Hey, nice tutorial. The speach and the demonstration are quite good! Everything worked as it should! For those, who are wondering, the GPU for guest will be all the time off, unless you add some scripts (also called hooks) for single gpu passthrough (that bounds/unbounds gpu drivers from vfio).
After I run the script, I receive the message that my Non-primary GPU is now vfio-pci. But after reboot lspci -k still says that my videocard still has "Kernel driver in use: Nvidia".
Ran thru this. Enabled SVM, IOMMU, in bios. the script ran without a hitch. but my secondary graphics is still show up as using amdgpu instead of vfio-pci. I have an rx580 that i use primary, and a 5600G that i'd like to pass the integrated graphics to the VM. is it possible?
thx for this scripts and tutorial, finally I was able to passtrought my 3060ti, sadly MacOS Catalina doesn't detect it D; I would like to try on the future on windows vm's... but im afraid gaming would not be entirely possible thanks to anti cheats systems
Does this still work currently? i tried running the script and it said that non primary GPUs were bound to vfio-pci, but when i reboot my nvidia card says that my drivers in use are "nouveau" still.
@@PavolElsig Is it possible to make two vm based on discrete GPU? Of course not in the same time but it's my goal to have a windows and a Linux guest. Is it possible?
@@MajidKamran Yes, it is possible. You just have to create another VM and add that GPU. A warning will pop up, but you can ignore it since you're not running them concurrently. If you have multiple GPUs available, you can even run multiple VMs at the same time: th-cam.com/video/rWuNAIHTIJA/w-d-xo.html
Having a bit of an issue, everything seems to be working. The script works without a hitch or any errors, lspci shows all of my RTX 2060 kernel drivers as vfio-pci, and when I start up the VM it does start running and it looks fine when I use spice to check it on my host desktop, but I'm getting absolutely no signal from the passthrough GPU. Nada. The GPU seems to be running fine, the fans are spinning and the PC can still see it. But there's just nothing being outputted. I'm using an i7-8700k and a RTX 2060 on a MSI Z370 GAMING PLUS motherboard EDIT : Fixed by remoting into the VM from host and installing GPU drivers.
@@JohnJigsaw420 Believe I fixed it by remoting in through a Display Spice first (the thing that lets you control the VM desktop from the host), and then installing the GPU drivers on the VM. :)
the QEMU/KVM manager is stuck at "Creating Virtual Machine" part. it says creating domain... and never moves forward from that point on. any help would be awesome!!
I have desktop with i5-4460 and nvidia 1050 ti. Pop Os has set nvidia as primary, can I use the Intel gpu just for virtmanager ? I can't seem to get to windows installation. After I do "press any key" then it goes to "Loading Files [][][][][][[]". I wont get to see the TianoCore logo. Its just black screen if you scroll top you would see "BdsDxe: starting Boot0001 "UEFI QEMU DVD-ROM QM00003 " .....
@@PavolElsig Yes I deleted all virtual gpu parts and added the intel gpu and audio pci. "Graphical console not configured for guest" is result of this. Actually, what I want to have is inside my host PopOS(using primarily nvidia gpu), is virt manager with windows 10 guest running on the intel hd graphics.(I haven't connected any monitor to the motherboard just the 1050 ti). I only need to use one monitor cuz that is all I have. Its just like if I am using virtualbox/vmware. I'm not trying to use the virtualization for another physical desktop. Just one to have intel graphics performance on vm. Am I misunderstanding gpu passthrough?
In a GPU passthrough setup you’d normally have your linux outputting though the Intel GPU. Your VM would get the secondary GPU passed through to it. Your VM would be outputting video through that GPU to a monitor connected to it. That message about graphical console not being connected is what you should get. There is no video being sent back to Linux.
Laptops are generally pretty difficult to do passthrough on. If you have 2 GPUs (Intel iGPU would be enough) you might be able to do it. I would recommend a laptop specific guide though.
i have a fresh install of pop os with nvidia drivers and only one hdmi to the gpu. 1050 ti with i7 8700. but after the reboot the audio driver is snd_hda_intel and vga is still nvidia
moose43h Is linux using the Nvidia card? You might need to use the Intel GPU as the boot GPU. That way linux would be outputting from the iGPU and the guest OS would use the Nvidia.
