Windows Server: Multiple A16 GPUs may cause Blue Screen Error During a PCI Scan
Summary: This article talks about that in Windows Server 2019, or 2022, with multiple A16 GPUs a blue screen error may show up during a PCI scan.
Symptoms
User may notice a Blue Screen Error with stop code SYSTEM_THREAD_EXCEPTION_NOT_HANDLED (7e) if there are multiple A16 GPUs installed.
Note: The system can boot back to the operating system after the blue screen error.
Note: Windows Server 2016 is also affected but is End-of-Life.
Steps to Reproduce:
Install two or more units of NVIDIA A16 in the server.
Install the Windows Server 2019 or Windows Server 2022 operating system.
Install the chipset driver, SWRAID (S140/S150/S160) driver, or perform PCI scan through Device Manger.
Cause
For Windows Server 2022 or previous operating system versions, the OS follows a certain algorithm for ARI devices.
If the child’s Max Payload Size (MPS) is smaller than the parent’s, the upstream port can send instructions that the child cannot answer.
If that happens, the endpoint produces an error and results in either a device disconnect or a blue screen error. In the failing case, the GPU displays MPS of 256 while the parent (USP and Rootport) is supporting the values of 512 for MPS.
Resolution
Windows Server 2022 Fix: March 12, 2024—KB5035857 (OS Build 20348.2340) - Microsoft Support
HCI 23H2 Fix: March 12, 2024—KB5035856 (OS Build 25398.763) - Microsoft Support
