My C870 now has a default monitor

[bemobali]

Hi,

I just installed the GeForce driver package 182.08, which installes GeForce driver version 7.15.11.8208 on my 8800GTX card. The C870 driver version is still 7.15.11.7781. In my Vista 32 system, I now notice that Vista shows a default monitor device on my Tesla C870 card. I find this weird because the C870 card does not have a monitor output. At least I have not seen it :)

Has anyone else observe this? Is there a reason for this default monitor?

Thanks,

[MisterAnderson42]

Is there a reason for this default monitor?

As I understand it, Vista considers GPUs without monitors to be utterly useless and won’t even let the drivers access them. NVIDIA’s drivers therefore must fake a monitor on Tesla cards in order for them to be usable.

[Mr_Nuke]

As I understand it, Vista considers GPUs without monitors to be utterly useless and won’t even let the drivers access them. NVIDIA’s drivers therefore must fake a monitor on Tesla cards in order for them to be usable.

That’s wierd. I have a couple of systems running Server 2008 with GPUs with no montors attached, and it detects them easily. Could it be a Vista-only limitation?

[tmurray]

Here’s the secret sauce for enabling CUDA on non-display cards in Vista or Server 2008 using 18x.xx drivers. It’s what the PhysX control panel does for one card (sort of, I think this is more thorough and also makes the card show up as no timeout enabled? but you still need to disable the TDR timeout to make that work correctly, I think):

• In the registry, go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Video(GPU ID)\0000 (you might also have to hit 0001 on a D870 or Plex D2)
• add the following two DWORDs and set them to 1:

DisplayLessPolicy
LimitVideoPresentSources

• reboot.

I don’t know of a good way to look up GPU IDs, so look at the name entries in each to figure out what you should be touching.

[Keldor314]

Here’s the secret sauce for enabling CUDA on non-display cards in Vista or Server 2008 using 18x.xx drivers. It’s what the PhysX control panel does for one card (sort of, I think this is more thorough and also makes the card show up as no timeout enabled? but you still need to disable the TDR timeout to make that work correctly, I think):

• In the registry, go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Video(GPU ID)\0000 (you might also have to hit 0001 on a D870 or Plex D2)

• add the following two DWORDs and set them to 1:

DisplayLessPolicy

LimitVideoPresentSources

• reboot.

I don’t know of a good way to look up GPU IDs, so look at the name entries in each to figure out what you should be touching.

Strange - on my system LimitVideoPresentSources already exists, but is a REG_BINARY. What’s with that?

[tmurray]

Strange - on my system LimitVideoPresentSources already exists, but is a REG_BINARY. What’s with that?

I’ve seen that too, it might be how the PhysX control panel creates it.

[morten]

Here’s the secret sauce for enabling CUDA on non-display cards in Vista or Server 2008 using 18x.xx drivers. It’s what the PhysX control panel does for one card (sort of, I think this is more thorough and also makes the card show up as no timeout enabled? but you still need to disable the TDR timeout to make that work correctly, I think):

• In the registry, go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Video(GPU ID)\0000 (you might also have to hit 0001 on a D870 or Plex D2)

• add the following two DWORDs and set them to 1:

DisplayLessPolicy

LimitVideoPresentSources

• reboot.

I don’t know of a good way to look up GPU IDs, so look at the name entries in each to figure out what you should be touching.

Hi,

I have tried this on Windows 7 x64 with one GTX 295, but only one core is known to Cuda, despite 2 are known to the operating system/Nvidia.

Morten

[Nitrobass24]

Anybody figured out how to lookup GPUID’s?

I have a system with only one GPU a 9800gt, but I cant figure out what registry value to edit, because there are so many GPUID’s under Video?
Also do you need to extend the desktop for this to work?

[evanevery]

Here’s the secret sauce for enabling CUDA on non-display cards in Vista or Server 2008 using 18x.xx drivers. It’s what the PhysX control panel does for one card (sort of, I think this is more thorough and also makes the card show up as no timeout enabled? but you still need to disable the TDR timeout to make that work correctly, I think):

• In the registry, go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Video(GPU ID)\0000 (you might also have to hit 0001 on a D870 or Plex D2)

• add the following two DWORDs and set them to 1:

DisplayLessPolicy

LimitVideoPresentSources

• reboot.

I don’t know of a good way to look up GPU IDs, so look at the name entries in each to figure out what you should be touching.

In my UNEDITED Registry “LimitVideoPresentSources” is a binary value “01 00 00 00” at the 0000 level, and, “DisplayLessPolicy” was found at the {GPUID} parent level. I’m running 186.18 drivers. Do you know why “DisplayLessPolicy” appears at at a location other than what you specify or why “LimitVideoPresentSources” is a binary value rather than a DWORD? Is there any documentation on what these do specifically? (I’m trying to troubleshoot a related issue with 4 x GTX-295 in a benchmark config)

[silveradocyn]

I am seeing the same values in my Win7 x64 System.

I am using driver 191.07.

It is interesting that in Win7, the GUID key does not seem to be generated unless a monitor has been attached to the port.

Has anyone been able to workthrough this?

• • Silveradocyn

[tmurray]

can you activate it and extend a fake desktop to it? I did that once with Win7 (screen resolution → detect → display VGA? something like that), so I think that might work if it actually doesn’t generate the GUID section (which seems weird, that should be generated at driver install time).