So i got the vm working and everything. But my host pop os won't boot when theirs a display plugged into the passed graphic card. I have to unplug the monitor. Boot to pop os. re plug the motor and then boot to windows. I followed your guide is, is there anyway to fix this?
Anyone tried running the script lately on Pop!_OS 22.04 LTS? After running it and rebooting the kernel driver in use is still 'nvidia' here. Edit: Seems the script adds an empty vfio-pci.ids"" line to the kernelstub which I can't seem to remove with kernelstub -d. It just moves around in the list when doing so. So maybe there is something reentring it right after I delete it? I've tried adding my own vfio-pci.ids with the appropiate GPU ids from the lspci -k list, but I don't know if the empty vfio-pci.ids entry cancels it out. Still can't set the kernel driver in use to vfio.
hi I struggle to have this working. I followed all the step and my secondary GPU still load the Nvidia driver instead of the vfio-pci I have VT-X enabled on my bios but I cannot enable VT-D (if I do so, my system doesn't boot anymore) Can my problem be related to VT-D being disabled? My motherbord is a ROGSTRIX Z370-E GAMING (I just update bios version from 1002 to 2401) My CPU is an Intel® Core™ i5-8600K Thanks in advance for your answer
@@PavolElsig Thanks for your answer. I think I am stuck with 2 computers in this case. Since no operating system (tried on Windows10 and Pop0S 20.10 ) want to boot when I activate VT-D in bios
The script is in plain text. Open it in a text editor to understand the changes that it makes. At the end it creates an uninstall.sh, to revert the changes.
If the 1060 is your primary GPU, it can't be passed through. Linux would need to use a different GPU for display output and then it should work. If you have an Intel iGPU, it should be enough. Otherwise you'd need to get another GPU for the host.
@@PavolElsig I have an intel iGPU that I plugged my HDMI monitor and my DVI-D monitor into it. It still says when I load the script that the boot GPU is my Nvidia GTX 1060. I have VT-d enabled in my bios so that wouldn't be an issue also. I manually put the ids of the NVIDIA GTX 1060 in the commandline and just caused my popos install to hang at startup.
@@PavolElsig For my motherboard (the Gigabyte B365M DS3H) the BIOS will automatically detect any intergrated graphics and will have an option in there. It has 3 options (Auto, Enabled, Disabled) I have it on enabled, I don't know where I would select a boot gpu in my BIOS.
When i use my intel UHD graphics i cant see my gpu anymore if i type lspci -k. I got a RTX 2060 Super and an Intel I5-8600k on an Asus Z390 TUF Gaming, i have no idea why this happens. I selected the CPU as the main graphics in the BIOS.
@@PavolElsig I found the problem, i have to turn it on via system76-power graphics power on. But now i have a different problem, when i try to boot the VM i get stuck at the Press any key to boot from CD or DVD. Whenever i press anything it stops and i have to force shut it down, any clue why this happens?
@@crylia8666 Make sure the keyboard is connected/routed to the VM. You could also try to install Windows without a GPU and add it later to a fully working VM and install the drivers.
You're saying in this video that secondary GPU should work with that script, could you please confirm that sentence because your previous Fedora video you were saying in the comment that it disabled any other GPU while guest VM was running
The primary/boot GPU always stays with the host. Secondary GPUs are bound to vfio-pci and can’t be used by the host. Only guest OSes can use those GPUs. Both the Fedora guide and the PopOS guide work like that.
Is the GPU and it’s audio portion controlled by vfio-pci? If not blacklisting nouveau might help. Is it in it’s own IOMMU group? If nothing else helps using a different pcie slot might help.
Could this be done with a singel GPU and intel iGPU where the iGPU is the host and the PCie GPU is the guest slave? using looking glass, 8700k + z390 motherboard?
