Hello everyone,
I'm running a Ryzen 9 9955HX (Zen 5) with an integrated Radeon 610M GPU passed through to a Windows 11 VM using VFIO on Proxmox VE. I'm following a slightly modified version of the isc30/ryzen-gpu-passthrough-proxmox guide (specifically the method described in issue #131 using a hookscript for GPU reset).
The Problem:I've isolated a stability issue related to the CPU model used in the VM configuration.
My Setup:
What I've Tried:
The Goal:I need the stability offered by kvm64 but with the modern CPU features (instruction sets, virtualization enhancements) provided by host or EPYC-v4/v3. The kvm64 model is clearly missing features that the guest driver/OS needs for long-term stability under load when combined with the QEMU/VFIO passthrough environment on this specific hardware.
Request for Help:Has anyone else encountered similar instability with modern AMD CPU models (host, EPYC-*) on Zen 5 hardware (especially with integrated graphics passthrough)? Are there specific CPU flags I should be enabling on top of EPYC-* models, or perhaps specific QEMU arguments not exposed by the Proxmox GUI that could bridge the gap? Could this be related to a QEMU/KVM bug with Zen 5 features, or perhaps an interaction with the specific integrated GPU (Radeon 610M)? Any insights or suggestions for further debugging would be greatly appreciated.
Thank you!
I'm running a Ryzen 9 9955HX (Zen 5) with an integrated Radeon 610M GPU passed through to a Windows 11 VM using VFIO on Proxmox VE. I'm following a slightly modified version of the isc30/ryzen-gpu-passthrough-proxmox guide (specifically the method described in issue #131 using a hookscript for GPU reset).
The Problem:I've isolated a stability issue related to the CPU model used in the VM configuration.
- Configurations that FAIL (Severe Artifacts/Crashes): Using cpu: host, cpu: EPYC-v4, or cpu: EPYC-v3 results in severe video artifacts, complete screen blackouts, and AMD driver crashes in the guest OS, especially under load (e.g., playing multiple 4K60p videos).
- Configuration that "Works" (Stable but Limited): Using cpu: kvm64 initially causes minor artifacts that clear up after the guest boots and caches settle. After this initial period, the system is stable and handles demanding video workloads without the catastrophic crashes seen with the other models.
My Setup:
- Host CPU: AMD Ryzen 9 9955HX (Zen 5)
- Host Motherboard: MS-A2 (BIOS version? Latest microcode?)
- Passed GPU: AMD Radeon 610M (Integrated into 9955HX)
- Guest OS: Windows 11
- Proxmox VE Version: [Fill in your version]
- QEMU Version: [Fill in if known, or pveversion output]
- Hookscript: Modified version of isc30 guide (resets GPU to vfio-pci/amdgpu)
- IOMMU: Enabled (iommu=pt in kernel command line)
What I've Tried:
- Different CPU models (host, EPYC-v4, EPYC-v3, kvm64).
- Different CPU flags combinations (disabling spec-ctrl, ssbd, pcid with EPYC-v4).
- Switching machine type (i440fx, q35).
- Using hugepages (hugepages: 2M).
- Ensuring stable power management settings in the guest OS (Windows High Performance plan, AMD driver set to High Performance).
The Goal:I need the stability offered by kvm64 but with the modern CPU features (instruction sets, virtualization enhancements) provided by host or EPYC-v4/v3. The kvm64 model is clearly missing features that the guest driver/OS needs for long-term stability under load when combined with the QEMU/VFIO passthrough environment on this specific hardware.
Request for Help:Has anyone else encountered similar instability with modern AMD CPU models (host, EPYC-*) on Zen 5 hardware (especially with integrated graphics passthrough)? Are there specific CPU flags I should be enabling on top of EPYC-* models, or perhaps specific QEMU arguments not exposed by the Proxmox GUI that could bridge the gap? Could this be related to a QEMU/KVM bug with Zen 5 features, or perhaps an interaction with the specific integrated GPU (Radeon 610M)? Any insights or suggestions for further debugging would be greatly appreciated.
Thank you!