GPU Passthrough + Windows VM with Looking Glass #3455
Replies: 8 comments 25 replies
-
|
This is amazing! |
Beta Was this translation helpful? Give feedback.
-
|
Want to test GPU passthrough before it's merged? Follow these steps: # 1. Fetch PR branch
git -C "$OMARCHY_PATH" fetch origin pull/3454/head
git -C "$OMARCHY_PATH" checkout -B feat/omarchy-gpu-passthrough FETCH_HEAD
# 2. Switch to test branch
omarchy-update-branch feat/omarchy-gpu-passthrough
# 3. Test features
omarchy-gpu-passthrough info detect # Check hardware compatibility
omarchy-gpu-passthrough setup # Run setup (backs up config files)
sudo reboot # Apply kernel parameters
# 4. Verify configuration (after reboot)
omarchy-gpu-passthrough info verify # Check everything is configured correctly
# Optional: generate reports (helpful for troubleshooting)
omarchy-gpu-passthrough info report # Hardware configuration summary
omarchy-gpu-passthrough info diagnose # Full diagnostic report
# 5. Install Windows VM
omarchy-windows-vm install # Install Windows VM with Looking Glass
# 6. Return to master
omarchy-update-branch master
git -C "$OMARCHY_PATH" branch -D feat/omarchy-gpu-passthroughNotes:
|
Beta Was this translation helpful? Give feedback.
-
|
omarchy-gpu-passthrough-diagnostics-20251128_003120.txt Hello brother, greetings! I'm coming from YouTube since you asked me to share the report generated about the RAM error problem. I assigned 4GB and 6 cores to the VM. This is the error: ❯ ERROR: qemu-system-x86_64: unable to map backing store for guest RAM: Invalid argument |
Beta Was this translation helpful? Give feedback.
-
|
I come from the future, brother, and I've already solved it. All I did was edit the docker-compose.yml file and modify the amount of RAM allocated to Looking Glass. The default is 128M, but I assigned 32M, which according to the LG website is enough for 1080p, which is my case, and now the VM works perfectly. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the report - very helpful! And congrats on getting it running! You're right - 32MB is enough for 1080p. The 32MB option was missing. Adding it and fixing some bugs in the next update. |
Beta Was this translation helpful? Give feedback.
-
|
This works great - 4gb Sapphire + 12gb RTX 2060 I fixed this with adding the linux zen kernel, then ACS patch. Here are the full details on how I got it running on my older workstation. Environment
Issue 1: Setup Wizard Fails Silently (GRUB users)Problem: Cause: Script expects Limine bootloader, fails when GRUB is detected. Workaround: Manually create config file: sudo tee /etc/omarchy-gpu-passthrough.conf << 'CONF'
GPU_PCI_ADDR="04:00.0"
GPU_NAME="NVIDIA GeForce RTX 2060 12GB"
GPU_VENDOR_ID="10de"
GPU_DEVICE_ID="1f03"
GPU_AUDIO_ID="10f9"
GPU_DRIVER_ORIGINAL="nvidia"
GPU_IOMMU_DEVICES="04:00.0 04:00.1"
CONFSuggestion: Add GRUB support or clearer error message when non-Limine bootloader detected. Issue 2: IOMMU Group Not Viable (AMD 500 Series Chipset)Problem: Docker container crashes with: Cause: AMD 500 series chipset puts GPU in same IOMMU group as USB controller, SATA, NVMe boot disk, and Ethernet. Can't bind all to vfio-pci. Fix: Install linux-zen kernel (has ACS override patch) and add kernel parameter: # Install linux-zen
sudo pacman -S linux-zen linux-zen-headers
# Add ACS override to GRUB
sudo sed -i 's/GRUB_CMDLINE_LINUX_DEFAULT="\(.*\)"/GRUB_CMDLINE_LINUX_DEFAULT="\1 pcie_acs_override=downstream,multifunction"/' /etc/default/grub
sudo grub-mkconfig -o /boot/grub/grub.cfg
# Reboot into linux-zen kernelSuggestion: Detect bad IOMMU grouping during setup and warn user about ACS override requirement. Issue 3: Looking Glass Device Not FoundProblem: Looking Glass host app in Windows shows "Looking glass device not found" Cause: kvmfr module loaded but Fix: # Create kvmfr config
echo 'options kvmfr static_size_mb=128' | sudo tee /etc/modprobe.d/kvmfr.conf
# Reload module with size parameter
sudo modprobe -r kvmfr
sudo modprobe kvmfr static_size_mb=128
# Fix permissions
sudo chown $USER:kvm /dev/kvmfr0
sudo chmod 660 /dev/kvmfr0
# Create udev rule for persistence
echo 'SUBSYSTEM=="kvmfr", OWNER="'$USER'", GROUP="kvm", MODE="0660"' | sudo tee /etc/udev/rules.d/99-kvmfr.rulesSuggestion: Issue 4: Package Conflict (netcat)Problem: Installation fails due to gnu-netcat vs openbsd-netcat conflict. Fix: sudo pacman -S openbsd-netcat # replaces gnu-netcatSummary of Required Manual Steps (GRUB + AMD 500 series)
Files Created/Modified