I've got an rtx2060 max-q in my laptop and the win10 KVM only recognises it as a Microsoft VGA adapter, the latest NVIDIA driver doesn't recognize the card, how do I go about installing the driver? Thanks you in advance
Aiden Craig Allman Laptops are more difficult to do passthrough on. You might need some additional steps. Try following this guide medium.com/@Leduccc/simple-dgpu-passthrough-on-a-dell-precision-7450-ebe65b2e648e
Probably a stupid question, I'm using kernelstub --add-options "intel_iommu", do I need to add the "vfio-pci.ids=10de:10fa,10de:1fb9" too?, The VM seems to see the PCI devices
@@PavolElsig might have found the root cause of my issue, looks like the subsystem id is Changing as soon as KVM starts and takes control of the vfio driver, is there a way to hardcode these so I can install the NVIDIA driver?
If I have 2 monitors , how do you set them up? One monitor to each GPU?. Is possible to use spice display? Can you have a monitor setup so it connected to different gpu example display port to one gpu1 and HDMI to the other gpu2?
Each monitor is connected to one of the GPUs. It's totally up to you whether it's connected to a display port or HDMI. You could still use spice, but if that's what you want, check out Looking Glass because it's way better.
I don't believe so. What you are referring to is something called vGPU, which is virtualizing a GPU to be used by one or more VMs and the host. Think basically how virtualization works for other hardware like RAM and CPU resources. This is still not something unlocked in consumer GPUs. What this guide is showing is passthrough, which is excluding the GPU from use with the host system and making it dedicated to the VM. Hence why, if you read the script READ.me file on Github, it requires that you have two GPUs present in your system. This can be two dedicated graphics cards or one GPU and a CPU that has integrated graphics as well.
I'm having an issue when it comes to booting into windows for the first time. After I press any key to boot into windows, it just starts loading files and then keeps showing a black screen. I've tried downloading another copy of windows and hoping that works but its the same thing as well. Can someone help me out?
@@PavolElsig I have a NVidia 1060 6gb. The VM will not run without the default GPU which I find weird. the Nvidia parts are already read as vfio-pci but the VM doesn't want to behave properly
Make sure you delete all the parts associated with the virtualized GPU, spice and anything related. Use Q35 and UEFI, not Seabios. Hopefully that helps.
My install of pop os is using grub instead of systemmd should i follow the ubuntu 20.04 tutorial instead? i tried with the script in the description but it did not work
It does count as two GPUs. But since it’s a laptop, there will probably be some additional challenges. I’d recommend looking for a laptop specific guide.
i couldn't find a guide for laptop so followed your instruction and have vfio.pci-ids="". I ended up add in the graphic cards pci-ids myself and got an unbootable kernel. Followed by a 30 minutes recovery I managed to get my system back. Guess i'll just have to stop playing around before things gotten worse haha.
@@rockmanvnx6 Here are two sites talking about some of the issues facing laptop GPU passthrough. It really depends on the laptop you’re using. gist.github.com/Misairu-G/616f7b2756c488148b7309addc940b28 medium.com/@Leduccc/simple-dgpu-passthrough-on-a-dell-precision-7450-ebe65b2e648e
@@PavolElsig I went into the config file and manually changed the addresses to my GPU and it seems to have worked. Now the only problem I have now is the iommu group is bound with other devices. So apparently I need to do an ACS patch to isolate the 2 devices from that group. I'm just not exactly sure how. If you got any tips - please let me know. :D www.reddit.com/r/VFIO/comments/gu974u/gpu_passthrough_not_sure_how_to_work_acs_patch/
@@rsxhitman AFAIK if you CPU has an integrated GPU like intel does or the Ryzen with Vega graphics inside you can use it in the host by the HDMI exit in the motherboard while the guest uses the dedicated GPU. If you need to use dedicated GPUs on the host and in the guest you will need 2 GPUs.
Hey, I had this guide working on a test bench but when I tried this with my main PC, the nvidia gpu after running script still says it is using the nouveau driver and not vfio.
Make sure linux isn’t using the Nvidia GPU for display output. Linux should be using something else. The GPU you are passing through should be secondary/non boot.
Steven Porte Are the virtualization extensions (both of them) turned on in bios? And if that does not help blacklisting nouveau might help. Also make sure kernelstub -p (I think thats what it is) show the ids of your GPU.
This won’t work for the iGPU. You could use the iGPU for Linux and passthrough a dedicated GPU to the VM. If you want to passthrough an iGPU, you could use gvt-g or gvt-d.
