CUDA graphic cards requirements

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello everyone,

I am a new (potential) CUDA user and I want to setup properly my computer. I read in the CUDA 2.0 relase notes that CUDA programs should run on a NVidia card that does not drive a monitor due to the 5 sec Windows XP limitation. Currently I am using an ATI graphics card connected to my motherboard’s single PCI Express x16 slot.

I would like to ask if it would be possible to buy a NVidia card that connects to a PCI-E x1 (the tiny connector) and use it for my CUDA applications, while still keeping the ATI card for driving the display. Is there going to be a driver conflict?

Thank you very much for your time!

Kind regards,
dtheodor

2

You’ll be much better off doing both computation and display on a single, good card. Very few kernels actually hit the 5 second limit (it’s the maximum time a single kernel can take to execute, not the whole application - applications can run for hours) and the performance hit due to displaying a few windows here and there is negligible.

And yes, you’d likely have a driver conflict with those two cards.

3

Hello Big_Mac,

Thank you very much for your reply and useful information. It seems then that tomorrow I’ll buy one new NVidia card to replace the old ATI one, and start coding… :)

Kind regards,
dtheodor