Tesla V100 + Nvidia 455.32.00: UseDisplayDevice "None" is not supported with GRID

[qwofford]

Hello,

I'm using VirtualGL 2.6.2 with a Tesla V100 in headless mode, and NVIDIA driver v455.32.00. It appears that my config file, as generated by the Headless Nvidia mini How-To, is conflicting with the V100 as configured. 

(EE) NVIDIA(GPU-0): UseDisplayDevice "None" is not supported with GRID

(EE) NVIDIA(GPU-0):     displayless

Is this something you've seen before? Looking at how to disable GRID vGPU support now, which I hope will solve the problem...

Thanks,

Quincy

[Youssef GHORBAL]

Hi,

Last time I asked Nvidia about this very message they give me this feedback:

"This is due to a change in our driver. And you are right, you need to remove the Option "UseDisplayDevice" "none" from the Xorg conf file. See release not here: https://docs.nvidia.com/datacenter/tesla/tesla-release-notes-440-3301/index.html (section 1.3 Known Issues / NoScanout Mode)
As you pointed out this is harmless. Nevertheless, I am hoping this will be improved in a future release."

Bascially, the error message is somehow misleading and has nothing to do with GRID technology at all.

Youssef

--------------------

--
You received this message because you are subscribed to the Google Groups "VirtualGL User Discussion/Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to virtualgl-use...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/f94a8abe-22fd-442c-95ab-da7d6b61cbbbn%40googlegroups.com.

[DRC]

What do they propose instead?  Is headless operation no longer supported with Tesla GPUs?

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/E626C2EF-A528-443C-8246-3A8B03FD58E0%40pasteur.fr.

[qwofford]

The headless mode works for all GPUs I've tested so far without UseDisplayDevice "None", including V100's. I'm thinking the "--use-display-device=None" breaks some subset of headless GPU/driver combinations. So...beyond some Nvidia driver version, some GPU's will break with this option. All GPU's I've tested work OK without this option, at Nvidia driver 455.32.00.

DRC, let me know if you want me to do some troubleshooting for you. Happy to help the project where I can.  I am removing this from option from our xorg.conf deployment mechanism. If I get  any complaints, I'll dig in and let you know wht I find.

[DRC]

I don't understand how the driver knows to treat the GPU as headless if you don't specify 'UseDisplayDevice "None"' in xorg.conf.

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/16ede473-5b78-44ee-9248-22271bef10a9n%40googlegroups.com.

[Jason Edgecombe]

NVidia Tesla's are designed as GPU compute cards and Graphics accelerators. They literally don't have any VGA/DisplayPort/HDMI ports on the back of the card. lspci shows them as a "3D Controller" instead of a VGA display card.

Sincerely,

Jason

---------------------------------------------------------------------------
Jason Edgecombe | Linux Administrator
UNC Charlotte | Office of OneIT
9201 University City Blvd. | Charlotte, NC 28223-0001
Phone: 704-687-1943
jwed...@uncc.edu | http://engr.uncc.edu |  Facebook
---------------------------------------------------------------------------
If you are not the intended recipient of this transmission or a person responsible for delivering it to the intended recipient, any disclosure, copying, distribution, or other use of any of the information in this transmission is strictly prohibited. If you have received this transmission in error, please notify me immediately by reply e-mail or by telephone at
704-687-1943.  Thank you.

On Tue, Mar 2, 2021 at 12:46 PM DRC <d...@virtualgl.org> wrote:

[Caution: Email from External Sender. Do not click or open links or attachments unless you know this sender.]

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/cf378d92-6b5d-30d0-9ab6-8e927a6137ab%40virtualgl.org.

[DRC]

OK, so does that mean that 'UseDisplayDevice "None"' is still supported with GPUs that have display connections?  If not, then how would one specify headless mode with those GPUs?

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/CAAR6MGC-%3DVVQtWPoEMuAAC4NGptzm8b4jwPSkbPcsS9rmijYKA%40mail.gmail.com.

[Jason Edgecombe]

Yes, it looks like it.

Here is an Xorg snippet for display :1 from one of my RHEL7 servers with an NVvidia Quadro M2000 card.:

Section "Screen"
    Identifier     "Screen0"
    Device         "Device0"
    Monitor        "Monitor0"
    DefaultDepth    24
    Option         "AllowEmptyInitialConfiguration" "True"
    Option         "UseDisplayDevice" "none"
    SubSection     "Display"
        Virtual     1920 1200
        Depth       24
    EndSubSection
EndSection

The above server has a Matrox graphics card for the console and the M2000 handles VirtualGL.

FYI, There is a quirk where Xorg still blanks the console when the Xorg on :1 is active. I haven't figured out how to fix that.

Jason

---------------------------------------------------------------------------
Jason Edgecombe | Linux Administrator
UNC Charlotte | Office of OneIT
9201 University City Blvd. | Charlotte, NC 28223-0001
Phone: 704-687-1943
jwed...@uncc.edu | http://engr.uncc.edu |  Facebook
---------------------------------------------------------------------------
If you are not the intended recipient of this transmission or a person responsible for delivering it to the intended recipient, any disclosure, copying, distribution, or other use of any of the information in this transmission is strictly prohibited. If you have received this transmission in error, please notify me immediately by reply e-mail or by telephone at
704-687-1943.  Thank you.

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/df2bf3fe-3b43-f55e-c629-e01faeb554c2%40virtualgl.org.

[DRC]

Well, then I don't understand what action needs to be taken.  https://virtualgl.org/Documentation/HeadlessNV already says "Leave out --use-display-device=None if the GPU is headless, i.e. if it has no display outputs."  That seems like a correct statement, given what you're saying.

DRC

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/CAAR6MGBS9%3DDzBv6mC-q0Ct%2BvdWM-1k9_3LfEwZUnjEFJCn0tyw%40mail.gmail.com.

[Quincy Wofford]

Yes, I think you're right that the documentation is correct. 

My issue is a special case. I have one xorg config file that I want to use for many devices and GPUs. If I can avoid special case configurations, I want to. For me, it makes sense to remove this option and monitor user feedback from problems. Some quick tests indicate that, with Nvidia driver v455.32.00, this option is not required for any of the systems where the universal xorg config is deployed.

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/d33f1d01-209e-4511-b933-8ad6dc160c4f%40virtualgl.org.

[DRC]

I understand that 'UseDisplayDevice "None"' isn't required for headless GPUs, but what you're saying below implies that that directive isn't required for non-headless GPUs either.  That is the source of my confusion.

DRC

To view this discussion on the web visit https://groups.google.com/d/msgid/virtualgl-users/CA%2BD-2wnSq22GoSha%2B%2BWTLcyU8A98tevJjgRzE2LjCYu8xkGNTQ%40mail.gmail.com.