Issue 5: Looking Glass Client Exits Instantly (SPICE socket permissions)Problem: Cause: SPICE socket at Fix: sudo chmod 777 /var/run/omarchy-windows/spice.sockSuggestion: Docker container should create socket with user-accessible permissions, or add udev rule / tmpfiles.d config. Issue 6: IOMMU Group Numbers Change After ACS OverrideProblem: After enabling ACS override, GPU moves to different IOMMU group (e.g., 15 → 21, audio → 22). Docker compose config has hardcoded group number. Cause: ACS override splits groups, changing group assignments. Fix: Update # Find new group numbers
basename $(readlink /sys/bus/pci/devices/0000:04:00.0/iommu_group) # GPU
basename $(readlink /sys/bus/pci/devices/0000:04:00.1/iommu_group) # Audio
# Update docker-compose.yml devices section with new group numbersSuggestion: |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the detailed report and congrats on getting it running! This is helpful.
Testing non-standard configs (different kernels, bootloaders) takes much more time, so community feedback like yours really helps ❤️ Thanks for taking the time to document everything! |
Beta Was this translation helpful? Give feedback.
-
|
Try using the latest version and let me know: #3455 (comment) |
Beta Was this translation helpful? Give feedback.
-
|
Hi, first of all, thank you so much for your work, I really appropriate it! I ran into an issue during during the install script. After downloading windows, attempting to boot it causes the docker container to shut down and the install script to hang, only printing the used storage space every minute. The logs from the docker container are: https://pastebin.com/1ZDxP02w This is with the latest commits in your branch from 3 hours ago. Running the install script again, choosing to do a quick reinstall and disabling gpu pass through allows the container to start and windows to start installing. If there is anything else I can provide that would be useful, let me know! Thank you once again for working on this :) |
Beta Was this translation helpful? Give feedback.
-
|
I've pushed a fix for Intel 12th/13th gen + Looking Glass DMA mapping failures. BIOS settings (keep as-is):
To update and test: This adds |
Beta Was this translation helpful? Give feedback.
-
|
Okay, we have progress, windows is installing, I'll run some tests and let you know tomorrow how it behaves, thank you for taking some time to look into this, you are awesome :) |
Beta Was this translation helpful? Give feedback.
-
|
Windows installed and ran without any other issues. I've also re-enabled ReBAR. I've tested out some software and games that I usually run, there are the results: SoftwareSome of the more interesting ones would be:
GamesFurMark2 P1080 score is higher than the median score for my GPU - https://www.gpumagick.com/scores/1894031
OtherI also encountered some small stuff here and there which might not be actionable on your end, but I thought might be useful for you to know and maybe document. When I initially started testing this PR, I read though the requirements you posted further up here, enabled everything in the BIOS and then ran
I couldn't get it to run at my monitor's native resolution and refresh rate, even a after disabling the VM adapter and updating nvidia drivers.
The install process puts the windows recovery partition after the NTFS partition.
Well, this is what I managed to gather so far today. If anything interesting comes up, I'll let you know. Also, if you want me to test something, let me know. Once again, thank you so much for your work to make this happen, I really enjoy this setup and I can't wait to see these changes merged for everyone to enjoy. Cheers! :) |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for all this - really helpful! And congrats on getting it working!
Yes - I'll add a note to
That's a dockurr/windows limitation - the unattended install creates this layout. One workaround could be adding a second disk in docker-compose manually.
Planning to add "Custom" option for manual input. Same for RAM. Regarding Javelin - I've tested with SecureBoot and TPM 2 enabled, same result. Battlefield 6 shows Javelin error regardless. If you have time, a few questions:
And if you could share: No rush on any of this. Thanks again! |
Beta Was this translation helpful? Give feedback.
-
|
Thank you as well!
I used the highest option, I figured that since I'm running 1440p@165Hz, I might as well give it more memory.
