Windows.sh
My configuration and other files for QEMU to launch Windows VM with GPU passthrough (VFIO)
Install / Use
/learn @new-sashok724/Windows.shREADME
windows.sh
My configuration and other files for QEMU to launch Windows VM with GPU passthrough (VFIO)
Hardware:
CPU: Ryzen 3950X
Mobo: Asus X570 Hero WiFi
RAM: 4x16 GB (64GB) @ 3600MHz CL16
GPU (Host): RX 5700 XT
GPU (Guest): GTX 1080 Ti
/etc/modules-load.d/kvmfr.conf:
kvmfr
/etc/modprobe/kvm.conf:
# kvm_amd avic is incompatible with nested (TODO Enable back on 5.19+)
# vhost_net experimental_zcopytx causes host lockup under high bandwidth load
options kvm ignore_msrs=N report_ignored_msrs=Y
options kvm_amd avic=1 nested=0
options vfio_pci ids=10de:1b06,10de:10ef,1912:0015 disable_vga=1
#options vhost_net experimental_zcopytx=1
softdep nvidia pre: vfio_pci
/etc/modprobe/kvmfr.conf:
options kvmfr static_size_mb=128
/etc/udev/rules.d/20-kvmfr.conf:
SUBSYSTEM=="kvmfr", OWNER="root", GROUP="kvm", MODE="0660"
Specials:
- Found many useful flags in other configurations, as well as in qemu official doc
- AVIC, CPU-PM and preallocated RAM are enabled to reduce latency
- KVM state is not hidden, since NVidia allowed using their GPUs in VMs recently
- VirtIO devices are used for anything that can be paravirtualized
- kvmfr module is used for creating IVSHMEM device
- QXL display can be viewed until GPU is initialized
- Spice and JACK audio devices are both working perfectly with no stuttering
- Separate physical USB controller is also passed through; qemu-xhci is buggy when used with multiple same devices (e.g. with Valve HMD controllers)
- Secure boot and TPM2 are enabled
Related Skills
node-connect
341.0kDiagnose OpenClaw node connection and pairing failures for Android, iOS, and macOS companion apps
frontend-design
84.4kCreate distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
openai-whisper-api
341.0kTranscribe audio via OpenAI Audio Transcriptions API (Whisper).
commit-push-pr
84.4kCommit, push, and open a PR
