optimus-vfio-docs VS vbios_from_reg

You can follow jscinoz's guide to see if you can get it going for yourself in some very limited cases, but overall it's quite an involved task with a lot of work to succeed.

Hi. I have the same model and it is definitely possible to passthrough the NVIDIA GPU. I don't think USB-C port is connected to the GPU and I always use the HDMI one. You need to connect with a monitor or a HDMI plug otherwise the GPU will not work. I also patched OVMF to include ROM (https://github.com/jscinoz/optimus-vfio-docs/issues/2) and a fake battery. I have a hook script to switch GPU on and off using bbswitch. I use this rom since other didn't work or have some problems. I allocated 3 of 4 cores to the VM. For me, the performance is okay for games that I play. I tried Valorant a few months ago with Looking Glass and performance is not that good, however.

This whole thing started 2 years ago when I tried passing my nvidia dGPU to get it working on a Windows guest for Autodesk programs (aka, that joyful company that hates everything non-Windows and free), but lo and behold, it didn't work! Shocker! So I followed through 2 sources: - This issue thread - This guide

Here's a nice little tool if you're having trouble extracting your own vbios.

I used this method: https://github.com/kyokenn/vbios_from_reg

Extra stuff you might want to try if you still have issues: - Extract the vbios from windows (GPU-Z or with this script to get it from registry in windows. - Check with GOPUpd script to see if you have a UEFI GOP (check post2 of the forum for the script, you may also want to read the first post to understand what the script does, basically just drop your vbios into the .bat executable and follow through) - In case your vbios contains a UEFI GOP, you do not need to update, just use it as it is - In case your vbios does not contain a UEFI GOP, install one according to your device's family (Pascal) - add just above the bottom of the passed device and change /path/to/vBIOS.rom to where your vBIOS is (note that sometimes, qemu may not be able to access the file because of apparmor, so use some other place that qemu can use like /usr/share/edk2-ovmf/x64/ or /usr/share/vbios or something or add an exception for it in apparmor)

The keys for my setup were the proper PCIe location in the VM and passing a GOP updated vBIOS, otherwise I'll be met with Error 43 or no display. To get the M2000M working properly: (I assume you've read Archlinux's wiki about OVMF GPU Passthrough and how to add stuff like qemu commands in a virsh xml) - Dump the vBIOS (there are many ways, through the windows registry (manually or with a script) or by using VBiosExtractor following this guide. Note that you will need to put the full path to the BIOS updater executable, the VBiosExtractor script doesn't seem to do well with relative paths. - Add UEFI GOP to it (as it's PC-AT only) (I used Maxwell MXM) - Patch the OVMF image with the regular (or GOP updated) vBIOS by building it automatically (thanks to @Ashymad's original gist) if you're on an Arch-based OS (you'll have to download the whole gist) or check the PKGBUILD and make the patching manually (this guide has the same steps too, don't forget the dos2unix commands, otherwise the patches wont be applied). You might want to add the path of the OVMF in /etc/libvirtd/qemu.conf accordingly or edit the PKGBUILD to copy them to /usr/share/edk2-ovmf/x64/ instead (where the edk2-ovmf official package store them). - Make a VM with virt-manager as you would normally, add the dGPU and the other hardware that comes with it (for me only Audio, for others it may be also USB or others peripherals that come with a GPU nowadays), but in the XML section (make sure you enable editing), remove all PCI addressing lines (to avoid conflicting addresses) and add the proper address for the dGPU to the hostdev section (note that it usually just means that the source address and the hostdev address match). When applying, virsh will properly populate the other addresses. xml