How does the script decide which GPU to change kernel driver on? I've tried the script a few times and it keeps using the wrong gpu. Is there a way to specify which gets used?
Dude, one day a guide to single gpu-passthrough? Only with your videos have I managed, thanks for everything.
No problem, and maybe one day!
@@PavolElsig I would love that, I'm seriously considering a subtle gpu passthrough with popos but I'm a little scared haha. Great video btw
Awesome work ! I was trying to set up a GPU passthrough and failed until I followed your guide. Thank you so much !
You make in 22.04?
Can't wait to try this when i get 2 GPUs, thanks for the tutorial!
Let me know how it works out
The hero we dident want. But we needed
Seriusly you probably saved a lot of people butts. Included mine!
vector Thanks for the feedback. I’m glad it worked.
WoW I need this guide for PoP OS. Thanks in advance. Your other video pop up in recommendation and check the videos and see this
Thanks! Glad to hear it
Thank you very much! Your script allowed me to actually unmount gpu on my laptop! Lenovo y720 on Pop_OS 21.10
Heads up for anyone trying this out on a laptop, your GPU must have a dedicated output port on the laptop and show up in lspci as a VGA controller.
Hey, nice tutorial. The speach and the demonstration are quite good!
Everything worked as it should!
For those, who are wondering, the GPU for guest will be all the time off, unless you add some scripts (also called hooks) for single gpu passthrough (that bounds/unbounds gpu drivers from vfio).
After I run the script, I receive the message that my Non-primary GPU is now vfio-pci. But after reboot lspci -k still says that my videocard still has "Kernel driver in use: Nvidia".
LEGEND
Ran thru this. Enabled SVM, IOMMU, in bios. the script ran without a hitch. but my secondary graphics is still show up as using amdgpu instead of vfio-pci. I have an rx580 that i use primary, and a 5600G that i'd like to pass the integrated graphics to the VM. is it possible?
Lovely guide, well explained :)
If I add my NVIDIA PCI the VM won't even start anymore. I push start then it freezes until I REBOOT the PC
thx for this scripts and tutorial, finally I was able to passtrought my 3060ti, sadly MacOS Catalina doesn't detect it D; I would like to try on the future on windows vm's... but im afraid gaming would not be entirely possible thanks to anti cheats systems
the copy paste script part is not a solution it will change or break...
You should do a single GPU passthrough guide
how to reverse this ?
Does this still work currently? i tried running the script and it said that non primary GPUs were bound to vfio-pci, but when i reboot my nvidia card says that my drivers in use are "nouveau" still.
Thank you for making such an useful videos
Majid Kamran Thanks
@@PavolElsig Is it possible to make two vm based on discrete GPU?
Of course not in the same time but it's my goal to have a windows and a Linux guest.
Is it possible?
@@MajidKamran Yes, it is possible. You just have to create another VM and add that GPU. A warning will pop up, but you can ignore it since you're not running them concurrently. If you have multiple GPUs available, you can even run multiple VMs at the same time: th-cam.com/video/rWuNAIHTIJA/w-d-xo.html
it says driver in use: nvidia
Thank you so much worked great
Great to hear!
Would this work with pop os! 21.10?
How to passthrough igpu ? which is already primary i guess?
this will work on mint, yeah? it uses the apt package manager, and it's similar to ubuntu just like pop is, so i figure that it will.
Having a bit of an issue, everything seems to be working. The script works without a hitch or any errors, lspci shows all of my RTX 2060 kernel drivers as vfio-pci, and when I start up the VM it does start running and it looks fine when I use spice to check it on my host desktop, but I'm getting absolutely no signal from the passthrough GPU. Nada. The GPU seems to be running fine, the fans are spinning and the PC can still see it. But there's just nothing being outputted.
I'm using an i7-8700k and a RTX 2060 on a MSI Z370 GAMING PLUS motherboard
EDIT : Fixed by remoting into the VM from host and installing GPU drivers.
Me too. Did you ever figure it out? I have the same GPU.
@@JohnJigsaw420 Believe I fixed it by remoting in through a Display Spice first (the thing that lets you control the VM desktop from the host), and then installing the GPU drivers on the VM. :)
the QEMU/KVM manager is stuck at "Creating Virtual Machine" part. it says creating domain... and never moves forward from that point on. any help would be awesome!!
