CUDA & PhysX with Quadro FX 4600 and Tesla C870 Dual GPU not available in CUDA or PhysX

Hello All,

Well I seem to have a problem that others are having as well.

System Specs:

EVGA nForce 680i SLI motherboard
8GB Corsair DDR2 1066
Q6850 Core 2 Extreme

Quadro FX 4600 in Slot 1 PCI-E
Tesla C870 in Slot 2 or 3 PCI-E
Drivers: Nvidia CUDA drivers Vista 64 177.84, CUDA Toolkit 2.0, CUDA SDK 2.02.0811

Monitors: Two Dell 24” DVI

Operating System Vista 64bit SP1

Problem:

I have done an uninstall and register clean of previous drivers. I installed the CUDA drivers, SDK and Toolkit ( listed above). Windows Device Manager shows the Quadro FX 4600 and the Tesla C870.

When I run “Simple multi-GPU” from the SDK it only reports finding one device the Quadro FX 4600. When I run any of the Benchmarks or N-body it gives me the same results at having only the Quadro FX 4600 installed.

When I open the PhysX Properties is does not offer any selection for multi GPU.

Since the Tesla does not have any video connectors in the back I have no way of performing the “Vista Fix” that requires a monitor connected to the second GPU card. I have gone into the Windows Display Settings and set the 3. (Default Monitor) on Nvidia Tesla C870 the same resolution as the monitors on the Quadro FX 4600 (1920 x 1200) as well as check marked “Extend the desktop…”, but when I click Apply it does not apply. I think this trick only works if you have an actual monitor connected to the card. Again the Tesla has no connector to connect a monitor.

I have also experimented with moving the Tesla from the SLI slot #2 to PCI-E slot #3. This did not solve the problem. My motherboard does not have jumpers or settings in the BIOS to turn SLI ON/OFF. To be expected the Nvidia Control Panel does not offer SLI anyway since the two cards are not identical.

Please help to resole this issue.
Thanks

SilulationMAN

Well I have been doing some more research and I think this all boils down to Windows Vista and the present Tesla Driver not being compatible.

It’s funny, because when I go to the download driver page on Nvidia’s website and I select Tesla as the product, there is no selection for Vista drivers. On the otherhand, if I go to the CUDA webpage and download the CUDA driver for Vista 64 the driver recognises the Tesla C870 just fine (it’s in the INF file).

I think there is a loop-hole here that should not have happened. The fact that the CUDA driver installed the Tesla card gives me the false impression that the Tesla works in Vista.

Well I guess until Nvidia release a new driver that supports Tesla cards properly in Vista I have an expensive Supercomuter doing nothing <img src=‘http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/crying.gif’ class=‘bbc_emoticon’ alt=’:’(’ /> Unfortunately the software I have been writing works only in Vista (DX10) so I can’t switch back to obsolete XP.

If anyone has any help…that would be great… or any Idea when Vista will be supported ( I thought XP has sunset. Why isn’t Vista first :( )

Thanks

SimulationMan

Hi

We are actively working on supporting Tesla under Vista. We expect full support within a few months.

Sumit

Wow it looks like Vista gets a lot less priority than XP considering it’s been released for going on 2 years.

Please register another vote for people who would appreciate more Vista support.

Thank you.

I guess that the Windows Vista code base (including Server 2008) hasn’t seen that much use in large datacenters and high performance computing yet.

XP probably also not. I think it is much more of a matter of the changed driver model in Vista. Maybe that is causing problems.