Skitals

The Quadro P400 came today. Plugged it in and it just worked, I didn't need to tweak anything in my bios. UEFI booting unraid, factory settings for Above 4G Decoding, Resizable BAR, ASPM, C-States, PCIe gen, etc. I can finally stream the UHD discs that I ripped w/ makemkv to my phone without crippling my cpu Also working with comskip, so no cpu spikes after each dvr recording. Thanks for your assistance, and the plugin.

Still fell off the bus, just without the aer error. Really hoping it's just this card. I've been using it for a few years but I don't think ever with drivers in any os. Just been using it for basic vga output in unraid/gui and the odd vm here and there. Never had any issues with other pcie devices, and I ask quite a bit from this machine.

I think I have exhausted all options with the GT 710. CSM Support legacy boot unraid (uefi boot disabled). Forced Gen2 and Gen 3. Tried all of the above in a different pcie slot (cpu vs chipset lanes). GPU always falls off the bus. Only thing of note is a different error when pcie is set to gen2: Dec 17 17:50:10 Tower kernel: NVRM: GPU at PCI:0000:0f:00: GPU-2e461cb5-b925-eff5-da77-bc3ab28682b1 Dec 17 17:50:10 Tower kernel: NVRM: Xid (PCI:0000:0f:00): 79, pid=4756, GPU has fallen off the bus. Dec 17 17:50:10 Tower kernel: NVRM: GPU 0000:0f:00.0: GPU has fallen off the bus. Dec 17 17:50:10 Tower kernel: pcieport 0000:00:03.2: AER: Corrected error received: 0000:00:03.2 Dec 17 17:50:10 Tower kernel: pcieport 0000:00:03.2: PCIe Bus Error: severity=Corrected, type=Data Link Layer, (Transmitter ID) Dec 17 17:50:10 Tower kernel: pcieport 0000:00:03.2: device [1022:1483] error status/mask=00001000/00004000 Dec 17 17:50:10 Tower kernel: pcieport 0000:00:03.2: [12] Timeout Dec 17 17:50:10 Tower kernel: NVRM: GPU 0000:0f:00.0: RmInitAdapter failed! (0x25:0xf:1241) Dec 17 17:50:10 Tower kernel: NVRM: GPU 0000:0f:00.0: rm_init_adapter failed, device minor number 0 Dec 17 17:50:10 Tower kernel: NVRM: GPU 0000:0f:00.0: RmInitAdapter failed! (0x22:0x56:667) Dec 17 17:50:10 Tower kernel: NVRM: GPU 0000:0f:00.0: rm_init_adapter failed, device minor number 0

You are right, switching slots and forcing gen 3 could do the trick. Both things I will try, need be, when the new card comes in. I can probably swap the GT 710 and dell 310 without affecting my iommu/VM assignments, so I might try that in the meantime.

It is a pciex1 card in a pciex4 slot, so that seems unlikely. That slot gets its lanes from the chipset versus the cpu, so switching the card to a different (cpu) slot might make a difference. It took a lot of mucking around 2 years ago to dial in my settings and make gpu passthrough work with my radeon 5700xt (subject to amd reset bug). I suppose you are right it shouldn't make a difference to my VMs so I might pursue this. The motherboard only supports running in gen 3 or gen 4 natively, so it will run in backwards compat either way (card is gen 2). Afaik I can't change this on a per slot basis. I went ahead and ordered a quadro p400 for $50. It's pcie gen 3 and supported in the latest nvidia drivers so hopefully I have better luck with that. I will pick up troubleshooting when that comes in a week or so. Thanks for your thoughtful input!

Dec 15 18:34:58 Tower kernel: NVRM: GPU at PCI:0000:05:00: GPU-2e461cb5-b925-eff5-da77-bc3ab28682b1 Dec 15 18:34:58 Tower kernel: NVRM: Xid (PCI:0000:05:00): 79, pid=5103, GPU has fallen off the bus. Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: GPU has fallen off the bus. Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: RmInitAdapter failed! (0x25:0xf:1241) Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: rm_init_adapter failed, device minor number 0 Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: RmInitAdapter failed! (0x22:0x56:667) Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: rm_init_adapter failed, device minor number 0 Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: RmInitAdapter failed! (0x22:0x56:667) Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: rm_init_adapter failed, device minor number 0 Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: RmInitAdapter failed! (0x22:0x56:667) Dec 15 18:34:58 Tower kernel: NVRM: GPU 0000:05:00.0: rm_init_adapter failed, device minor number 0 Decided to install nvidia-driver to see just how bad my gt710 is at transcoding. It falls off the bus as soon as nvidia-smi is run. I can set this card as my primary output, boot unraid, run nvidia-smi (or open the plugin) and immediately my screen goes black. I am of course running driver version v470.141.03 for legacy support. I have tried everything suggested. I enabled Above 4G Decoding and Resizable BAR. I disabled ASPM. I disabled C-States. The only thing I didn't try was switching from UEFI to Legacy since I have no interest in that. I turned off docker and turned off my VMs for testing. Any tips would be appreciated. tower-diagnostics-20221215-1828.zip

