Double GPU Passthrough in Proxmox 8!? Play Baldur's Gate 3 and Minecraft On The Same Machine?

Get the most out of your hardware, why not create a gaming VM, or use a GPU to accelerate tasks such as video rendering and Plex transcoding!?
In this video I show you how to pass a GPU through to virtual machine in Proxmox 8, including the iGPU on an Intel CPU. The same applies to any PCIe device or USB that you want to passthrough.
Instructions: github.com/Jam...
Proxmox PCI Passthrough: pve.proxmox.co...
Find me on:
Discord: / discord
Twitter: / jimsgarage_
Reddit: / jims-garage
GitHub: github.com/Jam...
Matrix: #jims-garage:matrix.jimsgarage.co.uk
00:00 - Introduction to Passthrough in Proxmox
02:00 - Hardware Overview
03:46 - Passthrough 3 Steps
09:10 - Configuring Proxmox for Passthrough
20:30 - Create a VM
23:26 - Check the GPU Is Recognised
24:05 - Outro

[@muromets4929]

Thanks to your instructions, I was finally able to pass through a GPU to VM! I must have watched dozens of videos from other KZbinrs, read numerous ProxMox forum posts, and somehow could not make it work until your video. Really appreciate your style of communicating the information, and specifically describing what each parameter means. Thank you! Just for reference, here the system I was able to make this work on (also somewhat dated): Dell Precision T7610 Workstation Dual Intel Xeon CPU E5-2697 v2 (12C/24T) @ 2.70GHz 128GB (8x16GB) DDR3-1600MHz PC3-12800 ECC RDIMM GPU: NVIDIA RTX 3070

[@Jims-Garage]

That's awesome 😎, good job and thanks for the feedback.

[@vivekbadra]

Jesus. I wasn't expecting the sun glasses line and I spit my coffee all over the keyboard. Nice one Jim . New to your channel. Subscribed 👍

[@Jims-Garage]

Haha, sorry about that 😎 thanks!

[@ReefAndChill]

Thank you so much, watched many videos on this subject and could not get my integrated GPU to passthrough to my windows VM. Great detailed video, I am now a subscriber. Cheers!

[@ReefAndChill]

But now, I get a blue screen failure when I reboot. I did get it to boot the first time with no error 43 in device manager. I even tested this on a fresh install of windows 11. My integrated device is in the group 0. I tried all different configurations with rom and pcie set in advanced. I know your video said something about group 0, I am wondering if this is causing my error.

[@helioradial]

Another good one. Keep 'em going. As usual highly appreciated.

[@Jims-Garage]

More to come!

[@dr.sajjadmahmood86]

Thanks brother. Watched this vide three times. and was able to passthrough my rtx 3090 without a single glitch. Your crystal clear instructions really helped.

[@Jims-Garage]

That's great, thanks.

[@kenshi1985br]

I'm doing this for some time using Xorg multiseat, but I'm going next step doing it on proxmox. Also, I need some VM's for working, so I will definitively do it! Nice vid!

[@Jims-Garage]

Thanks, let me know how you get on with it.

[@JunYamog]

Thanks for this guide. I was able to pass through 4 Nvidia 1070 to the guest os. Very helpful guide.

[@Jims-Garage]

4! Wow, nice one. What do you have planned?

[@JunYamog]

@@Jims-Garage it was my friends old miner rig, got it for free. Got PyTorch running on it already, most ML models can’t use multi gpu on inference but I think I can split off data and do parallel on training to use all 4. Anyway better get away from the keyboard and back to the grill. Happy holidays.

[@Jims-Garage]

@@JunYamog that sounds awesome 😎 same to you!

[@wheeliebinladin1416]

I have watched the video twice now (not that it wasn't very clearly presented) I'm at exactly the same point as yourself with the IGPU passthrough. I have it in device manager (more often than not with the dreaded code 43) samish hardware as yourself) igpu at 00:00:02 only its an Intel UHD 630 I7-9700. Being one of the better walkthroughs i have seen i was hoping that you got somewhere with the workaround. Appreciate the time taken to make the video!

[@Jims-Garage]

Thanks. It did work in Linux but not Windows. You might have some luck with: github.com/strongtz/i915-sriov-dkms

[@wheeliebinladin1416]

@@Jims-Garage Same exact issue, I have Kali running first time every time, just doesn't sit very well with the other half being on the living room "smart" TV. I guess its time to throw in the towel and buy another GPU. I have been at it for weeks. I will have a look at the github link first, I don't think I have seen that particular one. Thanks again for the great content and taking the time to respond. Have a great day!

