[Plugin] Nvidia-Driver


ich777

Recommended Posts

14 minutes ago, Thx And Bye said:

Would it be feasible to get a checkbox to run nvidia-persistenced after the driver install on system startup to keep the driver loaded? 

I've already thought about that but that is a bit more complicated than a simple checkbox...

 

For example:

If someone has two GPUs installed and wants to use one in a VM and the other one for Docker and has not bound the card to VFIO this can cause some serious trouble, from not starting the VM all the way to crashing unRAID.

If yutostart from the VM is enabled and unRAID crashes right at the start that would be pretty bad.

 

There are so many things that can happen that's why I always recommend to enable nvidia-persistenced in the go file or with the User Scripts plugin.

 

I hope you understand that...

 

EDIT: I know nvidia-persistenced very well btw... 😉

  • Thanks 1
Link to comment
15 minutes ago, LTech said:

I would like to use a Tesla K80 with Unraid and with your Plugin. But after a short look in the compatible GPU list, I found out that this GPU is not compatible.

Exactly, because this is a Datacenter card.

 

To be honest I even don't know if Docker or their container toolkit is compatible with those cards.

 

What is the exact use case for thus card in your server?

 

19 minutes ago, LTech said:

Is there a possibility to change the current Driver with an Nvidia Data Center driver Manually?

Not easily because it's not as easy as changing a URL.

The drivers needs to be compiled for each individual unRAID version, then the container packages get added, then it is packed up and uploaded to Github.

Link to comment
8 minutes ago, LTech said:

Is there maybe a Tutorial how to compile the Nvidia drivers?

The driver is relatively easy to install but keep in mind every time you reboot unRAID the driver is uninstalled again and you have to do everything from beginning.

 

The part thats a bit more difficult is to get everything working for Docker.

 

If you want to I can look into this but not today because it's my sons birthday... 😁

 

Keep one thing in mind, I don't know if this is working what you are trying to do to use one card for Docker containers and the other in a VM.

  • Like 1
Link to comment

Then first, Happy Birthday to your son! 🎉🎊

It would be really nice if you could look in to it when you have some spare time. The K80 is a dual GPU, so it hopefully should work. Can't I use a script for User scripts to do that automaticity for me?

Thanks in advance

 

Link to comment
5 hours ago, LTech said:

Can't I use a script for User scripts to do that automaticity for me?

Yes and no, this is a bit more complicated than you think and I really don't know if it's worth the effort since the Tesla K80 isn't even capable of transcoding H265 because the NVENC chip on there doesn't even supports it.

The next thing is I don't think you even can use one from the two cards in a VM because it doesn't even have a display output and from what I know display drivers won't install properly in VMs, but I could be wrong about that...

 

I've now created a complete custom package for you but the installation is a bit different than usual, please write me a short PM and I will hook you up with instructions on how to get the driver working. Also this custom package will only work on 6.9.2 and I even don't know if this will work...

 

May I ask what system do you have? Do you have maybe a Intel system that is 7th gen or newer with a iGPU? The iGPU can handle up to 4x 4K streams without a hitch.

If not maybe selling the card and buy something like a Nvidia T400 for transcoding (you can get them for about 140,- brand new) and a second card for the VM.

 

This is only a idea, over here in middle Europe Tesla K80s sell for good money where you can buy two cards without paying extra.

Link to comment
15 hours ago, ich777 said:

The next thing is I don't think you even can use one from the two cards in a VM because it doesn't even have a display output and from what I know display drivers won't install properly in VMs, but I could be wrong about that...

You can use this card in a VM because I do. There is a little trick on how to do it, but nothing too complicated. Because it's a computing card, you need to put it in WDDM mode, but it works.image.thumb.png.47862b7d93a1735fabb3bab802de5b9f.pngimage.thumb.png.29e4c1aa51394e0c1a8f337bf01f78e2.png

15 hours ago, ich777 said:

I've now created a complete custom package for you but the installation is a bit different than usual, please write me a short PM and I will hook you up with instructions on how to get the driver working. Also this custom package will only work on 6.9.2 and I even don't know if this will work...

