Jackalblood

No solution unfortunately just disable audio. I'm afraid I'm not around to check for sure but a quick Google should bring up the path

I did exactly the same thing USB audio but obviously that sucks overall. I am glad to hear it's not just me! I'm sorry we are in this boat together. Happy new year to you

Hey Guys I am running into a hell of an issue with my new build. I originally had a 3070 5600X build and had no issue with any of this. This build has completely stumped me. I am assuming its unfixable but I am hoping someone will know better. Here is a detailed outline of all I've tried. The end result is I can only boot the VM a 2nd time after a full shutdown IF I disable the GPU audio. TL;DR RTX 5090 passthrough on Unraid is stable only when the NVIDIA HD Audio function (01:00.1) is removed. The audio device fails FLR on VM shutdown, dead-bridges the PCIe slot, and triggers Error 127 on restart. With PCIe power management disabled, vfio_pci.disable_idle_d3=1, GPU-only passthrough, and a VBIOS ROM supplied, the VM can start/stop/restart indefinitely. Trade-off: no HDMI/DP audio. Looking to confirm whether this is an expected limitation on RTX 50-series GPUs under Unraid, or if a supported audio-safe approach exists. SystemHost: Unraid (current stable) GPU: NVIDIA RTX 5090 (GB202) VM: Windows 11 Hypervisor: KVM / QEMU Passthrough: VFIO Motherboard: ASRock X870E Taichi ObjectiveAchieve repeatable, stable GPU passthrough of an RTX 5090 to a Windows 11 VM without requiring a host reboot between VM restarts. Observed FailureVM boots correctly on first start Any VM restart after shutdown fails GPU remains unusable until a full host power cycle Errorinternal error: Unknown PCI header type '127' for device '0000:01:00.0' Root CauseThe failure is reproducible and isolated to: 01:00.1 – NVIDIA HD Audio Controller The device fails FLR on VM shutdown This leaves the PCIe slot in a dead state This matches the known NVIDIA reset bug, now observed on GB202. Platform / Firmware ConfigurationThe following motherboard and firmware settings are applied and verified. Component Setting Category Setting Required Value / Status Purpose Motherboard – ASRock X870E Taichi – Platform in use CPU / Virtualization AMD CPU Features SVM Mode Enabled Enables CPU virtualization IOMMU AMD Chipset Features AMD IOMMU Enabled Enables interrupt and memory remapping Power Management ACPI / Power Global C-States Disabled Prevents deep CPU sleep states Boot Boot Features Fast Boot Disabled Forces full hardware initialisation Final Working ConfigurationKernel Parameterspcie_acs_override=downstream,multifunction pcie_aspm=off pcie_port_pm=off vfio_pci.disable_idle_d3=1 Notes: vfio_pci.disable_idle_d3=1 is required in this configuration Prevents the GPU from entering PCIe D3cold/D3hot states Without it, the PCIe link can fail to recover reliably This does not resolve the reset bug on its own Stability still requires removal of the audio function VBIOS ROMA VBIOS ROM file is supplied: <rom file='/mnt/user/isos/rtx5090.rom'/> Used for consistent GPU initialisation only. Unrelated to the FLR/reset failure. Audio Function01:00.1 (NVIDIA HD Audio Controller) permanently removed Only 01:00.0 (GPU) passed through This resolves: Error 127 PCIe dead-bridge condition Requirement for host power cycles Current StateWorkingGPU passthrough is stable VM can start/stop/restart indefinitely No PCIe errors Not AvailableHDMI / DisplayPort audio (expected due to removal of audio function) QuestionAt this point the system is stable. Looking to confirm whether: This is expected behaviour for RTX 50-series GPUs under Unraid There is any supported way to retain GPU audio without reintroducing reset failures If the answer is simply “not currently possible”, that’s fine — just looking to confirm the limitation. SummaryRTX 5090 passthrough is reliable and repeatable Stability requires: PCIe power management disabled vfio_pci.disable_idle_d3=1 NVIDIA audio function removed Result is a working VM with no GPU-provided audio Any insight from others running GB202 would be useful.