7 minutes ago, UnraidNewbie2211 said:

Is it possible to install nvcc with drivers & make it system-wide and accessible inside containers?
I noticed, that i can't enable HW accel. for ffmpeg inside homebridge, because it looks at nvcc command existence at build time: https://github.com/homebridge/ffmpeg-for-homebridge/blob/latest/build-ffmpeg

Not on the host, unRAID is not a general purpose server, it is a application server.

 

You should consider deploying a Docker container with a ffmpeg binary compiled with Nvidia support, if you then pass through the card to the container everything will work as expected in the container with Nvidia hardware acceleration.

 

The same goes for nvcc, I think someone already asked such a question here IIRC.

I think almost all Cuda containers from Nvidia support nvcc if you want to build a container, but you of course have to pass through your GPU at build time I think, I never played around with nvcc.

7 minutes ago, UnraidNewbie2211 said:

And the last question is why GPU panel is not appearing at the dashboard?

Do you mean GPU Statistics? This is a dedicated plugin, search for GPU Statistics in the CA App, install and configure it.

I'm still on 6.11.5 on one of my servers (on 535), and I need the 550 driver for a new application.
Do I need to upgrade the Unraid OS for this driver? or is the plugin not showing available versions for some other reason?

Thanks

1 hour ago, tjb_altf4 said:

Do I need to upgrade the Unraid OS for this driver?

Yes.

 

I don't compile new drivers for such old Unraid versions any more (if they would even compile against such an old Kernel version).

Hey @ich777, thanks for this plugin - it has worked great for me for some time. I've started to run into some issues lately, and I'm unsure if it is related to your plugin or not.

 

Essentially, after my appdata backup completes on Monday mornings, I get an error that Plex could not start. I've narrowed it down to this plugin not being able to read what GPU is plugged in, and it only happens after appdata backup. When Plex tries to run with the Nvidia Visible Devices variable, it errors out so I need to restart the server to have the GPU be read by your plugin again.

 

I've attached my diagnostics, cheers!

ghost-diagnostics-20240916-0814.zip

1 hour ago, ctrlaltd1337 said:

it has worked great for me for some time.

There is your error:

Sep 12 03:15:14 ghost kernel: NVRM: GPU at PCI:0000:04:00: GPU-62807760-f89e-18e1-6f51-86c75eea568a
Sep 12 03:15:14 ghost kernel: NVRM: Xid (PCI:0000:04:00): 79, pid='<unknown>', name=<unknown>, GPU has fallen off the bus.
Sep 12 03:15:14 ghost kernel: NVRM: GPU 0000:04:00.0: GPU has fallen off the bus.
Sep 12 03:15:14 ghost kernel: NVRM: A GPU crash dump has been created. If possible, please run
Sep 12 03:15:14 ghost kernel: NVRM: nvidia-bug-report.sh as root to collect this data before
Sep 12 03:15:14 ghost kernel: NVRM: the NVIDIA kernel module is unloaded.

 

You have a Xid 79 error which you can read here what it is.

 

However Xid 79 can be pretty much anything but what it was most of the time is, at least what was reported in this thread, a bad power supply, a bad riser cable, setting the bus speed from the PCI slot to Gen3 or a thermal issue (overheating).

 

I also see in your logs that the UPS had to jump in because of a power failure, please double check your power supply, even if it's a high power unit if it's old it might be time to change it...

 

Please make sure to enable Resizable Bar Support and Above 4G decoding in your BIOS since these settings are crucial.

Thanks! I'll look into this. The PSU is only a year and a half old, but I can do a few tests on it.

 

I can also look into "Resizable Bar Support and Above 4G decoding," but my motherboard is server-grade and over 10 years old, so I doubt there's anything there for me. I've been using this plugin and the LS "variation" before it since release, so it should be fine without.

 

I assume it errored out and Plex was able to continue with CPU transcoding, but after the backup when trying to restart the container, it becomes noticeable because the container errors out when trying to look for that visible device. I'll attempt to re-seat the GPU into the mobo, and re-plug the power.

 

Thanks again.

34 minutes ago, ctrlaltd1337 said:

I've been using this plugin and the LS "variation" before it since release, so it should be fine without.

So you where using the plugin for a relatively long time and now it stopped working?

