15 minutes ago, unrateable said:

Hi guys,

I need a quick sanity check regarding the transcoding utilization of the iGPU 770 in my i7-12700. Jellyfin Docker with passed through dev/dri. The file is 4K HEVC Dolby Vision Profile 5. I am streaming to my mobile phone via WLAN. Jellyfin says transcoding with 89fps, to 42Mbps MP4 H264 AAC . GPU stats in the Dashboard show the picture below. 3D Render Load 98% what the heck? Shouldnt I be able to transcode a couple of those 4K streams without an issue ? I would have expected maybe 10-20-30% range.....

Any hint what could be the reason; are you own experiences similar ? What am I missing. When I switch off Tone Mapping the Load decreases substantially. Could that be it ?

We do know that theyre sleeping together these days but im certain an intel igpu does not use the nvidia plugin :D

1 hour ago, unrateable said:

I need a quick sanity check regarding the transcoding utilization of the iGPU 770 in my i7-12700. Jellyfin Docker with passed through dev/dri. The file is 4K HEVC Dolby Vision Profile 5. I am streaming to my mobile phone via WLAN. Jellyfin says transcoding with 89fps, to 42Mbps MP4 H264 AAC . GPU stats in the Dashboard show the picture below. 3D Render Load 98% what the heck? Shouldnt I be able to transcode a couple of those 4K streams without an issue ? I would have expected maybe 10-20-30% range.....

To answer your question, what's wrong about that?

I don't know to explain that properly, usually you transcode to a phone in something like 1080p since the quality loss should be acceptable even on a folding device.

If someone speaks about transcoding, at least most of the times, they transcode from 4K "main" h265 profile to 1080p h264.

In no offense but it's not very efficient, in terms of the bandwidth, to transcode 4K h265 to 4K h264 since the h264 was made for 1080p, sure it can transcode to 4K but it's not as efficient if you transcode to h265 but do keep in mind even if you transcode to h265 it won't be more powerfull since 4K is a lot to transcode to.

However I find that more than acceptable what you are getting on the iGPU to transcode in such crazy high bitrates and even if you leave it like that, you should be able to transcode two or even three 4K Dolby Vison streams at a time since most media is 25fps.

Again, if you transcode 4K to 4K it is a big hit on your GPU, I would recommend that you transcode to something lower or at least if you want to transcode to 4K in a reasonable bitrate for 4K.

IIRC Netflix streams 4K in 16Mpbs but I'm not sure if they are now allowing 25Mpbs too.

I hope I explained that somewhat understandable.

Just an FYI, I upgraded my 960 TI to a RTX 5050 recently and ran into various issues

Motherboard is a Z97M-Plus

• Needed to enable Above 4G to resolve unraid NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running

• Once above was enabled, ran into no device found

• Changed from latest to the Open Source Driver and now it's all working

Surprised that the open source driver was what let it all work, but hey, I'll take it lol

2 hours ago, Codex said:

Surprised that the open source driver was what let it all work, but hey, I'll take it lol

5000 Series and up only supports the Open Source driver version.

11 hours ago, ich777 said:

I hope I explained that somewhat understandable.

Thank you for your take, I will look into that again and try different setups to find out what is causing the high load, and how it would handle additional streams.

12 hours ago, Mainfrezzer said:

We do know that theyre sleeping together these days but im certain an intel igpu does not use the nvidia plugin :D

Hehe. Yeah sorry mate, absolutely right. Thanks for pointing out. I must have been sleeping as well. Bringing it up here in the wrong place. Still appreciate @ich777 take on the subject.

On 10/14/2025 at 9:46 PM, Xploit61 said:

Hey everyone,

I think i am experiencing the same issue as mentioned above and have updated BIOS, checked all settings, made sure that its Gen 3 etc. and am getting the following. Any help would be greatly appreciated.