No problems here. A tip for anyone is looking to get comskip working in this container. I for whatever reason couldn't get comskip to compile with statically linked libraries. I found the pre-compiled comskip from channelsdvr (channelsdvr_nvidia in community apps) works in this docker. Put it in your emby appdata so it is persistent. Edit: Wow, I overcomplicated that. Thanks for the universal-package-install docker mod, you can install comskip really easily by adding two variables to your emby container. Variable Key: DOCKER_MODS Value: linuxserver/mods:universal-package-install Variable Key: INSTALL_PACKAGES Value: comskip 🤯 This was the one feature I was missing transitioning from plex. Comskip + ComSkipper emby plugin fixed me up.

I had great success with linuxserver's docker. I'm using my personal gmail for my email. smtp.gmail.com port 25. Only gotcha is you can't use your normal password, you need to generate an app password. Use this link to generate one and you are good to go.

Getting PIA working is as simple as using this utility to generate a config file. It took me a few attempts trying different endpoints before finding one that worked (or perhaps there is some failure rate), but it is possible to create a standard wg config file with PIA.

Nasty patches? It's a few prebuilt binaries and prebuilt ovmf file. If someone can't edit their own xml file I wouldn't suggest they use a test verion of unraid.

You need to add the tpm to your xml and start the vm for the first time for the files to be created. If you get "Could not run '/usr/bin/swtpm_setup'" error starting the vm, for fix is in this post above (chmod to fix permissions):

The swtpm module is not installed. You need to download the .txz files, put them in /boot/extra, and reboot. The tzx files in /boot/extra get installed at boot time so the binaries aren't installed until reboot.

TYVM. Retail ISO is out, just updated my win10 vm to win11 final no problem after getting TPM emulation working per the guide. I'm on Unraid 6.9.1.

As an aside, 6.8.0 rc5 with this kernel has been rock solid. I'm at 84 days uptime as I type, I use this machine as my daily driver (win10 gaming vm w/ 5700xt) with 10+ docker containers running (plex server and more). I'm still holding off updating until there is a 6.9 release candidate.

What temp readings are you referring to? There are no issues with unraid and hot cpu temps on these boards.

Still waiting for a 6.9 Release Candidate before I update from 6.8 RC5.

They pulled the navi patch (v1, Limetech never bundled patch v2) because it didn't work or caused more issues with certain hardware configurations. The navi patch itself should do absolutely nothing if you don't have an applicable card, it simply hard codes a special behavior, or quirk, only for a matching pci device. There are tons of these quirks for different pci devices bundled into the linux kernel. Sorry, though, I haven't tested nvidia gpu passthough.

I am all UEFI, no legacy. It might come down to your GPU and which mode(s) it supports, if you are talking about an older card.

I replied in that thread, this kernel won't fix your issue.

Whats your issue, specifically? Does your MacOS VM work, just looking for a vega reset fix? I've had no luck getting my 5700 XT working (at all, not reset related) in Catalina despite it having native bare metal support. AMD Mac OS is it's own can of worms I haven't tackled (yet).

You need to actually read the first post in this thread. You need to make sure your unraid usb is on the 0b:00.3 USB controller, and the occulus stuff is NOT on that controller. Then you need to pass through 06:00.0, 06:00.1, and 06:00.3. This can all be done very easily with the vfio-pci config plugin. Do not attempt to do anything else until you understand what you are doing.

This will happen if you don't read and stub the usb controller being used by unraid. Put your unraid usb in another computer and delete the file config/vfio-pci.cfg and your computer will "work" again. I would recommend doing your due diligence and read a whole lot more before proceeding.

Onboard audio now works on 6.8.0-rc5 with this kernel: