Getting setup (over Remote Desktop

I’ve got the 3.2 Toolkit and driver for Quadro 2000 installed, and I’m able to build bandwidthtest with VS 2008 SP1. I have run it (and particles) from my machine, and everything looks ok.

Now I’m at the point where I’m trying to write my own hello world program, and I’m getting strange behavior from basic things, like cudaGetDeviceCount and cudaGetDeviceProperties (which returns 0x26 = 38 decimal, which driver_types.h says means “No devices”.) So I went back to bandwidthTest, and I see the same behavior there. The difference? Now I’m coming in over remote desktop.

Is it possible that there would be no cuda devices available for remote desktop sessions? Even if just writing a simple, command-line app? I REALLY need a work around for this.


You need to use TCC, which means you need a dedicated Tesla card for compute. (Blame WDDM.)

The easiest work-around is a program called VNC if you can’t use the TCC drivers.

It is also possible to setup cygwin to allow you to connect with ssh. That might work for writing/running command line apps, but I’m not sure. You wouldn’t be able to use visual studio this way - you would have to use an editor like vim and compile everything from the command line.

There’s not a way to turn the graphics on the quadro off to make it just a compute engine, is there? I have an integrated gfx controller, which would work just fine. I just need the CUDA cores…

Down the rabbit hole I go…


Why nVidia forbids TCC feature for GTX cards ?

I would buy 2 GTX 580 - one for CUDA only - without display attached to it. It would be great to have TCC RDP feature for GTX cards as well.

Can you provide us some technical details why is such feature impossible to implement on GTX cards without display attached ?

many thanks