September 30, 20241 yr Reviewing: https://www.phoronix.com/news/NVIDIA-Open-GPU-Virtualization we are finally getting open source drivers for vgpu support. any chance we can get updates are checks on this for splitting a high end gpu for multiple vms/dockers?
December 4, 20241 yr 2 hours ago, bmartino1 said: add a area to install 3rd party drivers removed from the kerneal For development purposes you can do that on your own machine: https://github.com/ich777/unraid_kernel However I would recommend that you contact me if you want to automate builds from drivers since I already have a toolchain that automatically builds the drivers and uploads them to GitHub for easy use in Unraid driver plugins. I'm here to help and do this for the community to make Unraid even better as it is now.
December 4, 20241 yr Author Thank you. I appreciate this. I honestly don't know where to begin. I would love too, but I have to be careful with our nvidia licensing agreements and file access/sharing. While I currently have the access via my current job with the university. I can grab some files here and there... But some of this revolves around the VGPU unlock with a modified licensed downloaded run file and with use of other 3rd party applications that may need to bin installed.. example (rust unlock... mdevctl, dkms...) In truth, it would be nice to be able to add a custom nvidia run file to your current nvidia driver plugin. as a lot is already in there. As I would prob want to make something similar for VGPU. as the vgpu driver is not public but kinda shareable.(I can often find the vgpu driver and guest divers in china blogs with proxmox.) So currently Re-reading out licenses and asking what can and can't be shared... From what I'm finding, its more the guest drivers that use the host modified run file that get gray with nvidia licensing. Nvidia done't really acknowledge that a 2080 is xyz device and cant get a clear when using a unlocked run file to use the 2080 as a enterprise grid card... But My end goal is: That and for Unraid I think a unlocked driver to use example NVIDIA 2080 g card as split into # of xyz virtual g cards for dedicated memory towards dockers or vms for things like parsec and 3d acceleration in the vm. But there some weirdness and in the end I think this would be a niche case and not easily shared in the community... more on editing the showing mdevctl profiles after getting the driver running in unraid. Then going for and edits the xml to use the created mdev profile in a vm for implementation. this really is a good guide. and what I would try to imitate and implement within unraid/the plugin... https://gitlab.com/polloloco/vgpu-proxmox and what I would follow in terms of make a file somewhere and copying at boot /etc/vgpu_unlock/profile_override.toml for Vgpu Overrides I'm not sure if rust compiler is needed after the run files are patched...or driver is installed (Based on systemd servces and ther onversion to slackware) After that, install the rust compiler curl https://sh.rustup.rs -sSf | sh -s -- -y --profile minimal Nor if i want to try and support that on unraid. I'm fairly certain the plugin can run rc.d service instead of the systemd to load the vgpu_unlock-rs library when starting the nvidia vgpu services As you can see its a lot and not really sure what I'm doing when I forked the older vgpu driver plugin nor where and how far they got. That and I have not had time to devoted these into unraid... Id make a slackware baremetal machine first and run the stuff there as a test... (that way I atleat know I was this version with these edits to make this work... with theses xyz libs and files...) But I didn't have time and don't have the spare hardware atm... Then Health issues got worse(multiple kidney stone in both kidney a ongoing health issues since 2018...), so its was placed on the back burner. Maybe with time I can reach out and tinker with it in the next year.
December 5, 20241 yr 8 hours ago, bmartino1 said: I would love too, but I have to be careful with our nvidia licensing agreements and file access/sharing. Oh sorry, I thought that nouveau has this built in and nothing from the Nvidia side was needed. 8 hours ago, bmartino1 said: But some of this revolves around the VGPU unlock with a modified licensed downloaded run file Sorry but them I'm out. 8 hours ago, bmartino1 said: In truth, it would be nice to be able to add a custom nvidia run file to your current nvidia driver plugin. That won't happen if it's for that use case... 8 hours ago, bmartino1 said: Then Health issues got worse Sorry to hear that, hope everything turns good.