I have desktop with i5-4460 and nvidia 1050 ti. Pop Os has set nvidia as primary, can I use the Intel gpu just for virtmanager ? I can't seem to get to windows installation. After I do "press any key" then it goes to "Loading Files [][][][][][[]". I wont get to see the TianoCore logo. Its just black screen if you scroll top you would see "BdsDxe: starting Boot0001 "UEFI QEMU DVD-ROM QM00003 " .....
You see that on the second monitor connected to 1050ti? Make sure you’ve deleted all the virtual gpu parts (QXL, spice ...).
@@PavolElsig Yes I deleted all virtual gpu parts and added the intel gpu and audio pci. "Graphical console not configured for guest" is result of this.
Actually, what I want to have is inside my host PopOS(using primarily nvidia gpu), is virt manager with windows 10 guest running on the intel hd graphics.(I haven't connected any monitor to the motherboard just the 1050 ti).
I only need to use one monitor cuz that is all I have. Its just like if I am using virtualbox/vmware. I'm not trying to use the virtualization for another physical desktop.
Just one to have intel graphics performance on vm. Am I misunderstanding gpu passthrough?
In a GPU passthrough setup you’d normally have your linux outputting though the Intel GPU. Your VM would get the secondary GPU passed through to it. Your VM would be outputting video through that GPU to a monitor connected to it. That message about graphical console not being connected is what you should get. There is no video being sent back to Linux.
My Nvidia card a gtx 1050 mobile it is a 3D controller so the script doesn't work on it is there any way to rectify this
Laptops are generally pretty difficult to do passthrough on. If you have 2 GPUs (Intel iGPU would be enough) you might be able to do it. I would recommend a laptop specific guide though.
should this work if i have 2 identical gpus? the popos helper script doesnt seem to actually change the kernel driver in either of my gpus
what if i want to use the gpu in popos for example for python work . ? is there a quick way to switch ?
Virt-manager ist stuck at creating domain.
find a fix?
i have a fresh install of pop os with nvidia drivers and only one hdmi to the gpu. 1050 ti with i7 8700. but after the reboot the audio driver is snd_hda_intel and vga is still nvidia
moose43h Is linux using the Nvidia card? You might need to use the Intel GPU as the boot GPU. That way linux would be outputting from the iGPU and the guest OS would use the Nvidia.
So i got the vm working and everything. But my host pop os won't boot when theirs a display plugged into the passed graphic card. I have to unplug the monitor. Boot to pop os. re plug the motor and then boot to windows. I followed your guide is, is there anyway to fix this?
Is the GPU you’re trying to pass-through in the first pcie slot?
Yes. Due to limited space I can't move it. There is no uefi option to switch them either.
Thank you! 👍
No problem!
Is there a single tutorial for single GPU pass-throughs on Pop OS?
Guess not
Anyone tried running the script lately on Pop!_OS 22.04 LTS? After running it and rebooting the kernel driver in use is still 'nvidia' here.
Edit: Seems the script adds an empty vfio-pci.ids"" line to the kernelstub which I can't seem to remove with kernelstub -d. It just moves around in the list when doing so. So maybe there is something reentring it right after I delete it? I've tried adding my own vfio-pci.ids with the appropiate GPU ids from the lspci -k list, but I don't know if the empty vfio-pci.ids entry cancels it out. Still can't set the kernel driver in use to vfio.
Question , are you using 2 graphics cards on your pc? , i mean , are you using the onboard for popos and the nvidia has your passthrough video card ?
hi
I struggle to have this working.
I followed all the step and my secondary GPU still load the Nvidia driver instead of the vfio-pci
I have VT-X enabled on my bios but I cannot enable VT-D (if I do so, my system doesn't boot anymore)
Can my problem be related to VT-D being disabled?
My motherbord is a ROGSTRIX Z370-E GAMING (I just update bios version from 1002 to 2401)
My CPU is an Intel® Core™ i5-8600K
Thanks in advance for your answer
Yeah, it won’t work without vt-d being enabled.
@@PavolElsig Thanks for your answer. I think I am stuck with 2 computers in this case. Since no operating system (tried on Windows10 and Pop0S 20.10 ) want to boot when I activate VT-D in bios
what if i wanted to use my gpu on my host pop os again?
