A Windows Update screen announces an AMD GPU provider update replacing the deprecated MIGraphX provider.
Microsoft has released KB5136903, an update to the AMD MIGraphX Execution Provider. It moves the component to version 2.2611.1.0 on Windows 11 versions 24H2, 25H2 and 26H2. Microsoft's support note is short and includes no changelog. Microsoft's own developer documentation adds an important point: MIGraphX is being phased out in favor of a newer AMD provider.

A Windows Update screen announces an AMD GPU provider update replacing the deprecated MIGraphX provider. What KB5136903 delivers​

According to Microsoft's support article, published November 19, 2026, KB5136903 includes "improvements" to the MIGraphX Execution Provider AI component. Microsoft does not list which bugs were fixed, which operators changed, or whether performance improved. Any claim of faster AI on AMD graphics is a guess.

The basics:

  • Component: AMD MIGraphX Execution Provider
  • Version: 2.2611.1.0
  • Replaces: KB5132811
  • Delivery: downloads and installs automatically through Windows Update
  • Prerequisite: the latest cumulative update for your Windows 11 version must already be installed
  • Update history entry: "Windows ML Runtime AMD MIGraphX Execution Provider (KB5136903)"

Who gets it, and when​

Microsoft lists these editions as eligible:

Windows 11 releaseEditions covered
Version 24H2Enterprise and Education, Enterprise Multi-Session, IoT Enterprise
Version 25H2All editions
Version 26H2All editions

There is one timing condition. Microsoft says the update will not be offered to Windows 11, version 24H2 until its next servicing period on or after December 8, 2026. If your 24H2 machine doesn't show it yet, that's expected. The note gives no similar delay for 25H2 or 26H2.

In short: this is a quiet, automatic component update with an unspecified list of changes, and 24H2 devices have to wait until December.

What an execution provider actually does​

Windows ML runs AI models in the ONNX format through ONNX Runtime. ONNX Runtime doesn't know how to get the most out of every chip, so it relies on execution providers (EPs). An EP is a hardware-specific backend that handles the parts of a model it can run well and leaves the rest to other backends, such as the CPU. You can think of EPs as translators, each fluent in one vendor's hardware.

The ONNX Runtime project describes MIGraphX as using AMD's Deep Learning graph optimization engine to accelerate ONNX model on AMD GPUs. The provider has been around for a while. The original ONNX Runtime pull request that added it said it adds an initial execution provider using MIGraphX which supports the AMD ROCm stack. Its job on Windows is newer.

Microsoft's Windows ML documentation lists strict hardware requirements for MIGraphX: AMD RDNA 3 or later GPU, AMD GPU driver version 25.10.13.09 or later. It also says this execution provider is not supported for GenAI scenarios today. Those requirements come from Microsoft's developer documentation, not the KB itself, but they show which machines can actually use this update.

Installing the update also doesn't mean your apps will use it. Microsoft's Windows ML CLI documentation says an installed EP doesn't guarantee support for every operator, data type or shape, and that graph partitioning and fallback depend on the provider, the model and the session settings. An app also has to choose MIGraphX in the first place.

The deprecation notice the KB doesn't mention​

The KB doesn't say this, but Microsoft's Windows ML execution provider page does: MIGraphXExecutionProvider is deprecated in Windows ML 2.x. The same page says AMD recommends transitioning to AMDGPUExecutionProvider for all AMD GPU devices. MIGraphXExecutionProvider will be retired from the Windows ML API interface with a new Windows ML major release.

Microsoft describes the replacement as one that provides a unified interface that automatically selects the most appropriate AMD acceleration backend based on the available hardware and workload. In the Windows ML 2.x provider table, MIGraphX is labeled "(Deprecated)" and the AMD GPU provider is listed alongside it.

What this means in practice:

  • For end users: nothing to do. The update installs itself, and a deprecated provider still gets maintenance updates while apps depend on it.
  • For developers: KB5136903 keeps existing MIGraphX code paths working. It doesn't mean MIGraphX is the long-term choice. If you're starting a Windows ML 2.x project for AMD GPUs, follow AMD's advice and target AMDGPUExecutionProvider.
  • For IT teams: if a line-of-business app calls MIGraphX by name, ask the vendor about its migration plan before a future major Windows ML release removes the provider from the API.

