Remote Desktop Connection Compatible with CUDA?

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear All,

I’ve got the hardware ready to move over to CUDA (currently using Brook). There is just something that we need to know.

Basically, the 7900-GT and 8800-Ultra equipped machines I have are hidden away in our Linux cluster here at Manchester. I use Microsoft Remote Desktop Connection from my laptop to access both of them.

This was working fine with Brook for compiling purposes. However, when it came to running the program, I was presented with an error. It looks like my Brook applications are trying to run on the Trident 32Mb graphics chip I have here in this laptop, which of course does not work.

Has anyone here used CUDA using any form of remote access? My work is split between two establishments, and it would be better for the computer to stay in one remotely-accessible place. Would I need something deeper than RDC to do this?

Thanks,

Mike

2

Remote desktop indeed doesn’t work with cuda because the graphics adapter isn’t recognized. Using VNC removes this problem.

3

I thought we might have to go to VNC. Right, got some documentation to read. Thank you for your reply.

4

If you’re going to work over WAN VNC might be too slow. I’d recommend you to try RAdmin by Famatech which is much faster (but it’s not free:( )

5

I dont understand how RDC can create a problem.

In the setup I have here, I have 2 graphics cards. One is an in-built standard VGA card and other is a PCIE GEForce 8800 card. I have NOT connected the NVIDIA card to the monitor as I intend to use it only for CUDA purposes. The standard VGA is selected as the primary display device in the BIOS as well as in the OS. Now, I have connected to this machine via Remote Desktop Connection and working.

My program fails as “cudaMallocHost” function returns “Initialization Error” (3). Has this got something to do with “Remote Desktop Connection” ?? I would be baffled if it were so.

I have just finished installing the CUDA toolkits and the VC++ 9.0 compiler. THe machine is pending a reboot. Could the error that I am seeing be related to that?

Thanks for any inputs.

6

Per the CUDA release notes, remote desktop cannot be used with CUDA.

7

Well, I cant find that in either CUDA Release notes or the SDK release notes. How would RDC affect CUDA when the primary display on the machine is an onboard video card (not the NVIDIA one). This is baffling.

Probably, RDC shuts down all local DISPLAY devices when an RDC connection is made from outside ??? (Aaargh…)

8

This has been discussed several times.

CUDA WILL NOT WORK WITH RDP. RDP was designed for other things. If you don’t belive me — go and test it, maybe you’ll get positive results.

With Windows XP this probably won’t work becuase you have to install NVIDIA card as primary display adapter too, because Windows prior to Vista don’t allow you to have more than one display driver loaded at a time.

RDP actually doesn’t care about installed display adapters, it uses it’s own virtual display adapter driver.

9

This is totally un-called for and off-topic as you would like to call.

I ran into this problem and so I found a relevant thread in this forum and posted my observation. Not a question of believing you or not. btw, you are not god.