primary display and I use RealVNC to access it. I am able to run kernels that take as long as 9.26 seconds. I am not seeing any problems.
At the same time, I have also seen the watch dog timer kicking around after 4:00 secs when my (bad) kernels hang…
So, Is VNC actually saving my watchdog from firing when the kernel is working fine ? I have NOT connected any monitor to my NVIDIA 8800 GTX. Its primarly accessed by VNC.
I use a WIndowx XP box with CUDA 1.1