In short: this is maintenance for a provider that is on its way out, so check what your apps depend on.

Where it sits in Microsoft's monthly AI update cycle​

KB5136903 is one of several updates in a monthly round. Microsoft's release history for Windows 11 versions 24H2, 25H2 and 26H2 shows these November 19 execution provider updates: AMD MIGraphX Version2.2611.1.0KB5136903 | Version2.2611.0.0KB5136902 | Version2.2611.1.0KB5136901 | Version2.2611.1.0KB5136900 | Version2.2611.1.0KB5136899. Going by the table's column order, those are AMD MIGraphX, AMD Vitis, Intel OpenVINO, Nvidia TensorRT-RTX and Qualcomm QNN.

Windows 11 version 26H1 has its own numbering. The same history lists that branch's November MIGraphX update as Version2.2611.1.0KB5136908, which shows KB5136903 is not the KB for 26H1 devices.

The version history for this group also tracks the release dates:

ReleaseKBMIGraphX version
November 19, 2026KB51369032.2611.1.0
October 28, 2026KB51328112.2610.1.0
September 22, 2026KB51287742.2609.2.0

The numbers follow a pattern of a year-and-month code plus a revision. So 2.2611 is the November 2026 build, which makes it easy to tell how current a machine is. Microsoft's Windows AI docs on GitHub also say updated versions of the execution providers are made available via Windows Update's optional nonsecurity preview releases, a.k.a. "D week releases". That matches the late-month timing of these releases.

How to check whether you have it​

  1. Make sure the latest cumulative update for your Windows 11 version is installed. KB5136903 won't install without it.
  2. Open Settings > Windows Update > Update history.
  3. Look for Windows ML Runtime AMD MIGraphX Execution Provider (KB5136903).
  4. On 24H2, don't expect it before the servicing period on or after December 8, 2026.

If it isn't there​

  • Wrong Windows version? On 26H1, look for KB5136908 instead.
  • 24H2 consumer edition? Microsoft's list of eligible 24H2 editions names only Enterprise, Education, Enterprise Multi-Session and IoT Enterprise.
  • Missing cumulative update? Install the latest one first, then check again.
  • Managed device? Your organization's Windows Update policies may control when the update arrives. Check what was actually deployed instead of assuming.

Microsoft's KB gives no manual-download procedure, so this guide doesn't offer one either.

Developer notes: Windows Update isn't the whole story​

Microsoft's Windows ML documentation says vendor EPs, MIGraphX included, aren't part of the core Windows ML Runtime binaries. Apps get them in one of two ways:

  • Through the ExecutionProviderCatalog APIs. These are Windows-certified providers, and Microsoft says they update automatically.
  • Bring your own. Developers ship and reference the provider binaries themselves, which suits offline environments, managed devices or strict version pinning.

If your app bundles its own MIGraphX build, KB5136903 doesn't change what your app runs. To see what's installed locally, the Windows ML CLI has inventory commands (winml sys --list-ep and winml sys --list-device), and MIGraphX appears under the alias migraphx.

Bottom line​

KB5136903 is a routine, automatic update for one AI component. It matters mostly to systems with RDNA 3 or later AMD GPUs running Windows ML apps that use MIGraphX. Microsoft hasn't said what it fixes, so don't expect a measurable speed boost. The bigger news is in the developer documentation: MIGraphX is deprecated in Windows ML 2.x, and AMD's newer AMDGPUExecutionProvider is the recommended replacement. Have you already moved your Windows ML projects to it, or are you still using MIGraphX? Tell us in the forums.

 

References

  1. KB5136903: AMD MIGraphX Execution Provider update (version 2.2611.1.0) | Microsoft Support Microsoft Support 2026-11-19T00:00:00+00:00
  2. KB5136903: AMD MIGraphX Execution Provider update (version 2.2611.1.0) | Microsoft Support support.microsoft.com
  3. Add amd migraphx execution provider to onnx runtime by pfultz2 · Pull Request #2929 · microsoft/onnxruntime · GitHub github.com