Server Version#: 1.32.5.7349
Player Version#: Multiple
My Nvidia GPU (Tesla T4) has worked great for two years and now PMS isn’t using it. Can’t attach log file because it says it’s too large…
So I found out externally that this is a well known issue. Thanks for the help 
Hi,
I have the same issue 
Plex Media Server Version 1.32.5.7349 (64-bit).
Last known working version: 1.31.1.6733
OS: Windows Server 2016
GPU: Nvidia Tesla A2, driver version 537.13.
I do not know what version it broke on. My Nvidia GPU is not listed in the transcoding device
menu:
I have tried
- Uninstalling PMS and reinstalling it.
- Upgraded to latest version of Nvidia drivers (537.13).
I hope it can be resolved soon 
Best regards,
Viktor
It’s not fixed in the new release either…
Sad to report, it is still not fixed 
I can’t get it to use either my Nvidia or my Intel 730. I removed the Nvidia thinking maybe having both was the issue but still…. Nothing. It sees the 730 in the dropdown menu but doesn’t use it.
No mention of fixing this issue in the latest update. I’m not going bother testing it.
@ViktorKarppinen Out of curiosity, what motherboard are you using?
Hi @kd6icz
I use an HPE Proliant DL380 Gen9 Server. So the motherboard is an HPE (Hewlett-Packard).
Best regards,
Viktor
@ViktorKarppinen I looked it up and can’t tell if that has any sort of OOB management controller (I.e. ASpeed 2400, 2500, 2600)?
I’m using a variety of Supermicro motherboards that all have IPMI and I’m wondering if the built in VGA controller is preventing PMS from seeing the Nvidia GPU?
Again, mine is working on an old version (7195) of PMS.
@ViktorKarppinen After further reading, it looks like HP has their own proprietary iLO? Same concept as the ASpeed chips. That iLO chip most likely has an integrated VGA controller that is the primary graphics for the server. This didn’t used to be a problem but now that PMS has added AMD support and the ability to “select” the GPU it has killed our Nvidia GPU.
Hi @kd6icz ,
Yes, that is correct. HP use ILO as its oobm-controller. Sins the Tesla-cards are only GPU-accelerators they are not possible to use as a normal graphics adapter sins it has no ports for external monitor (as you probably know). And it would not work with the oobm either.
I also believe that the feature to select the GPU is thing that breaks it. It did work perfectly without issues before the update of PMS.
Best regards,
Viktor
1 Like
Hi,
I did try the Nvidia GRID drivers, then I was able to use the Tesla as an hw transcoder device, but…
- I lost VGA output om my OOBM (iLO), only way to access my screen is via VNC.
- Performance is very poor comparing to before when it was working with the normal driver.
So, there was no point to continue using the GRID driver. I hope some kind of fix will be released. I only got about a month worth of very good hw transcoding from my new Tesla card worth around $3000 at the time.
Best regards,
Viktor
I’m working with @sa2000 on a private thread trying to pinpoint the issue. Maybe you could post some logs and system info that he could forward to have another example.
@ViktorKarppinen I’m using a X13 Supermicro motherboard with a Intel i7-13700T CPU and that’s not working either. So it’s across the board as far as GPU’s are concerned.
This motherboard also has an ASpeed 2600 BMC. So that’s the common thread between all of these systems. An OOB IPMI controller with VGA graphics for remote server administration.
It started when the hardware transcoding options were overhauled to support AMD GPU’s, headless Intel Windows systems, and the ability to select which GPU PMS uses for transcoding.
Hi @kd6icz,
Sure, tell me which logs you want and where to put them.
Also what test you want me to make before taking out the logs.
Kind regards,
Viktor.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.
