ACPI/NVRM Errors with NVIDIA RTX 3090 on TrueNAS Scale 25.10.0.1

TrueNAS General
, ,
1

Hi TrueNAS Community,

I’m running TrueNAS Scale 25.10.0.1 on the following hardware:

  • CPU: Intel Core Ultra 7 265K

  • GPU: NVIDIA RTX 3090 (24GB VRAM)

  • Mainboard: ASUS TUF Gaming Z890-PLUS WIFI

  • BIOS: Updated to latest version

Issue:
I’m experiencing persistent ACPI/NVRM errors (e.g., AE_ALREADY_EXISTS, knvlinkCoreShutdownDeviceLinks_IMPL) during startup, operation, and especially when using GPU-intensive services like Ollama or Open-WebUI. The errors appear continuously in the logs, and I’m concerned about potential performance degradation.

Question:
Has anyone else encountered similar issues with this TrueNAS Scale 25.10.0.1 + NVIDIA RTX 3090 combination? If so, did you find a workaround (e.g., kernel tweaks, driver updates, or BIOS settings)?

I’d appreciate any insights or shared experiences from the community. Thanks in advance!

2

UPDATE

RTX 3090 (GA102) repeatedly loses initialization / continuous knvlinkCoreShutdownDeviceLinks_IMPL messages

Hi everyone,

I’m trying to determine whether this is a hardware issue or a driver/kernel problem. After several days of debugging, I believe this is more likely related to the NVIDIA driver or its interaction with TrueNAS/Linux than to faulty hardware.

System

  • TrueNAS: 26.0.0-BETA.2

  • Kernel: Linux 6.18.23-production+truenas

  • CPU: Intel Core Ultra 7 265K

  • Mainboard: ASUS TUF Gaming Z890-PLUS WIFI

  • BIOS: Latest stable ASUS BIOS (updated during troubleshooting)

  • RAM: 32 GB DDR5

  • GPU: NVIDIA RTX 3090 (GA102, 24 GB)

Important:

The problem already existed before updating to TrueNAS 26.0.0-BETA.2. The BIOS update and TrueNAS upgrade did not change the behavior.

NVIDIA Driver



Driver Version: 590.44.01

modinfo nvidia

license: Dual MIT/GPL
version: 590.44.01

The system is using the NVIDIA Open Kernel Modules.

Main symptoms

The kernel continuously logs



NVRM: GPU0 knvlinkCoreShutdownDeviceLinks_IMPL:
Need to shutdown all links unilaterally for GPU0

approximately every few seconds.

Occasionally this is followed by



nvidia 0000:02:00.0:
loading /lib/firmware/nvidia/590.44.01/gsp-ga10x.bin failed with error -4

Direct firmware load for
nvidia/590.44.01/gsp-ga10x.bin failed with error -4

NVRM:
RmFetchGspRmImages:
No firmware image found

NVRM:
GPU 0000:02:00.0:
RmInitAdapter failed

rm_init_adapter failed

Sometimes, after these errors, nvidia-smi is no longer able to detect the GPU until the next reboot.

Things I have already checked

BIOS

Updated to the latest stable BIOS.

Verified:

  • Above 4G Decoding = Enabled

  • Resizable BAR = Enabled

  • PCIe Link Speed = Auto (not forced to Gen5)

  • ASPM disabled

None of these changes affected the issue.

PCIe Link

The PCIe link behaves correctly.

Idle:

  • PCIe Gen1 (2.5 GT/s)

During CUDA load:

  • PCIe Gen4

  • 16 GT/s

  • x16

So the PCIe connection appears completely healthy.

GPU stability

The GPU itself appears stable.

During CUDA workloads:

  • ~330 W power draw

  • ~75°C

  • ~90% SM utilization

The GPU computes normally without crashes.

No hardware-related errors

I could not find:

  • Xid errors

  • PCIe AER errors

  • ECC errors

  • VRAM errors

GSP

The firmware exists:



/lib/firmware/nvidia/590.44.01/gsp-ga10x.bin

Driver reports:



GSP Firmware Version: 590.44.01

Driver parameters include:



EnableGpuFirmware: 18
NvLinkDisable: 0

Interrupts

MSI interrupts are enabled and assigned correctly.

Why I currently don’t suspect hardware

The RTX 3090 successfully runs long CUDA workloads under full load.

PCIe trains correctly to Gen4 x16.

There are no Xid errors or PCIe AER errors.

The GPU only seems to fail during what appears to be an internal driver/GSP reinitialization.

Additional observations

The firmware file definitely exists, yet the driver occasionally reports



Direct firmware load failed

followed immediately by



RmFetchGspRmImages:
No firmware image found

which seems contradictory.

Question

Has anyone seen similar behavior with

  • RTX 3090 (GA102)

  • NVIDIA Open Kernel Modules

  • Driver 590.xx

  • GSP Firmware

Could this be a bug in the Open Kernel Modules or GSP handling rather than a hardware failure?

I have attached the complete nvidia-bug-report.log.gz generated using nvidia-bug-report.sh.

Any ideas would be greatly appreciated.

Thanks!