None so far, audio is combined between host and vm sources without issues to both wired and bluetooth outputs. Input also properly registers inside the vm.
Yes, bidirectional copy/paste is working out of the box.
I have, the vm seems to be unresponsive for about ~20 seconds after host resume, during that time, looking glass is also unresponsive. Trying to kill and start LG again during this short span of time results in LG being unresponsive with a still image of what was last seen in the from the vm or with a black screen. LG resumes working fine after this initial period passes. Also works fine if you don't have it running and start it after waiting for those ~20 secs to pass.
I no longer do, unfortunately. It would indeed be useful to have bare-metal benchmarks to compare against. That said, based on the FurMark score, which came out higher than the median for this GPU, and my overall experience so far, the performance feels genuinely native, despite any losses from virtualization, looking glass, the output signal coming form the MB instead of the GPU, and etc. I presume you could mitigate some of them by using the GPU output directly. I do run the VM with 16 cores and 16 GB of RAM and with a high-end GPU, so other people's mileage might vary. In addition, most games I’ve tested actually run better in the VM than through Proton, in some cases, I get nearly double the framerate. But maybe that's just a skill issue and I don't know how to use Proton properly.
Yes, it is, I've been using windows VMs for years, but this is the first time I actually run it in a docker container, with GPU passthrough on top of that. I was actually looking into how to do it like 10 years ago, but I never got around to actually giving it a shot.
I was looking into Omarchy, I really liked the modular approach to building this distro and I wanted to see what the community was up to. When I was browsing the PRs, yours was the newest, opened an hour ago, and it caught my eye. As I was running windows natively for some time now, mostly due to work, I was getting tired of all the AI slop and the increasing amount of bugs that nobody seems to address, so I am quite excited to finally be back on linux as my main OS, instead of running it in a VM or WSL, and this is in no small part thanks to you.
|
Beta Was this translation helpful? Give feedback.
-
|
Hey, boss! I have a new problem... I've been trying all day to get LG to access the virtual machine, but I can't get it to work. It connects with RDP. I followed the steps in your video to disable the Red Hat image controller. If I keep this controller active, I can access it with LG, but the performance is terrible... I only use a monitor connected directly to the motherboard with the iGPU. Something curious is that if I access it with RDP + keep alive, I can see the LG service in Windows running, but it doesn't connect to my host. I even downloaded it from the official B7 website, compiled it, and ran that exe file (stopping the process that comes with the B7 in your project), and I can access the client, but it doesn't allow me to control it at all. I could see the movements but not the mouse in the RDP window, and when I close RDP, the LG client window also closes. Here is my configuration in case you see anything. Thank you for everything. |
Beta Was this translation helpful? Give feedback.
-
|
Looking Glass needs your GPU to have a display connected - otherwise Windows won't render anything to send. Options:
I use 2 monitors on Linux, one also has a GPU cable for Windows VM. Looking Glass shows that GPU output as a window on my Linux desktop. After connecting display to GPU:
|
Beta Was this translation helpful? Give feedback.
-
|
Thank you very much, that was precisely the problem. I thought it was something else; I thought I didn't have to have it connected to the graphics card output. Thank you, now everything works perfectly. I'm going to try some games. You're a beast. |
Beta Was this translation helpful? Give feedback.