Thanks for your efforts, I will write you a short PM

15 hours ago, ich777 said:

May I ask what system do you have? Do you have maybe a Intel system that is 7th gen or newer with a iGPU? The iGPU can handle up to 4x 4K streams without a hitch.

Well, I use a dual Xeon E5-2630 on a Supermicro X9DR3-F Mainboard with 128 GB DDR3 Ram in a Supermicro SC846 Chassis (SAS Expander) and a Fujitsu d2607 as an HBA. All Server Hardware. I know that this constellation may not be the best for a File Server, but for around 400€ that was quite the deal.

15 hours ago, ich777 said:

This is only a idea, over here in middle Europe Tesla K80s sell for good money where you can buy two cards without paying extra.

I got the Tesla K80 for 115€ from eBay that was the only Reason I bought it. I am from Germany, but I thought writing in German inside this Forum post would be counterproductive.

  • Like 1
Link to comment
  • 2 weeks later...
1 hour ago, Arragon said:

Would it be okay to have only one GPU for VM Passthrough and Transcoding if both aren't running at the same time? 

you can try it, but for sure not recommended ... ;)

 

1st step, do not bind the card in vfio, install nvidia driver plugin, reboot, check if your card is still working in your vm in passthrough mode.

 

then go further, setup plex to use the card, test if this is working, and so on ...

 

just consider, it can end in a total system crash, freeze, ... (hard reboot), just to say you know what can happen ;)

 

and may before you didnt thought about it, no igpu available for transcoding ?

  • Like 2
Link to comment
44 minutes ago, alturismo said:

and may before you didnt thought about it, no igpu available for transcoding ?

I believe the AMD Ryzen 5 3600 has no integrated GPU and the board only has on PCIe x16 Slot (and 2 x1 Slots)

 

Edited by Arragon
Link to comment
5 minutes ago, Arragon said:

I believe the AMD Ryzen 5 3600 has no integrated GPU and the board only has on PCIe x16 Slot (and 2 x1 Slots)

Correct, no iGPU.

You could however buy a PCIe riser fro x1 to x16 so that you can use a cheap card for the display output, but no guarantee that this will work, depends also on the BIOS and if it supports that you can select which card you want to use for the primary display output.

Link to comment
On 2/11/2022 at 12:28 PM, ich777 said:

Yes, because Nvidia dropped the support for such old cards now.

Your card should be supported up to driver v470.103.01 (last stable "legacy" driver which supports these cards) which is actually available for 6.9.2.

 

Anyways, please note that your card can only transcode h264, not h265 (Source) that's why I recommended the T400 to you because this is actually a new released card that is affordable and supports h265 and is not that power hungry as the K600.

 

Thanks for this, I had an old HP Proliant ML310e G8 with a Quadro K2000 card which wasn't seen by Unraid when I upgraded to the latest Nvidia Driver. I even installed a Quadro K620 to see if that made any difference but I've since found that the ML310 doesn't support the K620 so I had to downgrade the driver back to the last stable Legacy and re-install the K2000.

 

It's only used to do some transcoding for Plex/Jellyfin so no real loss but thanks for confirming.

  • Like 1
Link to comment
On 3/6/2022 at 11:37 AM, CGRFC1872 said:

i am having the same error as before where msot of the webui doesnt work, i've attatched

If you run the server without the plugin it does work?

Can you SSH into the server? If yes what does the command 'nvidia-smi' give you if you can SSH into your server?

 

From what I can see in your syslog the driver crashes right after it is loaded.

Do you have Above 4G Decoding enabled in your BIOS?

Link to comment

I have an issue around NVidia GPU usage in idle mode. If I boot my Unraid Version: 6.10.0-rc2  server with my Nvidia A4000  in GUI mode with the NVidia driver installed natively (v510.54)  my GPU will idle at 9-10Watts. However  follign a Plex HW transcode it will only idle down to 18W. I use the GPU statistics dashboard and according to that Im in the same power state (P8) and the same clock and memory frequencies are being applied. Is this just a software glitch that I cant get around?