December 5, 20241 yr Author Originally posted when not much info was out and about on it... The community behind the open source driver is a bit weird and things are up in the air on it. The Nouveau driver did get some native support, yes, but there is alot of other groundwork things that need ironed out. More NVIDIA added open source to their driver run install file... Regardless, atm from my understanding it would be there vgpu line up only. and open source nvidia cards code name Ada Lovelace moving forward... Example, a PNY NVIDIA RTX 6000 is a Ada Generation Graphics Card...thats a $8000 enterprise card...(Not including their license for grid or driver access...) With open source driver just now starting to touch it, it be a long time before the community also opens it up for unlock to use vgpu on consumer cards. (if they can with the MIT license and how nvidia is using their software stack via the run file.... Other news, I Finally got a response but more dead end. Nvidia doesn't recognize consumer card VGPU and will not support it... "The NVIDIA vGPU approach is enterprise-focused and allows for splitting a physical GPU into virtual GPUs that can then each be assigned to multiple concurrently running virtual machines." There is also previous stuff online where nvidia has fought similar stuff in the past. There is some speculation behind the scene on open source vgpu that Nvidia might do as there Grid licensing is big money.(Where they been fighting to put the genie back int the bottle) As the original consumer side unlock started on quadro cards when a bunch of people started looking into the driver linux run file and adding features from other run files as quadro passed some checks to do those feature sets... it would be a good 5 ish years if not more before that can be leveraged for the next generation consumer cards via the community... Nvidia info: https://developer.nvidia.com/blog/nvidia-transitions-fully-towards-open-source-gpu-kernel-modules/ From my understanding, this is more to help WSL (Windows sub layer) and gpup using windows/Microsoft hyperV and NVIDIA move was more for windows... more on gpup. https://forum.level1techs.com/t/gpgpu-grid-unlock-your-nvidia-card/170668 Edited December 5, 20241 yr by bmartino1 typo - data
January 29, 20251 yr Very interesting discussions. A bit of hope, but not anytime soon It would be so great to be able to use vGPUs in UNRAID. Not for several years if I understand correctly?
January 29, 20251 yr Author 58 minutes ago, thymon said: Very interesting discussions. A bit of hope, but not anytime soon It would be so great to be able to use vGPUs in UNRAID. Not for several years if I understand correctly? That is my understanding of it yes. For official vgpu support in unraid. as unriad is built on Slackware linux. Slackware linux would first need to have other 3rd party installs applications installed(unraid may not have them) and then Slackware would need the capability to run the nvidia run file to install NVIDIA vgpu. Official GPUS for VGPU: https://docs.nvidia.com/vgpu/gpus-supported-by-vgpu.html one could try: Stumbled upon https://github.com/stl88083365/unraid-nvidia-vgpu-driver instructions: https://www.lxg2016.com/55831.html where others may or may not have had success in some form. There last github driver: nvidia-550.90.07-6.6.68-Unraid-1.txz Which is Nvidia VGPU 15... and per the other lxg forum post / unraid data worked in a previous v6 instance.... So In theroy it should still run on a unraid v6.9 instance, but there is no official support / limetech offical means on this... with dockers already able to do vgpu like things, I'm not aware of limetech / slackware expectation to truly support it. (Official means need to be done first) before any attempts, do try a VGPU Unlocked run driver...) With both mentioned and the need for a NVIDIA grid license access, this may not be feasible for limetech. What I was trying to bring attention to as Nvidia allowed some open sourcne -ness was to be able to 3rd party (but a lets say a tesla m card form ebay And I run Unriad)... With ICH777 nvidia driver. I can install the card. But ICH777 is expecting us to gpu pass it not use it on the host... Thus a separate plugin is needed for VGPU use case and needs... Then trying to add the xml vm code for mdevctl to use the vgpu card after the driver... (then comes the NVIDIA license per VM...) At which case why would a company run dev support to add a more at cost that doesn't go to them... Maybe we can get luck with v7 but I doubt
November 28, 2025Nov 28 Author ... bump: https://forums.unraid.net/topic/195204-unraid-virtualized-nvidia-vgpu-guest-driver/?&do=getLastComment
November 28, 2025Nov 28 2 hours ago, bmartino1 said: ... bump: https://forums.unraid.net/topic/195204-unraid-virtualized-nvidia-vgpu-guest-driver/?&do=getLastCommentIf vgpu unlock is required for this to use then it wont be implemented. Also doesn't vgpu need a licence server and an official license?
November 29, 2025Nov 29 Author 3 hours ago, SimonF said:If vgpu unlock is required for this to use then it wont be implemented. Also doesn't vgpu need a licence server and an official license?Vgpu unlock is a proof of concept only...vgpu unlock is not required...And yes there is a licensing in the hyper v and per VM... Edited November 29, 2025Nov 29 by bmartino1 Spelling
November 29, 2025Nov 29 Author 3 hours ago, SimonF said:If vgpu unlock is required for this to use then it wont be implemented. Also doesn't vgpu need a licence server and an official license?What I really what is the kernel files to build against... If I run a slack Linux VM or use ich777 docker I have no way to transfer or translate the files on unraid... The issue is getting or keeping kernel build files. That what's stoped me from running the Nvidia run file...