Oct 14 04:44:46 kernel: NVRM: RmInitAdapter: osInitNvMapping failed, bailing out of RmInitAdapter Oct 14 04:44:46 kernel: NVRM: GPU 0000:0c:00.0: RmInitAdapter failed! (0x22:0x51:763)

Oct 14 04:44:46 kernel: NVRM: GPU 0000:0c:00.0: rm_init_adapter failed, device minor number 0 Oct 14 04:45:46 kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1361
Oct 14 04:45:46 kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from nvStatus @ message_queue_cpu.c:242
Oct 14 04:45:46 kernel: NVRM: kgspInitRpcInfrastructure: GspMsgQueueInit failed
Oct 14 04:45:46 kernel: NVRM: kgspConstructEngineIMPL: init RPC infrastructure failed

After spending a lot of time troubleshooting this issue, I can confirm that I experienced the exact same behavior:

• GPU (RTX 3060) disappearing randomly from the system

• RmInitAdapter failed and later NV_ERR_NO_MEMORY errors in the logs

• Completely random timing — sometimes hours, sometimes days

• No VM passthrough, only Docker workloads (Jellyfin, etc.)

• Tried different NVIDIA drivers (including the open-source module)

• Swapped PCIe slots

• Changed GPU

• Tested different PSU cables / rails

• BIOS updated

• PCIe power saving disabled

• Above 4G decoding / ReBAR checked

• Even tested forcing PCIe Gen3

Nothing solved it consistently.

The only thing that finally stopped the issue 100% was enabling NVIDIA persistence mode, so the driver never tries to re-initialize the GPU during idle:

nvidia-smi -pm 1

After enabling this, the system has been stable for 24+ hours with zero GPU errors.

Add this to /boot/config/go to make it persistent at boot

It looks like the problem is related to GPU re-initialization on idle with newer NVIDIA drivers
Keeping the GPU initialized avoids the crash completely.

So if anyone is seeing random GPU "disappear / re-init / Out of memory / RmInitAdapter failed" issues on Ampere cards, this workaround fixed it for me.

Edited October 30, 2025Oct 30 by thymon

2 hours ago, unrateable said:

Thank you for your take, I will look into that again and try different setups to find out what is causing the high load, and how it would handle additional streams.

Dolby Vision is causing the high load and because you are transcoding from 4K to 4K unnecessary high bit rate, as explained above you usually transcode from 4K to 1080p.

Most posts here, and in other threads, talk about transcoding from 4K to 1080p or at least 4K to 4K "main" profile with a reasonable bit rate.

43 minutes ago, thymon said:

The only thing that finally stopped the issue 100% was enabling NVIDIA persistence mode, so the driver never tries to re-initialize the GPU during idle:

That's not how you use persistence mode by today's standards, please use:

nvidia-persistenced

in your go file instead.

Someone refresh my memory - planning to upgrade Unraid to 7.2.0; aren’t their special steps I must take with Nvidia plugin in relation to Unraid Upgrade?

3 minutes ago, blaine07 said:

Someone refresh my memory - planning to upgrade Unraid to 7.2.0; aren’t their special steps I must take with Nvidia plugin in relation to Unraid Upgrade?

Nope, just trigger the upgrade and wait for the message in your Notifications (Bell Icon) says it's now okay to reboot.

I am having issues when trying to change to any driver version. The only information in the logs is

Oct 30 08:16:43 Tower emhttpd: cmd: /usr/local/emhttp/plugins/nvidia-driver/include/exec.sh update_version 575.64.03

Also I am unable to change the Driver update notification

Edited October 30, 2025Oct 30 by mr2toyou
spelling

1 hour ago, mr2toyou said:

I am having issues when trying to change to any driver version. The only information in the logs is

Oct 30 08:16:43 Tower emhttpd: cmd: /usr/local/emhttp/plugins/nvidia-driver/include/exec.sh update_version 575.64.03

Also I am unable to change the Driver update notification

We're facing the same issue

1 hour ago, mr2toyou said:

Also I am unable to change the Driver update notification

24 minutes ago, netconnect said:

We're facing the same issue

Did you already saw the recommended post on top:

@ich777 re-installed and that fixed the updater.. Waiting for a disk rebuild / reboot to verify version download. Thank you again

Trying to update to Unraid 7.2 but I keep getting this error.

When I try to download the driver from the plugin itself I get this error.

29 minutes ago, Nanuk_ said:

Trying to update to Unraid 7.2 but I keep getting this error.