 

Link to comment
4 minutes ago, james godbold said:

Is this just a software glitch that I cant get around?

Do you have nvidia-persistenced enabled?

 

Are you sure that it actually uses more power or is this displayed on the dashboard, keep in mind this readings are often not that accurate.

 

If you are in the same power state (P8 - seems that you have nvidia-persistenced already enabled) it should idle like it does before the Plex transcode.

Link to comment

Thanks for responding!  Yes I have persistence enabled.  Whilst I cant objectively confirm that the card is using more power (i dont have a power meter) it is most definitely warmer / outputting warmer air. I should note that the 10 W state only exists if I log into the gui mode. Otherwise 18w is the lowest I can achieve (manually Appling the persistence syntax) 

Link to comment
3 hours ago, james godbold said:

I should note that the 10 W state only exists if I log into the gui mode. Otherwise 18w is the lowest I can achieve (manually Appling the persistence syntax)

Some cards need a little load on it so that it actually can even drop down to a lower power state, but if you've enabled nvidia-persistenced that is really weird.

Seems like a driver bug to me but can't tell for sure, what you can do is that you post on the Nvidia forums if someone already experienced such a behaviour.

Link to comment

Hello! 

 

I'm very new to Unraid so forgive me if there's something basic I'm missing here. I recently got everything set up and installed a Quadro P620 to use for transcoding with Plex. Everything was working fine! After I rebooted, however, I saw that the Plex Docker wouldn't launch any longer because of an invalid parameter call to load with nvidia support. Looking in the plugins tab, I saw that the Nvidia-Driver plugin was no longer listed as installed and under the Plugin File Install Errors, there was an entry with "/boot/config/plugins-error/nvidia-driver.plg." If I re-install the plugin, everything works fine until I reboot again. Any idea what might be going on? Attached diagnostics after my most recent reboot. Any help would be VERY much appreciated.

 

EDIT: Was just looking through the logs. It looks like it realizes that it already has the desired version downloaded and installs it. Then after the NIC initializes, it attempts to download the driver again and fails. Would this lead to the situation I'm in? Any idea why this might be happening?

 

 

 

tower-diagnostics-20220308-1349.zip

Edited by Zoidoid
Link to comment
54 minutes ago, Zoidoid said:

Was just looking through the logs. It looks like it realizes that it already has the desired version downloaded and installs it.

Where did you see that?


The plugin basically installs itself and then tries to download the driver package.

 

i really don‘t understand why that is happening on some systems because the driver is actually downloaded when you install the plugin for the first time.

 

58 minutes ago, Zoidoid said:

After I rebooted

Please try the following, delete the plugin from the failed tab within the CA App, reboot your server, install the plugin again from the CA App, try to reboot again and see if it is working.

 

Male also sure that you have enough space on your USB boot device.

 

Have you waited until the plugin finished downloading the driver package and waited for the Done button to appear?

 

Once the package is downloaded and no internet connection is available on boot the plugin will search for local available driver packages and install the appropriate package but that didn‘t happen in your case.

Link to comment
4 hours ago, ich777 said:

Please try the following, delete the plugin from the failed tab within the CA App, reboot your server, install the plugin again from the CA App, try to reboot again and see if it is working.

 

Thanks for this advice! Everything seems to be working as expected now. I didn't read the settings carefully enough originally. I didn't realize that selecting an option from the top few versions forced it to download rather than use the pre-downloaded version. Why the download failed is a little bit of a mystery to me -- I think it might be an order of operations thing where my network interface isn't ready by the time Unraid wanted to try to download the driver? Regardless, I'm happy for it just to use the perfectly-functional driver that's now on the usb stick. Appreciate the help!

  • Like 1
Link to comment
4 hours ago, Zoidoid said:

I didn't realize that selecting an option from the top few versions forced it to download rather than use the pre-downloaded version.

No, that's not how it works, it tries to fetch the latest version and if this fails it searches for locally predownloaded packages and if a package is found it will install the predownloaded one.

Even when your network is not available at installing the drivers and a predownlaoded file is found it will be installed, regardless of the selected version.

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.