I am giving up on VM Gaming | Deploy on Friday

← All posts<br>Saturday<br>I am giving up on VM Gaming<br>By Norbert Takács · Tuesday, 2 June 2026

6 years ago I got lured into the promises of VM gaming with the original 2 Gaming Rigs, 1 Tower Linus Tech Tips video. It described how one could use virtualisation software (Unraid or Proxmox ) on the host, spawn virtual machines with physical graphics cards using PCIe passthrough, and achieve near identical gaming performance. You could even slice up your existing storage and CPU easily or pass them as is.<br>Like many others I also owned an under utilised gaming pc. Most of the times the hardware was sitting idly waiting to be turned on for those hourly gaming sessions.<br>1 Gaming PC, 1 NAS<br>Unlike what Linus ended up doing I wasn't drawn to the idea because I had more than one player wanting to play games and was constrained by hardware. It was the potential to utilise my gaming pc as a network attached server mainly and gaming pc secondly. At the time I already owned the beefiest consumer 8 bay Synology 1817+ NAS money could buy and was not pleased with the performance it's Intel Atom C2538 CPU provided.<br>Unraid host OS<br>I opted for Unraid since it had a more mature PCIe passthrough experience out of the box compared to Proxmox and it was also the OS that the original video used. Both systems use the same KVM/QEMU/Virtio stack. The performance boost my docker workloads received after switching to an Unraid host gaming PC NAS combo were night and day compared to the Intel Atom Synology NAS .<br>Gaming VM issues<br>Only draw back was that VM gaming was a bit of a nuance to use. Some of it was expected since I was running an experimental setup, and I don't shy away from tinkering, I brushed them off as issues that can be solved.<br>💾 SOLVED: VM State management<br>To start out, managing the VMs state has to be done through the UI of the host OS. Every time you want to turn on/off/restart the VM, you have to preform a bit of click-ops on a website. There are creative ways around this using third party devices which would interact with the API directly, I have used home-assistant to control the state of the VM.<br>💽 SOLVED: USB passthrough<br>To be able to use USB devices you need to attach them to the VM. This can be done in the web UI, but it means every time you attach a new usb drive you have to click yourself through to attach the device. If the device glitches and disconnects, you have to do it over again. Instead if you pass through a PCIe USB controller, Fresco Logic FL1100 in my case, you get to have the plug and play experience you are used to.

Fl1100 can be found from 2 port to 8 port and works with VIRTIO!

🐧 SOLVED: NVIDIA VBIOS patching<br>This is not required anymore but it's worth mentioning. There used to be an arbitrary error from NVIDIA created to prevent users running GeForce GPUs in virtual environments. To bypass it you had to extract the VBIOS , patch it and load it on runtime in the host os, otherwise the NVIDIA driver would fail to install with an Error 43 .

Nvidia purposefully blocking the usage of the GPU in VM-s

🍨 UNSOLVED: GPU sharing with docker and VMs<br>I was hoping I could use the GPU if the VM was not used for docker containers. This is possible but not user friendly, the GPU has to be bound to VFIO at boot. Meaning if you have setup the GPU for the Gaming VM by binding it to VFIO, and want to test out some docker local LLM, you must first unbind it in the UI, then restart the computer. Trying to use the GPU for both use cases at the same time either results in the docker container not starting, or in worse case the system locking up.

Handy VFIO binding tool in unraid

🗂️ UNSOLVED: Multi VM GPU sharing<br>Another interesting use case would have been running multiple VMs using a single GPU. You can attach the GPU to multiple VMs but by default only one VM can be running which utilises the GPU. Running multiple VMs simultaneously requires a hardware splitting technology SR-IOV which is already used by many network cards. For GPU-s SR-IOV exists mostly on enterprise Nvidia Ampere level cards which I don't own, and the functionality is further locked into the NVIDIA vGPU stack and require annual licensing. There are supposedly vGPU workarounds but it adds another layer of annoyance on top of the enterprise GPU cost which might not be optimised for gaming, alternatively you could use vgpu_unlock and turn your consumer grade GeForce GPU into its Quadro equivalent.<br>🥀 UNSOLVED: Stability, lags, corrupt VMs and dread<br>There are plenty of fixes you have to perform to get the VM usable, from fixing audio glitches by setting interrupts correctly in MSI manager. CPU Pinning, reserving RAM and passing in real drives have helped with speed and stability. KVM XML tuning became a constant maintenance task to avoid regressions, lag and lockups. I experienced VM degradation, Windows VMs that used to work properly became unbearably slow for no apparent reason. It was also hard for...