[@Jims-Garage]

@@wheeliebinladin1416 You're welcome, let me know if you have a breakthrough! I believe custom driver packages are needed but I try not to walk too far off piste as an update invariably breaks things...

[@samzick6301]

For audio, you could run a Scream server at the host level and use a virtual network adapter. Then from your VMs, stream your audio to the host machine, then output there.

[@Jims-Garage]

Good shout, I will check that out.

[@samzick6301]

@@Jims-Garage I run a pretty cool workstation setup: I run Proxmox on the host and 2 VMs. The host machine is a 2nd gen threadripper build with a workstation amd gpu. My favorite part of this setup is being able to break my linux workstation, and restore the daily backup. VM 1) My "workstation" * EndeavourOS * GTX 970 running 4 displays * pcie usb card * I use Moonlight client to connect to the Sunshine service running on VM2, via a virtual NIC bridge. VM2) Gaming * Windows 11 * RTX 3060 with hdmi dummy plug * LSI SAS card with sata fanout cables to 8 ssd. * I use Stablebit Drivepool to store all my games on the 8 drives. On the other side of the room is my 42u rack, but that would be a rant in itself.

[@PicSta]

A couple of years ago, when I first time had contact with virtualization, I got fooled by the same thinking. Your CPU has 4 cores, and you have used 2 cores already, but you would need to set another 4 cores, 1 socket for a Windows machine. Then I have learned that the VMs just get access to use the cores and of course if there is no load why shouldn't they use all cores or those who need to run fluidly. Btw, a great video and well explained. Subscribed & liked.

[@Jims-Garage]

Thanks for the feedback. For Homelabs you can usually over provision your VMs without impact, it's a useful feature.

[@5speedfatty]

@14:30 I have a valid reason to pass an entire USB hub through but its kind of specific. suppose you want to run a file server / nas and all your drives are USB, also suppose you have dedicated pcie usb3 cards for them to talk to , and those are all thats on it. in that case passing through the host card as a whole should in theory mean better performance for the usb drives in the VM, but youd need to be carefull passing through an onboard one. its too easy to cut off unintended devices from the host that way.

[@Jims-Garage]

That does sound sensible. Thanks 👍

[@Ecker00]

21:43 This is not how CPU Cores work in Proxmox, you can over provision this and the CPU scheduler will handle it fine balancing more power to the one that has more provisioned of the available resources. So go ahead and unless your CPU base load is spiking already, you can put number of cores equal to the number of threads your physical CPU has. Proxmox it self requires very little CPU resources, and will prioritize itself when needed.

[@Jims-Garage]

Correct, you can do this every hypervisor I've used but you want to avoid it if you can. Competing pressure on vCPU cores can lead to instability.

[@wusaby-ush]

great video, pleas do one to passthrough GPU to containers

[@Jims-Garage]

I already have done for docker and kubernetes. Check them out, I use Jellyfin as an example.

[@JesseStrangfeld]

How do you connect to your VMs for gaming? Is it RDP or another remote access? For example, what are the specs or minimum requirements and software of the client workstation that you use daily?

[@Jims-Garage]

@@JesseStrangfeld treat like a normal pc. Add HDMI, keyboard and mouse.

[@ivanmaglica264]

Proxmox can yank the PCIe devices from underneath the running host to the VM. I once mistakenly passed though the wrong device, HBA instead of Video card, and ZFS panicked 😀

[@Jims-Garage]

Haha, yes, that'll do it! 😜

[@rudypieplenbosch6752]

Very well explained, I'm thinking about buying an RTX A4500, but i want to share this card over multiple VMs, apparently the technology for this to be possible has been developed a long time ago SR-IOV, but the manufacturers are blocking this, of course. I see Nvidia developed some proprietary stuff for their Quadro gpu's vDWS, virtual work stations, but it only seems to run on VME or Citrix, I'm using Xen orchestra myself. I used Proxmox, but experience several severe crashes when i had to reinstall the whole thing. Anyway sharing GPU's among VMs should have been here yesterday, would be interesting if you know more about it and can share it in a video.

[@Jims-Garage]

Thanks for the feedback. You are right with your explanation as far as I am aware, locked behind paywalls (quadro required I believe). Unfortunately I don't have a card to test, but perhaps one day someone would be kind enough to donate or sponsor. It's something I'd love to explore. What are you wanting to use GPU acceleration for? If it's Docker containers you could split between those (suspect it's VMs though).