-
|
Hey, gave it another go, here are the results: ❯ ERROR: qemu-system-x86_64: -device vfio-pci,host=0000:01:00.0: vfio 0000:01:00.0: group 2 is not viable Here are the diagnostic results: omarchy feat/omarchy-gpu-passthrough ❯ omarchy-gpu-passthrough info detect GPU Detection Detected 2 GPU(s): GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630] GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile] ✓ IOMMU: Enabled === Hardware Configuration Report === SYSTEM Omarchy: MOTHERBOARD BIOS SETTINGS GPUs [2] NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile] (01:00.0) IOMMU GROUPS (for passthrough) DISPLAY CONFIGURATION Monitors per GPU: DRIVER VERSIONS GPU PASSTHROUGH STATUS Generated: 2025-11-29 19:18:25 GPU Passthrough Diagnostics ℹ️ Collecting system information... [SYSTEM INFORMATION] [CPU] [Operating System] [Bootloader] [Motherboard] [Display Session] [GPU INFORMATION] [GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630]] DRM Card: card1 [GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile]] [IOMMU INFORMATION] IOMMU Status: Enabled Active VFIO IOMMU Groups: 1 [GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630]] [GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile]] [KERNEL CONFIGURATION] [Kernel Command Line] [IOMMU Parameters] iommu=pt: ✓ Enabled [LOADED KERNEL MODULES] [VFIO Modules] [NVIDIA Modules] [AMD GPU Modules] [DRIVER VERSIONS] [NVIDIA Driver] [AMD GPU Driver] [VFIO Driver] [GPU PASSTHROUGH CONFIGURATION] [GPU Configuration: /etc/omarchy-gpu-passthrough.conf] GPU Passthrough Config - Generated 2025-11-29GPU_PCI_ADDR="01:00.0" [Limine Configuration: /etc/default/limine] [VFIO Configuration: /etc/modprobe.d/vfio.conf] VFIO GPU Passthrough - Generated 2025-11-29options vfio-pci disable_vga=1 [Driver Blacklist: /etc/modprobe.d/blacklist-gpu-passthrough.conf] GPU Passthrough - Prevent NVIDIA auto-load at bootinstall nvidia /bin/false [System Environment (/etc/environment)] [PROCESSES USING GPU] [NVIDIA GPU Processes] [DRI Device Handles] [NVIDIA Device Handles] [RECENT GPU PASSTHROUGH LOGS] [Last 50 lines of /var/log/omarchy-gpu-passthrough/gpu-passthrough.log] RELEVANT BOOT MESSAGES (dmesg) [IOMMU Messages] [VFIO Messages] [Boot VGA Device] [PCI Passthrough Messages] [END OF DIAGNOSTICS REPORT] Report saved to: /tmp/omarchy-gpu-passthrough-diagnostics-20251129_191825.txt ✓ Report saved: /tmp/omarchy-gpu-passthrough-diagnostics-20251129_191825.txt Issues here imo: Device: 0000:00:01.0 |
Beta Was this translation helpful? Give feedback.
-
|
Please try the latest version. # Fetch latest changes
git -C "$OMARCHY_PATH" fetch origin pull/3454/head:feat/omarchy-gpu-passthrough
omarchy-update-branch feat/omarchy-gpu-passthrough
# Re-run setup (will now auto-detect USB controller)
omarchy-gpu-passthrough setup
# Reboot for binding to take effect
sudo reboot
# Remove old Windows VM
omarchy-windows-vm remove
rm -rf ~/.config/windows/
# Reinstall VM
omarchy-windows-vm installLet me know if it works. |
Beta Was this translation helpful? Give feedback.
-
|
Hi! GPU Detection Detected 2 GPU(s): GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630] GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile] ✓ IOMMU: Enabled === Hardware Configuration Report === SYSTEM HARDWARE GPUs
DISPLAY PASSTHROUGH STATUS Passthrough GPU: IOMMU Group 2: VFIO Modules: loaded Issues: DRIVERS Generated: 2025-11-30 15:08:07 GPU Passthrough Diagnostics ℹ️ Collecting system information... [SYSTEM INFORMATION] [CPU] [Operating System] [Bootloader] [Motherboard] [Display Session] [GPU INFORMATION] [GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630]] DRM Card: card1 [GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile]] [IOMMU INFORMATION] IOMMU Status: Enabled Active VFIO IOMMU Groups: 1 [GPU 1: Intel Corporation CoffeeLake-H GT2 [UHD Graphics 630]] [GPU 2: NVIDIA Corporation TU106M [GeForce RTX 2060 Mobile]] [KERNEL CONFIGURATION] [Kernel Command Line] [IOMMU Parameters] iommu=pt: ✓ Enabled [LOADED KERNEL MODULES] [VFIO Modules] [NVIDIA Modules] [AMD GPU Modules] [DRIVER VERSIONS] [NVIDIA Driver] [AMD GPU Driver] [VFIO Driver] [GPU PASSTHROUGH CONFIGURATION] [GPU Configuration: /etc/omarchy-gpu-passthrough.conf] GPU Passthrough Config - Generated 2025-11-30GPU_PCI_ADDR="01:00.0" [Limine Configuration: /etc/default/limine] [VFIO Configuration: /etc/modprobe.d/vfio.conf] VFIO GPU Passthrough - Generated 2025-11-30options vfio-pci disable_vga=1 [Driver Blacklist: /etc/modprobe.d/blacklist-gpu-passthrough.conf] GPU Passthrough - Prevent NVIDIA auto-load at bootinstall nvidia /bin/false [System Environment (/etc/environment)] [PROCESSES USING GPU] [NVIDIA GPU Processes] [DRI Device Handles] [NVIDIA Device Handles] [RECENT GPU PASSTHROUGH LOGS] [Last 50 lines of /var/log/omarchy-gpu-passthrough/gpu-passthrough.log] RELEVANT BOOT MESSAGES (dmesg) [IOMMU Messages] [VFIO Messages] [Boot VGA Device] [PCI Passthrough Messages] [END OF DIAGNOSTICS REPORT] Report saved to: /tmp/omarchy-gpu-passthrough-diagnostics-20251130_150807.txt ✓ Report saved: /tmp/omarchy-gpu-passthrough-diagnostics-20251130_150807.txt |
Beta Was this translation helpful? Give feedback.
