Skip to main content
Solved

Metis M.2 card on a Gigabyte B550I AORUS PRO AX system

  • July 23, 2026
  • 4 replies
  • 35 views

I'm having trouble getting a Metis M.2 card to be detected on a Gigabyte B550I AORUS PRO AX system, and I'm trying to determine whether this is a compatibility issue or if the card itself may be faulty.

System configuration

  • Motherboard: Gigabyte B550I AORUS PRO AX (latest BIOS)

  • CPU: AMD Ryzen 5 5600

  • Operating System: Windows 11

I tested the card in both available M.2 slots:

  • M2A_CPU (front slot)

  • M2B_SB (rear slot)

To rule out an issue with the motherboard, I first installed a known-good NVMe SSD in each slot. The SSD is detected correctly by both the BIOS and Windows in either slot.

When I replace the SSD with the Metis card, using the same BIOS configuration, the behavior changes completely:

  • The BIOS does not detect any device in the M.2 slot.

  • Windows Device Manager does not show a new PCIe device, not even as an unknown device.

  • The MetisOmega INF driver installs successfully, but reports that it was installed on 0 devices, which makes sense since Windows never detects the card.

I've also checked the following:

  • Both M.2 slots are configured to use PCIe (Auto), not SATA.

  • Lane-sharing limitations described in the motherboard manual have been taken into account.

  • The exact same slots and BIOS settings work perfectly with the NVMe SSD.

Based on these tests, it seems that the Metis card is not establishing a PCIe link with the motherboard.

Could you please let me know:

  • Whether there are any known compatibility issues between the Metis M.2 card and the Gigabyte B550I AORUS PRO AX.

  • If there are any additional BIOS settings or diagnostics I should try (PCIe link status, LEDs, etc.).

  • Whether these symptoms are consistent with a defective card and if I should proceed with an RMA.

Thank you for your time. I appreciate any guidance you can provide.

Best answer by Samuel.Maldonado

Hi! Thanks

Just wanted to update you that it’s working now! Moving it to the M2B_SB slot did the trick after a couple of BIOS adjustments:

  • Power Saving: Tweaked the PCIe power-saving options to avoid sleep state conflicts.
  • iGPU Memory: Reduced the shared memory allocated to the CPU's integrated graphics.

Once those settings were applied, the Metis enumerated properly as a PCIe device.

This was likely happening because on this B550I motherboard, Slot 1 (M2A) links directly to the CPU lanes, whereas Slot 2 (M2B_SB) routes through the chipset/southbridge. That bridged path seems to handle the accelerator's initialization and lane distribution much better.

Appreciate the guidance on this!

4 replies

Spanner
Axelera Team
Forum|alt.badge.img+3
  • Axelera Team
  • July 23, 2026

Hi ​@Samuel.Maldonado, nice work on the systematic testing. A couple of notes and next steps:

First thing that might help is that the BIOS not listing the Metis in the M.2 storage screen is probably normal. Because Metis is a processing accelerator rather than an NVMe drive, I don’t think most BIOSes will show it there. Can’t say for sure especially with newer BIOSes, but FYI.

Wha we’re looking for is whether it enumerates as a PCIe device, and I’m not sure it currently does.

In which case, let’s think about a few things to try!

  1. In the BIOS, can you set the M.2 slot's link speed explicitly to Gen3 (Metis is a Gen3 x4 device and some boards fail link training on Auto/Gen4). While you're there, also disable any "PCIe ASPM" power-saving options for that slot.
  2. A full cold boot is surprisingly helpful sometimes. Power off completely, switch off the PSU for ~30 seconds, then power on, etc. Metis sometimes needs a genuine cold power cycle to link up.
  3. Maybe test with a Linux live USB (e.g. Ubuntu) and run lspci? Look for 1f9d:1100 or Axelera AI Metis AIPU (rev 02). It just removes the Windows driver layer from the equation and is a giid conclusive test. If the card enumerates under Linux, it suggests a Windows/driver issue; if it doesn't, we’re looking at a link-level problem. Bit of a faff, but a great diagnostic tool.
  4. Since it's an AMD platform, if the card does appear under Linux but misbehaves, amd_iommu=off is a known fix, but that's a later step.

Let me know if any of that helps!


Hi! Thanks

Just wanted to update you that it’s working now! Moving it to the M2B_SB slot did the trick after a couple of BIOS adjustments:

  • Power Saving: Tweaked the PCIe power-saving options to avoid sleep state conflicts.
  • iGPU Memory: Reduced the shared memory allocated to the CPU's integrated graphics.

Once those settings were applied, the Metis enumerated properly as a PCIe device.

This was likely happening because on this B550I motherboard, Slot 1 (M2A) links directly to the CPU lanes, whereas Slot 2 (M2B_SB) routes through the chipset/southbridge. That bridged path seems to handle the accelerator's initialization and lane distribution much better.

Appreciate the guidance on this!


Spanner
Axelera Team
Forum|alt.badge.img+3
  • Axelera Team
  • July 24, 2026

Ah, nice work! That’s worth knowing, about how the two slots behave a little differently. Thanks for sharing!

What’s the project you’re building, by the way?


Hi! It's an test object detection project, so we're training custom YOLO models.

We've run into a few issues deploying our custom YOLO11 model with the Axelera SDK. The deployment eventually succeeds, but we've seen some warnings and behavior we're still investigating (for example, around the deployment/calibration process).

To narrow things down, we're going to try the same workflow with a YOLOv8 model using the default Axelera configuration to see whether the issues are specific to YOLO11 or to our custom setup.