[@rudypieplenbosch6752]

@@Jims-Garage I need GPU acceleration for Blender, I also have a bunch of security cams on my network with dvr agent, it could use some gpu support, for recognition you can use the tensor cores for processing images. General GPU support in each VM is also welcome, since from a graphical perspective some vms are pretty slow. I have jellyfin now running on my main PC since i don't have a good graphical card in my server to do the decoding. What i would want, but of course this is not possible yet, would be a very flexible assignment of GPU power, when i render this power is moved towards a vm where blender is running, when i don't render, move that power back to a lower priority but usefull assignment. At the moment these gpu assignments are all fixed in configurations like you showed, this can done much more flexible upon demand, but this is the future i hope, sr-iov certainly would support these kind of use-cases. Anyway there are plenty usecases for flexible graphical horsepower nowadays and this will only grow, its just that the software is years behind it seems, i think a major cause are these manufacturers that want us to buy a separate gpu for each VM 😃, no way.

[@Jims-Garage]

@@rudypieplenbosch6752 absolutely, I hate software locks (but I doubt it'll change in the near future...)

[@rudypieplenbosch6752]

@@Jims-Garage Yes they will always keep trying, but with the emergence of hypervisors in the home environment the demand for more flexibility will grow, at some point they can't stop it anymore i hope. The hypervisor suppliers themselves could make a start, by providing a mechanism to make those settings you showed us, automatically. For instance, each time you access the desktop of a vm, the graphic card gets passed through, to that activated desktop. This should not be that difficult and would already help a lot and would be a good start for some more advanced forms of sharing of graphical power.

[@mattiavadala7870]

Thanks for this video! Just a question: at 17:25 the group of HD Graphics shouldn't be the 00:02? I probably didn't understand how group nomenclatures work, thanks!

[@antoninoromeo4015]

Me too. 😊

[@DrMarcArnoldBach]

I am running a standard Debian as a kvm QEMU server at home. The big thing behind it is that I can dispatch my hardware resources like cpu and memory between guests VMs and in case of CPU cores sharing between VMs as well. Passthrough of a GPU is the exact opposite. A limited resource is assigned to one unique VM… I do not see the point . Especially as my server would need to run that GPU 24x7 Not your fault but a great achievement would be splitting a GPU and its memory into N vGPU

[@alvarobinimelis2800]

I got error 43 on windows device manager with my uhd 630. do you know whats could be wrong?

[@Jims-Garage]

Be sure to check blacklisting the drivers and device, also check iommu group. Sometimes the hardware simply isn't compatible.

[@krdesigns]

is it possible to send iGPU to 1 windows and the GPU to another windows? that is something I'm interested in doing. Unsure if this possible

[@Jims-Garage]

Yes, that's possible. Depends on iommu groups but should be ok

[@Raima888s]

The thing that's stopping me is if the hardware being pass through plays up, it will require a hard reboot to reset the component. This means it will take out every VM running on proxmox.

[@Jims-Garage]

Potentially, it depends if you have multiple nodes and failover. In my experience I haven't witnessed hardware failing once booted. It basically boots and if it works it works.

[@93ct]

Hello! I have been watching your proxmox video and I must say you are very thorough and clear. I am following all your instructions. Thank you very much! I know this video is many months old but I hope you might respond. Can I check that at time-stamp 17:33 you type lspci -n -s 02:00. Is that supposed to be 00:02? And when you say grouping, it is the separate 2 sets of 2 digits right? So 00:02 is a different grouping to 00:06 for example? And you are not referring to different grouping based on the first set of 2 digits only right? Also you had a config setting of disable_vga=1 but before that you didn't have the ID for the iGPU in the immediate parameters before (ids=8086.....). Can you clarify? Thank you. I am using an i7-12700H and my id for that is also 8086:46a6. And finally the blacklist driver would also be i915? Many thanks!!

[@antoninoromeo4015]

Hi Jim i got vfio-pci 0000:00:02.0: not ready 1023ms after flr; waiting using igpu 630 intel and 10500 processor. It lock whole proxmox. Any advice? Many thanks.

[@Jims-Garage]

What are your iommu groups for the iGPU?

[@antoninoromeo4015]

@@Jims-Garage group 0. Thks

[@Jims-Garage]

@@antoninoromeo4015 if this is after enabling iommu etc then unfortunately your hardware won't work with passthrough, it needs a distinct group. You might have luck with forcing it with the kernel (ACS I believe).

