npu_kmd Causing Windows Kernel Crashes

[Jonathan_P]

Hello, I am having an issue where I have multiple HP EliteBook 640 14 inch G11 Notebook PCs that are Kernel Crashing

I have been updating the drivers and patching the OS, as well as wiping and reimaging the laptops. 

Tha

Here is the current info:

System info

• OS: Windows 11 25H2, Build 26200.8457
• CPU: Intel Meteor Lake Intel Core Ultra 7 165U
• NPU Driver: 32.0.100.4778 (installed via Intel Driver & Support Assistant)

Root cause from the dumps

• Faulting component: npu_kmd (Intel NPU kernel-mode driver)
• Every dump contains the same sequence:

  Feature change detected  IOMMU wait timeout  Guest code led to instruction timeout. Terminating partition.

• Fence counters (LastSubmitted, LastCompleted, LastReported) are all zero at crash time — NPU appears to fail before completing any work
• IOMMU errors began appearing with driver 4778

[Robin_Intel]

Hi ,

Thank you for reaching out to the Intel Community Forum. To assist us in investigating the issue, please provide the following details:

1. What NPU driver version were you running before 32.0.100.4778? Did the crashes start immediately after updating to this version?
2. Have you tried rolling back to the previous NPU driver version? If so, did the crashes stop?
3. What BIOS/UEFI firmware version are these EliteBooks running? Is it the latest available from HP?
4. How many total units are affected? Is this happening on ALL EliteBook 640 G11s or just some?
5. What specific applications or workloads are running when crashes occur? Any AI/ML applications, Windows Studio Effects, or NPU-accelerated features?
6. Are these domain-joined machines with specific group policies that might affect hardware acceleration or NPU functionality?
7. What imaging/deployment method are you using? Are drivers being injected during imaging or installed post-deployment?

If you have any questions, please let us know. Thank you.

Best regards,

Robin G.

Intel Customer Support Technician.

[Robin_Intel]

Hi Jonathan_P,

Thank you for reaching out to the Intel Community Forum. To assist us in investigating the issue, please provide the following details:

I wanted to check if you had the chance to review the questions I posted. Please let me know at your earliest convenience so that we can determine the best course of action to resolve this matter.

Best regards,

Robin G.

Intel Customer Support Technician.

[MurrayG]

Posting to add a data point and confirm this isn't limited to Meteor Lake. The previous report cites the same npu_kmd crash signature on Core Ultra 7 165U (Meteor Lake) HP EliteBooks. I'm seeing what looks like the identical issue on a different machine and a different CPU generation, so the common factor appears to be NPU driver 4778 rather than any specific platform.

System

• CPU: Intel Core Ultra 9 285H (Arrow Lake-H, Series 2)
• GPU: Intel Arc 140T
• RAM: 96 GB
• OS: Windows 11 Pro 25H2, build 26200.8655
• NPU driver: 32.0.100.4778 (installed via Intel Driver & Support Assistant)
• NPU firmware: dated 20260422, MTL_CLIENT_SILICON branch
• Virtualization stack active (Hyper-V / WSL2 enabled), which may be relevant to the IOMMU timeout.

Symptom
Unexpected full-machine kernel crash and reboot, not a hang or a single-VM failure. Two crashes on consecutive days produced two different bugcheck codes:

• 0xF7 (DRIVER_OVERRAN_STACK_BUFFER)
• 0x4E (PFN_LIST_CORRUPT)

Different stop codes, but both minidumps contain the same npu_kmd secondary-dump block, which is what ties them together.

Root cause from the dumps
Every dump contains the same sequence from the NPU driver's bugcheck callback:

• npu_kmd!KmFenceIdRingDumpDetails: [ERROR] ... Job Queue submissions before TDR — the NPU hung and TDR (timeout detection & recovery) fired.
• Fence counters all zero at crash time (LastSubmitted = 0, LastCompleted = 0, LastReported = 0) — the NPU fails before completing any work.
• npu_kmd!NpuCriticalTraceRead: [ERROR] VPU critical trace header canary values corrupted — the driver's own trace buffer is corrupted, consistent with the memory-corruption bugchecks above.
• On my Hyper-V-heavy setup the failure also surfaces an IOMMU wait timeout, matching the other thread's report that IOMMU errors began with 4778.

What I've tried
Disabled the NPU device in Device Manager (Neural processors → Intel AI Boost → Disable). Did that just now - can report back later if it helped. 4778 is the current public driver, so there's nothing newer to update to; the only options are disable the device or roll back to a pre-4778 build (e.g. 32.0.100.4514 WHQL).

Question for Intel: Is a fix for the 4778 NPU TDR / IOMMU-timeout issue in progress, and is there a recommended interim driver version? Given this now reproduces on both Meteor Lake and Arrow Lake, it looks like a driver regression rather than a platform-specific fault.

Happy to share minidumps if useful.

[Robin_Intel]

Hi Jonathan_P,

Thank you for reaching out to the Intel Community Forum. To assist us in investigating the issue.

I will be further discussing this with a team and get back to you as soon as we have an update.

If you have any questions, please let us know. Thank you.

Best regards,

Robin G.

Intel Customer Support Technician.

[Robin_Intel]

Hi Jonathan_P,

Thank you for your continued cooperation.

To help us further investigate the issue and better understand your system configuration, could you please provide the Intel® System Support Utility (SSU) log? This will allow us to review your system details and configuration.

Additionally, could you please provide the following information:

• How many units are affected by this issue?
• Does the issue occur on all affected units, or only on a specific unit?

We would also like to ask you to test if the issue persists when using the older Intel graphics driver version 32.0.100.4724. Please install the driver, test the behavior again, and let us know the results.

Your feedback and the requested information will help us continue troubleshooting and identify the appropriate next steps.

We look forward to your update.

If you have any questions, please let us know. Thank you.

Best regards

Robin G.

Intel Customer Support Technician.

[Robin_Intel]

Hi Jonathan_P,

Thank you for your continued cooperation.

I wanted to check if you had the chance to review the questions I posted. Please let me know at your earliest convenience so that we can determine the best course of action to resolve this matter.

If you have any questions, please let us know. Thank you.

Best regards

Robin G.

Intel Customer Support Technician.