You can also try to use an older driver and see if that fixes your issue but i doubt it...

 

38 minutes ago, ctrlaltd1337 said:

but my motherboard is server-grade and over 10 years old

Yes, I saw that, please keep in mind that hardware can fail and maybe something is wrong with the motherboard, but that's only assumptions. Troubleshooting for this kind of error is hard.

You can also try to switch the PCIe slots on your motherboard and see if that helps.

 

Maybe it is "just" the power supply or a thermal issue of some sort, I hope that nothing serious is broken.

Hello, I had a Tesla P4 for transcoding and an older Nvidia video card for an external monitor in my unraid server. I needed an extra PCIE slot so removed both cards and installed a GTX1050Ti. I can see the card listed in system devices,

IOMMU group 32:[10de:1c82] 03:00.0 VGA compatible controller: NVIDIA Corporation GP107 [GeForce GTX 1050 Ti] (rev a1)

[10de:0be9] 03:00.1 Audio device: NVIDIA Corporation GF106 High Definition Audio Controller (rev a1).

Bind selected VFIO is unselected

But the NVIDIA driver plugin gives the error NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. 

I've tried changing the driver and rebooting, but no luck. Is there something else I am missing?

 

 

 

49 minutes ago, Neitfall said:

Is there something else I am missing?

Yes, you forgot to post your Diagnostics so that I can go through them and see what's causing the issue.

1 hour ago, ich777 said:

Yes, you forgot to post your Diagnostics so that I can go through them and see what's causing the issue.

Sorry about that, I had hoped there was a simple step I had missed. Diagnostics attached. I have since put the P4 back in just so that my plex Docker is able to launch and plex is running for the moment. The 1050Ti is also installed and I do have CLI output to the monitor, but it still does not show up in the nvidia driver plugin.

bonaire-diagnostics-20240923-1526.zip

Edited September 23, 20241 yr by Neitfall
typos

7 hours ago, Neitfall said:

The 1050Ti is also installed and I do have CLI output to the monitor, but it still does not show up in the nvidia driver plugin.

May I ask where you have the card from?

It seems like this is a counterfeit card...

 

You can easily see that here:

03:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP107 [GeForce GTX 1050 Ti] [10de:1c82] (rev a1)
	Kernel modules: nvidia_drm, nvidia
03:00.1 Audio device [0403]: NVIDIA Corporation GF106 High Definition Audio Controller [10de:0be9] (rev a1)

 

It says that it's a GP107 at the first line but as you can see at the second line which is the Audi Controller it says it's a GF106 which is from a GeForce GT4xx/5xx series card manufactured in 2010, see here.

 

7 hours ago, Neitfall said:

P4

As far as I can see the Tesla P4 is working as expected correct?

10 hours ago, ich777 said:

May I ask where you have the card from?

It seems like this is a counterfeit card...

 

You can easily see that here:

03:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP107 [GeForce GTX 1050 Ti] [10de:1c82] (rev a1)
	Kernel modules: nvidia_drm, nvidia
03:00.1 Audio device [0403]: NVIDIA Corporation GF106 High Definition Audio Controller [10de:0be9] (rev a1)

 

It says that it's a GP107 at the first line but as you can see at the second line which is the Audi Controller it says it's a GF106 which is from a GeForce GT4xx/5xx series card manufactured in 2010, see here.

 

As far as I can see the Tesla P4 is working as expected correct?

I ordered it from Amazon. I'll order another card and try that. I can always use this one as a tester card. Will see what happens with a new card. 

 

Thanks for the help.

11 minutes ago, Neitfall said:

I ordered it from Amazon.

Oh no, I would report it immideately to Amazon that this is a counterfeit card, you are not the first one who is posting diagnostocs with such a card but usually they are sold on Ebay, Aliexpress or such sites.

 

Do you have a link from the card or can you post a picture how the card looks like?

3 minutes ago, ich777 said:

Oh no, I would report it immideately to Amazon that this is a counterfeit card, you are not the first one who is posting diagnostocs with such a card but usually they are sold on Ebay, Aliexpress or such sites.

 

Do you have a link from the card or can you post a picture how the card looks like?

13 minutes ago, Neitfall said:

This is most certainly a counterfeit card.

