Hosts running legacy NVIDIA drivers experience PSOD

Products

• VMware vSphere ESXi

Issue/Introduction

• Hosts experience frequent Purple Screen of Death (PSOD).
• The PSOD backtrace indicates a page fault (Exception 14) or IOMMU fault detected in the nvidia kernel module.
• vmkernel.log shows fatal PCIe bus errors on the slot containing the NVIDIA GPU.
• The following XID errors are recorded: NVRM: Xid (PCI:0000:3b:00): 31, 32, or 43
• Virtual Machines with vGPU enabled fail to start or crash with the error: VGPU message 6 failed
• Hosts are running legacy NVIDIA drivers from the 7.0.2 (v535) branch.

Environment

Vsphere ESXI 8.0x

 

Cause

Legacy NVIDIA vGPU drivers (7.0.2 branch) are incompatible with the ESXi 8.0.x IOMMU and interrupt handling logic, leading to memory translation failures and kernel panics. "VGPU message 6" specifically indicates frame buffer exhaustion on 1B vGPU profiles under high utilization.

Resolution

 Fixed in NVIDIA vGPU software release 15.5, 16.2, and higher.

1. Upgrade the host NVIDIA VIBs to a supported ESXi 8.0 branch (e.g., vGPU 16.x).
2. See Download Broadcom products and software for steps to download these releases.
3. For VMs experiencing XID 43/Message 6 errors, increase the vGPU profile from 1B to 2B to provide additional frame buffer.
4. Ensure host BIOS and GPU firmware are updated to the latest versions supported by your hardware vendor.
5. If you need to direct a customer specifically to a support phone number, see Contact Support. Scroll to the bottom of the page and click on your respective region.
6. If you need further assistance, please reference our main Contact Broadcom Support article.