The script is in plain text. Open it in a text editor to understand the changes that it makes. At the end it creates an uninstall.sh, to revert the changes.
Do I need to have a separate keyboard and mouse for this? Will it open the vm as a window or do I need to connect a secondary display?
Thanks
what if the script sets the wrong gpu to run vfio? how do i choose which gpu i want to run the vfio driver?
having this exact issue
tried to do this guide, the script was never able to get the ids of my Nvidia GTX 1060 6GB and put them in the commandline
If the 1060 is your primary GPU, it can't be passed through. Linux would need to use a different GPU for display output and then it should work. If you have an Intel iGPU, it should be enough. Otherwise you'd need to get another GPU for the host.
@@PavolElsig I have an intel iGPU that I plugged my HDMI monitor and my DVI-D monitor into it. It still says when I load the script that the boot GPU is my Nvidia GTX 1060. I have VT-d enabled in my bios so that wouldn't be an issue also. I manually put the ids of the NVIDIA GTX 1060 in the commandline and just caused my popos install to hang at startup.
@@mulct4727 You might need to change the boot GPU in BIOS. Select the iGPU.
@@PavolElsig For my motherboard (the Gigabyte B365M DS3H) the BIOS will automatically detect any intergrated graphics and will have an option in there. It has 3 options (Auto, Enabled, Disabled) I have it on enabled, I don't know where I would select a boot gpu in my BIOS.
@@mulct4727 You should be able to change it under peripherals and initial display output.
When i use my intel UHD graphics i cant see my gpu anymore if i type lspci -k.
I got a RTX 2060 Super and an Intel I5-8600k on an Asus Z390 TUF Gaming, i have no idea why this happens.
I selected the CPU as the main graphics in the BIOS.
Try running “lspci | grep NVIDIA” that GPU should still be there
@@PavolElsig I found the problem, i have to turn it on via system76-power graphics power on.
But now i have a different problem, when i try to boot the VM i get stuck at the Press any key to boot from CD or DVD. Whenever i press anything it stops and i have to force shut it down, any clue why this happens?
@@crylia8666 Make sure the keyboard is connected/routed to the VM. You could also try to install Windows without a GPU and add it later to a fully working VM and install the drivers.
@@PavolElsig it is, I now deleted the VM and made a new one. No idea what was wrong but thanks for the help it's working now
You're saying in this video that secondary GPU should work with that script, could you please confirm that sentence because your previous Fedora video you were saying in the comment that it disabled any other GPU while guest VM was running
The primary/boot GPU always stays with the host. Secondary GPUs are bound to vfio-pci and can’t be used by the host. Only guest OSes can use those GPUs. Both the Fedora guide and the PopOS guide work like that.
@@PavolElsig I lust've read it wrong then thank you for your comment and content !
stuck creating domain any idea why?
What hardware are you running? IOMMU groups could be another cause. Try creating a VM without a GPU just to make sure it’s GPU passthrough causing it.
@@PavolElsig works w/o gpu, nvidia geforce 1060 3gb
Is the GPU and it’s audio portion controlled by vfio-pci? If not blacklisting nouveau might help. Is it in it’s own IOMMU group? If nothing else helps using a different pcie slot might help.
Could this be done with a singel GPU and intel iGPU where the iGPU is the host and the PCie GPU is the guest slave? using looking glass, 8700k + z390 motherboard?
That would work just fine. Just make sure that the iGPU is set as your boot GPU in BIOS.
@@PavolElsig Awesome, thats actually made my day, was thinking i needed to go buy another "lower end" gpu xD
I've got an rtx2060 max-q in my laptop and the win10 KVM only recognises it as a Microsoft VGA adapter, the latest NVIDIA driver doesn't recognize the card, how do I go about installing the driver? Thanks you in advance
Aiden Craig Allman Laptops are more difficult to do passthrough on. You might need some additional steps. Try following this guide medium.com/@Leduccc/simple-dgpu-passthrough-on-a-dell-precision-7450-ebe65b2e648e
Probably a stupid question, I'm using kernelstub --add-options "intel_iommu", do I need to add the "vfio-pci.ids=10de:10fa,10de:1fb9" too?, The VM seems to see the PCI devices
Kernelstub needs the GPUs you’re passing through. You’d have to list them one by one. Those devices’ll be bound to vfio-pci.
