ESXi 8.0 PSOD with IOMMU Fault Reason 0x79 involving NVIDIA Driver 580.126.08 and Management Daemon 580.65.05

Products

• VMware vSphere ESXi

Issue/Introduction

• ESXi 8.0 hosts may experience a recurring Purple Screen of Death (PSOD) when utilizing NVIDIA vGPU profiles. This occurs when the IOMMU detects an invalid memory translation entry.

• The host crashes with a purple screen displaying the following signature:
  
  Log File Path: /var/run/log/vmkernel-zdump.log

YYYY-MM-DDT HH:MM:SS cpu#:#######)@BlueScreen: IOMMU Fault detected for (vmgfx#/nvidia-gpu) IOaddr: ############ Reason: 0x79 (Invalid Read/Write permission(R=W=0) for second-level paging entry) Domain: ############

Environment

VMware vSphere ESXi 8.0

GPU Driver: NVD-VMware_ESXi_8.0.0_Driver version 580.126.08-1OEM.800.1.0.20613240

Management Daemon: nvdgpumgmtdaemon version 580.65.05-1OEM.700.1.0.15843807

Cause

The NVIDIA 580.x driver series contains a logic error where memory pages assigned to the GPU are occasionally marked with "No Read" and "No Write" permissions (R=W=0) in the second-level page tables.

When the IOMMU hardware attempts to process a DMA (Direct Memory Access) request against these pages, it triggers a fault, leading to the ESXi host crash

Resolution

• Verify the version of driver installed by running below command:

1. 1. esxcli software vib list | grep -i nvidia
   2. Confirm if both 580.126.08 and 580.65.05 are installed.

• Consult the NVIDIA Support for a driver update (e.g., version 580.131 or newer) that addresses "IOMMU Fault 0x79" for ESXi 8.0