[@phen0boy]

Really great video about how to passthrough in proxmox. Please can you make a video about demystifying SR-IOV with proxmox? I have Intel N100 and currently using virtgpu drivers for win and linux. They're ok but lack many features like hw video accelerators. Would be awesome to run both W11 and Linux with shared GPU.

[@gg-gn3re]

you can run many games, not just a simple two.. and you don't even need virtualization. Linux is built as a multi-user system (windows is a single user system) and you can connect multiple keyboards & mice to the same machine and play on separate monitors/users of the same machine.. again, no virtualization necessary.

[@Jims-Garage]

Thanks, I'm planning to cover this further down the road.

[@zira_fluff4105]

Thanks so much for this presentation on GPU pass through. I have a threadripper board where I want to run 3 or 4 individual GPU cards. They are all AMD RX5600X. I want leave one non-blacklisted for the host ProxMox. At 19:14 of the video, the blacklist.conf is generated. I don't want to disable the drivers for the lone GPU for ProxMox. Is it an all or nothing situation - i.e. must I disable AMD drivers completely or is there a way to generate the blacklist.conf and selectively disable AMD drivers for selected discrete GPU's ? Again thanks.

[@Jims-Garage]

I think you might be ok just by choosing the 3 device IDs to blacklist. Leave the one you want to use.

[@barrybay3696]

What do i do with error code 43?

[@Jims-Garage]

Double check you have blacklisted all the hardware and drivers, and have a separate iommu group. If the problem still persists you likely cannot passthrough on that hardware.

[@antoniodagata1265]

Very useful and well made video! If i understand it correctly it's not possible to use an igpu for both host and vm? I'd like to use proxmox with a container with media transcoding and a win10 vm for small gaming with my ryzen 7 5700u laptop, what do you think? Certainly would be faster installing only windows 10 with docker media container (plan B) but i'd like to try this hypervisor!

[@Jims-Garage]

With consumer equipment you cannot share hardware, you'd need 1 dGPU/iGPU per VM.

[@greenprotag]

I love these set ups. I built out a Dual x99 rig with 5x GPUs for 5 virtual arcades remote accessible through Parsec. Does everyone use a Rosewill custom 4U? I have a feeling that nearly any homemade "custom 4U" is almost always a rosewill.

[@Jims-Garage]

That's such a cool setup!

[@cxklann]

@Jims-Garage - thanks for the great content, thorough explanations - in the same class as @TechnoTim! At ~19:30, you point out that if the homelabber has their HDMI cable plugged in and they're currently using the iGPU, there won't be any video displayed (makes sense)...I assume that means that if one is using a PiKVM for remote console access, there' won't be any video. If that's correct, is there a way to provide video to the PiKVM? Thanks again!

[@Jims-Garage]

Thank you, that's high praise, I love Tim's videos. If you're blocking the iGPU by not loading drivers I don't think it will work, you would not be able to access the bios etc.

[@techknowbabble]

That tittle had me thinking he actually got to different dedicated GPUs passthrough because passing 2 gpus is supposed to be harder than just one. Got to the end of the video with a the typical KZbinr: "technically if i wanted to". If you are reading this somehow before watching, don't bother the video will not give you what the tittle promised you. But honestly otherwise great video, too bad the title is a lie.

[@Jims-Garage]

Not entirely true. I'm limited by the hardware used, the process is accurate and I have this working on a different machine. It's completely potluck with your hardware.

[@eglinfo1]

try installing debian 12 UEFI and I think you might change to how to install virtualbox instead

[@marvinalone]

So the monitor should connect to the discreet graphic card directly?

[@Jims-Garage]

Correct

[@NJun21]

Sorry one question Is that able to run game that has EAC (Easy anti-cheat) ? Like new world and rust

[@Jims-Garage]

No, sadly not.

[@rrodriguesv2]

hello jim thanks for the video, one question if possible, does the cpus that you passthru to a windows vm machine, when running a benchmark such as cinebench r23 or cpu-z do they all work with default cpu turbo boost? or the cpu stays on default speed (lower clocks) ? if possible show us on a video how to turn on turbo boost for ryzen and intel cpus (inside a windows vm)

[@Jims-Garage]

A good question, I'll investigate and report back. I suspect that the CPU turbos, and you would likely be able to see it on the host, but it probably isn't reflected inside the VM. So, you should receive a performance uplift, but you wouldn't be able to assert that against a visible clock speed change.