@@PavolElsig might have found the root cause of my issue, looks like the subsystem id is Changing as soon as KVM starts and takes control of the vfio driver, is there a way to hardcode these so I can install the NVIDIA driver?
TKS!!
No problem
If I have 2 monitors , how do you set them up? One monitor to each GPU?. Is possible to use spice display? Can you have a monitor setup so it connected to different gpu example display port to one gpu1 and HDMI to the other gpu2?
Each monitor is connected to one of the GPUs. It's totally up to you whether it's connected to a display port or HDMI. You could still use spice, but if that's what you want, check out Looking Glass because it's way better.
@@PavolElsig Thank you, will do haven't got it to working yet. All your videos are helpful.
I want to ask, can I continue using my Nvidia GPU on Linux after I close my vm?
I don't believe so. What you are referring to is something called vGPU, which is virtualizing a GPU to be used by one or more VMs and the host. Think basically how virtualization works for other hardware like RAM and CPU resources. This is still not something unlocked in consumer GPUs. What this guide is showing is passthrough, which is excluding the GPU from use with the host system and making it dedicated to the VM. Hence why, if you read the script READ.me file on Github, it requires that you have two GPUs present in your system. This can be two dedicated graphics cards or one GPU and a CPU that has integrated graphics as well.
Is there any way to get GPU pass through working on a laptop with two gpus
It’s possible on some laptops, but it’s more complicated than on a desktop. Look for a laptop specific guide. There are a couple out there.
I'm having an issue when it comes to booting into windows for the first time. After I press any key to boot into windows, it just starts loading files and then keeps showing a black screen. I've tried downloading another copy of windows and hoping that works but its the same thing as well. Can someone help me out?
Maybe try installing Windows with the default virtual GPU (the one I deleted) and add the actual GPU once Windows VM is up and running.
@@PavolElsig I have a NVidia 1060 6gb. The VM will not run without the default GPU which I find weird. the Nvidia parts are already read as vfio-pci but the VM doesn't want to behave properly
Make sure you delete all the parts associated with the virtualized GPU, spice and anything related. Use Q35 and UEFI, not Seabios. Hopefully that helps.
I had the same problem. My solution is that the second graphics card that you are trying to pass to the VM needs to be connected to a monitor
My install of pop os is using grub instead of systemmd should i follow the ubuntu 20.04 tutorial instead? i tried with the script in the description but it did not work
James Yes, Ubuntu 20.04 guide should work in your case.
i have an integrated intel graphic chip and a gpu 970m. Does it count as 2 graphic cards or no?
It does count as two GPUs. But since it’s a laptop, there will probably be some additional challenges. I’d recommend looking for a laptop specific guide.
i couldn't find a guide for laptop so followed your instruction and have vfio.pci-ids="". I ended up add in the graphic cards pci-ids myself and got an unbootable kernel. Followed by a 30 minutes recovery I managed to get my system back. Guess i'll just have to stop playing around before things gotten worse haha.
@@rockmanvnx6 Here are two sites talking about some of the issues facing laptop GPU passthrough. It really depends on the laptop you’re using.
gist.github.com/Misairu-G/616f7b2756c488148b7309addc940b28
medium.com/@Leduccc/simple-dgpu-passthrough-on-a-dell-precision-7450-ebe65b2e648e
Thank you 😊
Hi! Do you know if this works with a laptop? I have a MSI PE70 7RD and I'm so anxious to make this work on it! Thank you so much for this! Keep going!
I haven't tested it on any laptop. I'm guessing it won't.
@@PavolElsig I'm able to passthrough my NVIDIA GTX 1050 (no audio device) but windows guest doesn't detect it. I'm wondering if I missed something...
This might help: medium.com/@Leduccc/simple-dgpu-passthrough-on-a-dell-precision-7450-ebe65b2e648e
What are the options if my graphics cards are the same?
justin credible The guides for Fedora, Manjaro and I think Suse would work in that situation.