-
|
I've looked at the diagnostic report. The config shows GPU_IOMMU_GROUP="unknown" and GPU_IOMMU_DEVICES="", which means IOMMU detection failed during setup. This is why the USB controller ID wasn't added to vfio.conf. I need more data to understand why detection fails on your system. Could you run these commands and share the output? # Check if IOMMU symlink exists
ls -la /sys/bus/pci/devices/0000:01:00.0/iommu_group
readlink -f /sys/bus/pci/devices/0000:01:00.0/iommu_group
# List devices in IOMMU group 2
ls /sys/kernel/iommu_groups/2/devices/
# Check if xhci_hcd is built-in
cat /lib/modules/$(uname -r)/modules.builtin | grep xhci
# Current mkinitcpio MODULES
grep "^MODULES" /etc/mkinitcpio.confThen re-run setup with full terminal capture: # Fetch latest
git -C "$OMARCHY_PATH" fetch origin pull/3454/head
git -C "$OMARCHY_PATH" checkout -B feat/omarchy-gpu-passthrough FETCH_HEAD
omarchy-update-branch feat/omarchy-gpu-passthrough
# Re-run setup (copy entire terminal output)
# When asked about reboot, choose NO
omarchy-gpu-passthrough setup
# Check these files before reboot
cat /etc/modprobe.d/vfio.conf
cat /etc/omarchy-gpu-passthrough.confAfter that, reboot and test: sudo reboot
# After reboot, generate reports
omarchy-gpu-passthrough info verify
omarchy-gpu-passthrough info report
omarchy-gpu-passthrough info diagnose
# Try VM installation
omarchy-windows-vm installUpload all outputs to pastebin.com. This will help narrow down whether it's a permission issue, kernel config, or something specific to your hardware. |
Beta Was this translation helpful? Give feedback.
-
|
Try using the latest version and let me know: #3455 (comment) |
Beta Was this translation helpful? Give feedback.
-
|
Hello, I'd like to contribute the output of my attempt at setting up passthrough on my computer (Lenovo Legion 5 laptop in hybrid graphics mode) After running omarchy-gpu-passthrough info detect |
Beta Was this translation helpful? Give feedback.
-
|
Try using the latest version and let me know: #3455 (comment) |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
GPU Passthrough + Windows VM with Looking Glass
Got GPU passthrough working with Looking Glass for Windows VM. This is for the request in #3307.
PR: #3454
Demo: YouTube - full setup, Fortnite, CS2, benchmarks
How it works:
You can bind your GPU to the VM for gaming, then switch it back to the host whenever you want. Looking Glass gives you low-latency display through shared memory so there's basically no lag. I can play Fortnite or CS2 in the VM, then switch the GPU back to the host and use LM Studio while the VM is still running.
There's hardware spoofing built in - it auto-detects your SMBIOS info so Windows sees real hardware instead of VM stuff. Anti-cheat works because of this. Fortnite and CS2 both run fine.
Setup is an interactive wizard that handles IOMMU config, kernel parameters, driver stuff, and sets up Hyprland to use your integrated GPU so your desktop stays on the iGPU.
What I'm running:
Community tested hardware:
What hasn't been tested yet:
Important: Looking Glass needs a display on GPU
Your dedicated GPU needs a display connected - otherwise Windows won't render anything. Options:
Want to try it?
You need 2 GPUs - one for the VM, one integrated for your desktop. Setup wizard does most of the work.
If you test it, run
omarchy-gpu-passthrough info verifyafter reboot to check everything is configured correctly. Let me know:omarchy-gpu-passthrough info diagnoseand share the outputIf it doesn't work that's also good to know - helps me figure out what hardware has issues.
Heads up:
This is experimental. There's a chance of black screen on first boot (there's a TTY2 recovery thing in the PR if that happens). Back up your stuff before trying.
Thanks for any testing or feedback!
Beta Was this translation helpful? Give feedback.
All reactions