[@rrodriguesv2]

yes the biggest issue that we dont install some softwares in a vm is the lack of power from the cpus not entering in turbo boost mode@@Jims-Garage

[@Jims-Garage]

@@rrodriguesv2 I've just tested a workload and the CPU was boosting in the hypervisor. Failing that you could just pin the clocks to turbo boost frequencies in the BIOS.

[@SeregaTikhonov]

Is there an API? I would like to send messages from my smart home to Matrix.

[@therainstorm809]

Thanks for your nice video! But I meet a problem, Balur's gate3 crash every time it enter the main menu. Did anyone meet Balur's gate3 crash problem in VM? I tried KVM win11, it crashed too. My profile: pve8, win10, amd 5800x(cpu select host mode) + GTX 1070 pcie passthrogh.

[@Jims-Garage]

Thanks. Do you see the GPU in device manager? Is there an error 43?

[@cort]

Great video, thanks. If I want to install Linux like Mint/Ubuntu/Debian with GPU Passthrough do I need to add TPM, change machine to q35 and BIOS to OVMF(UEFI). Meaning do I need all of them or can skip some for example TPM?

[@Jims-Garage]

Thanks, I think you need to do the same process. Test and let me know.

[@cort]

@@Jims-Garage I saw other people on YT doing GPU PassThrough without these options so the question arose. Hoped you could know but will have to test it myself.

[@Jims-Garage]

@@cort double check the official documentation as that is what I referred to.

[@evanmarshall9498]

Can you do the same video or point me in the right direction on how to do this with an LXC Container instead of a VM?

[@Jims-Garage]

Hi, I will likely do a video on this at a later time. This post seems to have it working: forum.proxmox.com/threads/gpu-passthrough-to-lxc-container.114106/#post-529758

[@evanmarshall9498]

@@Jims-Garage You're the best! Thank you

[@radoskrib7155]

Just spent almost two hours why this command "dmesg | grep -e DMAR -e IOMMU" was not showing iommu enabled. Its because on AMD system you should run this: journalctl -b 0 | grep -i iommu

[@Jims-Garage]

@@radoskrib7155 thanks for sharing, good to know

[@radoskrib7155]

@@Jims-Garage yeap, thats why i put it here. And thanks for this guide, probably the best I found on youtube.

[@Jims-Garage]

@@radoskrib7155 appreciate it

[@marlonnoto4116]

Hey Jim, thanks for your content. Could you possibly make a video using pihole as a local DNS server for traefik services? I'm sure there are folks that aren't comfortable exposing services to the internet.

[@Jims-Garage]

I already have a PiHole video up. Have you watched it?

[@sebasdt2103]

Well That's a bit anoying IOMMU Isnt enabling for my system. Even when set in bios to enabled it reverts back to Auto.. time to go another route

[@Jims-Garage]

Hmm. Have you checked the CMOS battery? Perhaps it's dead and the bios is resetting each time?

[@sebasdt2103]

@@Jims-Garage That's a good one that most people would forget. it doesn't explain why other options on the same page don't revert back. looking into it but it seems like IoMMU is detected. " [ 0.763522] pci 0000:00:00.2: AMD-Vi: IOMMU performance counters supported [ 0.764598] pci 0000:00:00.2: AMD-Vi: Found IOMMU cap 0x40 [ 0.764829] perf/amd_iommu: Detected AMD IOMMU #0 (2 banks, 4 counters/bank). " It doesn't explicitly say it is enabled..

[@nomercyriding]

​@@sebasdt2103 double check you are editing the correct boot file. For instance I followed a video from KZbin which had me edit I think a file called "commandline" and added the iommu and I got the same output as you posted. It wasn't until I did more digging that I also had to add the quiet/iommu options to another boot file (which depends on what your system uses to boot ((GRUB or otherwise )) that I got it to show iommu enabled (in proxmox). I don't know about your bios issue though.

[@lakshya238]

Is there a possibility to pass through an integrated GPU? Thank you in advance 😄

[@Jims-Garage]

Yes, follow the steps in the video. I was simply unfortunate with my hardware. The process is the same as illustrated.

[@lakshya238]

@@Jims-Garage thank you, I will give it a shot

[@Jims-Garage]

@@lakshya238 let me know how you get on

[@declanmcardle]

Basic question: I have no graphics cards. I only have the GPU insider the CPU. Does that work for passthrough? If so, does that mean the console or the spice window in proxmox would be really fast?

[@Jims-Garage]

In many cases you can passthrough the CPU integrated GPU (iGPU), many people do this. It is hardware dependent though and relies on IOMMU grouping. This will mean you lose video output for your machine and will have to use a Web UI or SSH.