 

Maybe try to get a Nvidia T400, T600 or even T1000, a Nvidia T400 is enough for 3 to 4 simultaneous 4K transcodes depending on the bitrate.

 

 The T series Turing based and I would say more future proof than a 10xx series card, draws way less power and is well enough for transcoding.

I'm having an issue where my GPU is detected by the plugin shortly after booting the server, but then disappears after page reload and says "No Devices Were Found"

 

I have tested 2 GPUs a 1070 and a 1080ti.

 

The device shows up in lspci and in System devices, I have also verified the device is not bound at boot.

 

Tested different IOMMU groups (even though I'm fairly confident this would be for issues with VMs)

 

Tested CPU multi monitor ON/OFF in the UEFI.

 

Any suggestions for additional troubleshooting would be greatly appreciated.

5 minutes ago, GigaGrim said:

Any suggestions for additional troubleshooting would be greatly appreciated.

It would be super cool to have Diagnostics...

I'm on mobile, but I'll upload when I'm home from work.

23 minutes ago, GigaGrim said:

I'm on mobile, but I'll upload when I'm home from work.

No worries, just to let you know I'll go to sleep in the meantime. 😴

7 hours ago, ich777 said:

It would be super cool to have Diagnostics...

Attached, appreciate the reply and help.

Sleep tight don't let the bedbugs bite <3 lol

poseidon-server-diagnostics-20240924-2111.zip

2 hours ago, GigaGrim said:

Attached

may some points ahead

 

Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: GPU at PCI:0000:01:00: GPU-9ca54a51-4546-a724-ffa0-10f03b790518
Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: Xid (PCI:0000:01:00): 79, pid=4874, GPU has fallen off the bus.
Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: GPU 0000:01:00.0: GPU has fallen off the bus.

 

take a look if the Power supply is sufficient and its sitting properly in the socket

 

may slso 2 more software questions

 

1/ you are using the "old" 470 driver Version ? tried an actual driver ?

2/ pcie acs downstream enabled ? tried with default setting ? or why did you enable it ?

4 hours ago, GigaGrim said:

Attached

In addition to @alturismo's answer, this pretty much also applies to your error:

(except for the part with the UPS)

14 hours ago, alturismo said:

may some points ahead

 

Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: GPU at PCI:0000:01:00: GPU-9ca54a51-4546-a724-ffa0-10f03b790518
Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: Xid (PCI:0000:01:00): 79, pid=4874, GPU has fallen off the bus.
Sep 24 21:00:41 POSEIDON-SERVER kernel: NVRM: GPU 0000:01:00.0: GPU has fallen off the bus.

 

take a look if the Power supply is sufficient and its sitting properly in the socket

 

may slso 2 more software questions

 

1/ you are using the "old" 470 driver Version ? tried an actual driver ?

2/ pcie acs downstream enabled ? tried with default setting ? or why did you enable it ?

This was purely for troubleshooting, I attempted the New Feature branch as well, it just happened to be on the 470 driver for this diag.

 

I considered the power supply as the issue, it's a 650w Gold on an otherwise fairly low power system, but it's possible there's something wrong with it. It's been through a few systems at this point. I tried the other GPU cables I had for it and in both GPU outputs from the PSU.

 

As for PCIe ACS downstream, again just for troubleshooting. I don't have a real reason I enabled it other than grasping at straws for why the GPU would continue to disappear after booting.

 

I'll update the driver, disable PCIe ACS downstream, and pull a power supply out of another machine to see if that fixes my problem.

 

 

Thanks!

I initially installed the plugin and correctly modified the container settings, and the GPU could be used properly. However, after I rebooted the Unraid server, the containers using the GPU could not be started, with the following error: 'Error response from daemon: unknown or invalid runtime name: nvidia.' (also can be seen in the screenshots below).

 

 

I tried changing 'Enable Docker' from 'Yes' to 'No' and 'No' to 'Yes,' but it still didn't work. The plugin page shows as usual, and the nvidia-smi command returns results, but containers using the GPU still cannot be started.

 

 

 

Thank you for your support, please let me know if you need any further information!

1 hour ago, tonywang1201 said:

Thank you for your support, please let me know if you need any further information!

Please post your Diagnostics, it is really hard to troubleshoot without them.