November 29, 2025Nov 29 Author a tesla p4 a pascal card is a vgpu card that still works in NVIDIA grid v16 lts channels and lattest v6 kernels...Nvdia driver line 535...its would be nice to get cheap telsp4 and still be able to use them... @SimonF is an example of a vgpu card that DOESN'T REQUIRE UNLOCK!that with the v16 535 nvdia gridd driver installed can use and support vgpu..vgpu_gridd would then need a license file in /etc/nvida/client somthing..mdevctl would then add options for use the 3d acelator telsa p4 in brances for VMs.example debain linnux which would also need a grid license. but on a legal license sytems its is posble to get a vgpu instance working...ATM VGPU on unraid is posble in v6 and the github dev claims it stall works on v7 ...after major kenral changes and witht eh intrduciton of limetech adding opensoure novea drivers is when theses thigns broke...Where One can have unraid see a vgpu card bare metal and make vgpu instances for docker, VMs, etc on unraid... https://github.com/stl88083365/unraid-nvidia-vgpu-driverhttps://forums.unraid.net/topic/195204-unraid-virtualized-nvidia-vgpu-guest-driver/This was more during the series 4000 and 5000 cards where nvida tyring to close markets.... 1 eye open 1 eye closed... and with the AI boom the keep Unraid a bit mroe relvant and a option I thought it worth a mention...As this topic was more on teh open driver static which is curently implemented!but ther are 4 nvdia linux drives stuff.nouvea - the kerneal upstream nvdia driver and open source drivers.nvdia run file consumernvdia run file Griddnvida run file grid_guest driverthe issue lies in trying to implemnt and use gridd on unraid due to kernal access to build against... last offical legacy working instance was back on v7.0.1 which had the disk bug and was pulled from archive...And with the introduciton of nouvea movbeing forward this breaks teh nvidia run install options and workarounds...
December 2, 2025Dec 2 On 11/29/2025 at 9:43 PM, bmartino1 said:a tesla p4 a pascal card is a vgpu card that still works in NVIDIA grid v16 lts channels and lattest v6 kernels...Nvdia driver line 535...its would be nice to get cheap telsp4 and still be able to use them... @SimonF is an example of a vgpu card that DOESN'T REQUIRE UNLOCK!that with the v16 535 nvdia gridd driver installed can use and support vgpu..vgpu_gridd would then need a license file in /etc/nvida/client somthing..mdevctl would then add options for use the 3d acelator telsa p4 in brances for VMs.example debain linnux which would also need a grid license. but on a legal license sytems its is posble to get a vgpu instance working...ATM VGPU on unraid is posble in v6 and the github dev claims it stall works on v7 ...after major kenral changes and witht eh intrduciton of limetech adding opensoure novea drivers is when theses thigns broke...Where One can have unraid see a vgpu card bare metal and make vgpu instances for docker, VMs, etc on unraid... https://github.com/stl88083365/unraid-nvidia-vgpu-driverhttps://forums.unraid.net/topic/195204-unraid-virtualized-nvidia-vgpu-guest-driver/This was more during the series 4000 and 5000 cards where nvida tyring to close markets.... 1 eye open 1 eye closed... and with the AI boom the keep Unraid a bit mroe relvant and a option I thought it worth a mention...As this topic was more on teh open driver static which is curently implemented!but ther are 4 nvdia linux drives stuff.nouvea - the kerneal upstream nvdia driver and open source drivers.nvdia run file consumernvdia run file Griddnvida run file grid_guest driverthe issue lies in trying to implemnt and use gridd on unraid due to kernal access to build against... last offical legacy working instance was back on v7.0.1 which had the disk bug and was pulled from archive...And with the introduciton of nouvea movbeing forward this breaks teh nvidia run install options and workarounds...Nouveau is a kernel module that was added to support virgl. It is blacklisted by default. How does it break the vgpu driver?
December 2, 2025Dec 2 Author 20 minutes ago, SimonF said:Nouveau is a kernel module that was added to support virgl. It is blacklisted by default. How does it break the vgpu driver?the run drive detects is abort of nvdia driver insall. as nvdia vgpu requires the nvdia drive to run and at run time it will black list the nouveau driver.Nvidia done't really no what breaks where or atleast hasn't been upfront about it.
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.