[@declanmcardle]

@@Jims-Garage The iGPU is in 00:02 - 00:02.0 VGA compatible controller [0300]: Intel Corporation JasperLake [UHD Graphics] [8086:4e61] (rev 01) - did you say you have a workaround for that? GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt rd.driver.pre=vfio-pci pcie_acs_override=downstream,multifunction nofb nomodeset video=vesafb:off,efifb:off" (some extra options added from other blogs but I can always modify them) vfio.conf is options vfio-pci ids=8086:4e61 vfio-pci disable_vga=1 and blacklist.conf has blacklist i915 in it. Still no joy. 😞

[@declanmcardle]

@22:59 00:01 "I'll come back to this" it what you mentioned. Also @24:52

[@Jims-Garage]

@@declanmcardle that looks right from a cursory glance, but double check online. Sometimes unfortunately it's just not possible due to the way the mobo is layed out. Bit annoying!

[@Jims-Garage]

@@declanmcardleYes, planning to do a follow up video to see if I can fix it.

[@Catge]

Would passing through a GPU cause issues for an AMD based system where there are no built in graphics? I have read that it won't work but that a few years ago

[@Jims-Garage]

You mean a discrete GPU in a PCIe slot? That should be fine.

[@Catge]

@@Jims-Garage Yeah a 3700x cpu and quadro p1000 gpu. I don't have all the hardware yet to test but I will give it a try. Good info

[@Jims-Garage]

@@Catge that should work perfectly, especially with a workstation GPU.

[@Catge]

cheers mate, I will sub for more content@@Jims-Garage

[@Jims-Garage]

@@Catge Great, much appreciated

[@xSniper_LoL]

i follow the guide letter by letter... 13 gen 1360p igpu does not want to pt.... any suggestions? or can you make a guide for it? Thank you.

[@Jims-Garage]

It's hit and miss depending on your hardware unfortunately. Largely depends on iommu groups. Have you checked your groups?

[@xSniper_LoL]

@@Jims-Garage group 0 for igpu. But is not showing me IOMMU enable… i tried everything.

[@Jims-Garage]

@@xSniper_LoL sounds like it isn't enabled properly. Double check BIOS is correct

[@xSniper_LoL]

@@Jims-Garage I did… is enabled… if you know intel nuc bios is very simple and easy to use. I dont know why is just not pt… if you want I can give you access to it and you can look, maybe you can do some video for channel I dont mind if you are up for it :)

[@Jims-Garage]

@@xSniper_LoL thanks, drop some screen grabs in Discord support forum. Likely diagnose it through numbers ha

[@BigFourHead]

should be a FYI that not every game will work under this, example would be PUBG, doesnt let me work cause of the anti-cheat system they have in place.

[@Jims-Garage]

Correct, some games have VM detection.

[@rayfushgo]

i thought overprovisioning CPU cores is not a problem, because Proxmox anyway manages the load

[@Jims-Garage]

It's not a problem in the sense that it will work, it's just that the scheduler won't know what to prioritise when over provisioned which might have a downstream impact on your workloads. Likely not a Homelab problem, but something to bear in mind for production setups.

[@vikrantsingh3923]

thank you for the explanation. I followed this process but inside my vm i get an error when i try to run nvidia-smi "NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running." I'm doing this on a supermicro server that has 4xNvidia A40 GPUs.

[@Jims-Garage]

Let me take a look. I think for server GPUs (non consumer) you may need to go down the vgpu route. I'm surprised GPU passthrough doesn't work though.

[@vikrantsingh3923]

@@Jims-Garage i have spent literally months trying Proxmox to work and must have followed a dozen tutorials from Proxmox 7 to 8. Non worked for me so far.

[@aidanm7225]

Drag on much

[@Jims-Garage]

Thanks 😂

[@jeffreyschlieve590]

Great video, should I have more than the Meta-driver running? I noticed a long list on yours and I ONLY have that meta-driver. I ask because after I edited the moduals and then updated the initramfs I got a few errors. Running hook script 'zz-proxmox-boot'.. Re-executing '/etc/kernel/postinst.d/zz-proxmox-boot' in new private mount namespace.. No /etc/kernel/proxmox-boot-uuids found, skipping ESP sync. update-initramfs: Generating /boot/initrd.img-6.5.11-8-pve Running hook script 'zz-proxmox-boot'.. Re-executing '/etc/kernel/postinst.d/zz-proxmox-boot' in new private mount namespace.. No /etc/kernel/proxmox-boot-uuids found, skipping ESP sync.