Hi Guys,
I am running CUDA blas on a 9800GX2 card. My application is graphics so it also invloves rendering.
So, I figured out that it will be better to use one core of my 9800GX2 for rendering and the other core for matrix operations.
The problem is that when I run the cuDeviceGetCount command, I always get 1 (I expected it to be 2).
I am using Vista with VS2005.
Any one have a clue?
Thanks!
Turn off SLI, make sure you have PhysX enabled on the other device.
I turned off SLI and still got 1 device only.
I thought that maybe a restart is needed.
After restart, I was unable to load Windows. At the stage that the welcome screen is supposed to appear, the display disappeared and there is no signal from the graphics card.
The only way out of it is to startup in safe mode, uninstall and then install the driver again.
By the way, what do you mean by “PhysX enabled on the other device”? Which one is the other device? How do I control which device is running PhysX?
Why do I need PhysX at all?
There’s an option in the NVIDIA control panel for enabling PhysX on a second card. This really means “enable CUDA on a card that is not used for display.” If you don’t enable that, you won’t be able to see the second device.
(double check that you’re using a 18x.xx driver and 2.1)
Thank you very much for your advices.
I have the latest CUDA enabled drivers with CUDA 2.1 running on Vista.
I do not see any other option in the NVIDIA control panel beside turning on/off Physics and turning on/off SLI.
There is no option for determining that Physics should run on a certain device.
The more troubling issue is that once I turn SLI off (regardless to the state of Physics), I cannot restart my computer!
Any suggestions?
I bought a new computer yesterday.
Installed a different OS (Vista 64 rather than my old 32 bit) and pluged in the 9800GX2 card.
In addition, I installed another card - Nvidia Tesla 1060 with 4GB memory.
Sli is turned off and physics is enabled on the TESLA.
When I call cuDeviceGetCount I get 2 but further investigation shows that the first CUDA device is the Tesla and a second device is one of the 9800GX2 cores.
So after all, I am still unable to “see” the two seperate devices (cores) of the 9800GX2.
Does anyone has an experience with using both cores of a 9800GX2 for CUDA?
By the way, When Sli is turned off; does it means that for rendering purposes, I am using just half power of the 9800GX2 or maybe, since it is one physical card, sli is always enabled in the driver for this specific card?