When I try to download the driver from the plugin itself I get this error.

Looks like you can just uninstall the plugin, do the update and then just reinstall the plugin. I'll update here if I turn out to be wrong.

15 minutes ago, Nanuk_ said:

Looks like you can just uninstall the plugin, do the update and then just reinstall the plugin. I'll update here if I turn out to be wrong.

You can do that or do the following:

1. Uninstall the plugin

2. Reinstall the plugin

3. Execute this command from a Unraid terminal:
   

   wget -qO- https://raw.githubusercontent.com/ich777/unraid-plugin_update_helper/master/force_package_update | bash

4. Reboot

7 hours ago, ich777 said:

You can do that or do the following:

1. Uninstall the plugin

2. Reinstall the plugin

3. Execute this command from a Unraid terminal:
   

   wget -qO- https://raw.githubusercontent.com/ich777/unraid-plugin_update_helper/master/force_package_update | bash

4. Reboot

Thanks! I'm going to save that command!!

3 hours ago, Nanuk_ said:

Thanks! I'm going to save that command!!

You don't need to save that usually it will work OOB, I really don't know why for some people the plugin breaks, I assume something messes with the plugin files after a long uptime.

On 10/30/2025 at 12:30 PM, thymon said:

After spending a lot of time troubleshooting this issue, I can confirm that I experienced the exact same behavior:

• GPU (RTX 3060) disappearing randomly from the system

• RmInitAdapter failed and later NV_ERR_NO_MEMORY errors in the logs

• Completely random timing — sometimes hours, sometimes days

• No VM passthrough, only Docker workloads (Jellyfin, etc.)

• Tried different NVIDIA drivers (including the open-source module)

• Swapped PCIe slots

• Changed GPU

• Tested different PSU cables / rails

• BIOS updated

• PCIe power saving disabled

• Above 4G decoding / ReBAR checked

• Even tested forcing PCIe Gen3

Nothing solved it consistently.

The only thing that finally stopped the issue 100% was enabling NVIDIA persistence mode, so the driver never tries to re-initialize the GPU during idle:

nvidia-smi -pm 1

After enabling this, the system has been stable for 24+ hours with zero GPU errors.

Add this to /boot/config/go to make it persistent at boot

It looks like the problem is related to GPU re-initialization on idle with newer NVIDIA drivers
Keeping the GPU initialized avoids the crash completely.

So if anyone is seeing random GPU "disappear / re-init / Out of memory / RmInitAdapter failed" issues on Ampere cards, this workaround fixed it for me.

A massive thank you for this, I really appreciate it!

Interestingly I was going through the motions after @ich777 kindly provided last bit of info, but as there was new versions of Unraid to 7.2 sable I kept testing. After a few hours being up, it had lost the card info and gave me the usual no devices found. Entering your command "Enabled Legacy persistence mode for GPU 00000000:0C:00.0. All done." - straight away brought the card back online and showing in both Nvidia plugin and GPUStats

I will continue to monitor, but this already seems to be back on the right track.

Thank you again @thymon

37 minutes ago, Xploit61 said:

I will continue to monitor, but this already seems to be back on the right track.

Please use this instead:

nvidia-persistenced

1 hour ago, ich777 said:

Please use this instead:

nvidia-persistenced

Ok thank you, I will do that. Should I add to the Go file as well?

27 minutes ago, Xploit61 said:

Ok thank you, I will do that. Should I add to the Go file as well?

Just that one line.

What @thymon said is the old way of enabling persistenced.

So I noticed the other day that Immich is showing up in the GPU stats on the dashboard but I'm not running it. Only containers I have the GPU passed through to are Plex and WhisperAI. NVTOP show Whisper running and nothing else.

1 minute ago, shiftylilbastrd said:

So I noticed the other day that Immich is showing up in the GPU stats on the dashboard but I'm not running it. Only containers I have the GPU passed through to are Plex and WhisperAI. NVTOP show Whisper running and nothing else.

Sorry, but this is the wrong support thread since this is GPU Statistics related.
However I link @SimonF here, he will be most certainly be able to fix that.