@@PavolElsig Wow thank you for the prompt reply! I'll report back if I get anything working.
Any idea why the secondary gpu's kernal driver is not changing to VFIO-PCI?
Is your BIOS set to use the iGPU as boot GPU? That’s one thing I can think of.
@@PavolElsig I went into the config file and manually changed the addresses to my GPU and it seems to have worked.
Now the only problem I have now is the iommu group is bound with other devices. So apparently I need to do an ACS patch to isolate the 2 devices from that group. I'm just not exactly sure how. If you got any tips - please let me know. :D
www.reddit.com/r/VFIO/comments/gu974u/gpu_passthrough_not_sure_how_to_work_acs_patch/
Daniel Wheeler I’m planning to release a guide for ACS patching this week. Other than that, you could just move one of them to a different slot.
@@PavolElsig Thanks man - keep up the good work! :D
How do you undo this?
There should be an uninstall.sh in the folder
do i need to connect my monitor to the onboard gpu or the standalone?
Both, the output from the host comes out of the onboard GPU. Your VM will be using the standalone one.
Pavol Elsig so you need 2 gpus for this correct? Is there a way to use one only
@@rsxhitman AFAIK if you CPU has an integrated GPU like intel does or the Ryzen with Vega graphics inside you can use it in the host by the HDMI exit in the motherboard while the guest uses the dedicated GPU. If you need to use dedicated GPUs on the host and in the guest you will need 2 GPUs.
Hey, I had this guide working on a test bench but when I tried this with my main PC, the nvidia gpu after running script still says it is using the nouveau driver and not vfio.
Make sure linux isn’t using the Nvidia GPU for display output. Linux should be using something else. The GPU you are passing through should be secondary/non boot.
@@PavolElsig Linux is using and booting with an Amd gpu
Steven Porte Are the virtualization extensions (both of them) turned on in bios? And if that does not help blacklisting nouveau might help. Also make sure kernelstub -p (I think thats what it is) show the ids of your GPU.
@@PavolElsig Do you mean vt-d and vt-x? As well, yes i see the ID's of my gpu that I want to passthrough in the Kernel Boot Options.
Yes, those need to be enable in BIOS. Without vt-d and vt-x it won’t work.
is there a version with one gpu
yahia ghadiry Yes, look for single GPU passthrough. It is significantly more difficult, though.
@@PavolElsig Ok thanks
Curious about doing this in the future. Is there any input lag or does it feel like playing natively?
No noticeable input lag. There’s a very minimal loss of performance.
@@PavolElsig Sweet! I wonder if it would work with windows LTSC version that has no bloatware and more privacy options?
I think it should work just fine with LTSC
Hey, i have follow all of ur video but i get this error when i try to install windows : zupimages.net/up/20/41/yof7.png
Have u any idea ?
Thank's
This is some random Windows error. Option 3 -repair might help. Downloading a fresh iso from Microsoft's website might be a good start too.
creating domain
doesn't work
maxximus How many GPUs do you have? Are all the virtualization extensions enabled in BIOS?
@@PavolElsig I have 2 GPUs. After it prompts to press any key to boot from CD-ROM, all I get is a black screen.
maxximus Is it an AMD GPU? Try BIOS instead of UEFI and see whether that helps.
@@PavolElsig ryzen CPU with integrated graphics with nvidia GeForce GTX 2060 GPU
My iGPU only shows intel 915 when doing lspci?
I connected it in the VM, don’t do shit
This won’t work for the iGPU. You could use the iGPU for Linux and passthrough a dedicated GPU to the VM. If you want to passthrough an iGPU, you could use gvt-g or gvt-d.
@@PavolElsig I only find stuff about gpu switching for iGPU/NVIDIA? I have an AMD GPU.
@@obinator9065 How many GPUs do you have?
@@PavolElsig 1 amd and the iGPU of the 8600k
@@obinator9065 As long as the host/linux is using the iGPU (not the dedicated GPU), this guide should work.
How does the script decide which GPU to change kernel driver on? I've tried the script a few times and it keeps using the wrong gpu. Is there a way to specify which gets used?
Does it pick the card with the most PCIE bandwidth? Because I am using a 2060 and a 750, attempting to passthrough the 2060 (in top gpu slot)