Microsoft has quietly published KB5072093, a targeted component update that bumps the AMD MIGraphX Execution Provider to version 1.8.35.0 for Windows 11, and Microsoft will deliver it automatically through Windows Update to devices running Windows 11, version 24H2 and version 25H2. The public KB entry is deliberately short — it confirms the update’s purpose (“includes improvements to the MIGraphX Execution Provider AI component”), the supported OS versions, and the delivery mechanism, and it notes that the update requires the latest cumulative update for the corresponding Windows 11 branch before it will apply. This release is part of Microsoft’s ongoing practice of componentized AI/runtime updates that improve hardware-accelerated inference without waiting for full cumulative rollups.
The KB is an incremental but meaningful part of the broader push to optimize on-device AI through modular runtime updates. For organizations that depend on AMD acceleration, this update is worth validating quickly — but treat it as a systems-level change and follow a staged rollout with careful functional checks and telemetry to avoid surprises.
Source: Microsoft Support KB5072093: AMD MIGraphX Execution Provider update (1.8.35.0) - Microsoft Support
Background
What is AMD MIGraphX and why Windows ships it
MIGraphX is AMD’s graph optimization and compilation engine designed to accelerate neural-network inference on AMD GPUs and NPUs. It integrates with frameworks such as ONNX Runtime and Torch-MIGraphX to compile and schedule model execution for AMD compute stacks, taking advantage of ROCm and other AMD runtime components. On Windows, Microsoft sometimes distributes vendor execution providers as modular components so system-level AI features — from photo transforms to low-latency assistant tasks — can route model execution onto vendor-optimized engines. The ONNX Runtime project documents the MIGraphX Execution Provider and its role in accelerating ONNX models on AMD hardware; AMD’s ROCm docs provide complementary build and runtime guidance for MIGraphX itself.Why Microsoft uses small component updates
Over the last few years Microsoft has shifted many hardware- and AI-related runtime pieces into component updates that can be updated independently of the monthly cumulative cadence. This approach lets Microsoft and hardware partners ship targeted performance, stability, and compatibility improvements for execution providers — such as MIGraphX, OpenVINO, QNN and others — more frequently and directly via Windows Update. The KB format for these component updates is often concise by design, leaving technical deep-dives to vendors or to separate engineering notes. Community analysis and prior KBs show the same pattern: small KB text, automatic Windows Update delivery, and a prerequisite LCU (latest cumulative update) requirement.What KB5072093 actually says — concise summary
- Applies to: Windows 11, version 24H2 and Windows 11, version 25H2 (all editions).
- What it does: Updates the Windows ML Runtime AMD MIGraphX Execution Provider component to version 1.8.35.0 and “includes improvements” to the component; Microsoft provides no line-by-line changelog in the KB.
- Delivery: Automatic via Windows Update; visible under Settings → Windows Update → Update history after install.
- Prerequisite: The device must have the latest cumulative update (LCU) for the relevant Windows 11 version installed before the component will apply.
- Replacement: Microsoft states the update does not replace any previously released update.
How MIGraphX fits into the Windows AI stack
Execution Providers and ONNX Runtime
ONNX Runtime is the cross-vendor inference runtime widely used in Windows and in many AI-enabled applications. It supports Execution Providers (EPs) — modular backends that implement operator kernels and scheduling for target hardware. MIGraphX is one such EP tailored to AMD hardware; it implements graph-level optimizations, operator kernels, and platform-specific compilation flows that enable faster model inference on AMD GPUs and NPUs. When Microsoft ships the MIGraphX EP as a Windows component, system and application workloads that rely on ONNX Runtime can transparently take advantage of AMD acceleration without each app bundling its own EP version.What the EP actually does at runtime
In practical terms, the MIGraphX EP:- Parses ONNX graphs and applies graph-level optimizations and operator fusions.
- Compiles selected subgraphs into device-native kernels and schedules them for AMD devices.
- Manages kernel caches and compiled artifacts to reduce time-to-first-execution on subsequent runs.
- Exposes configuration options and environment variables to tune device selection, cache directories, and log verbosity.
Real-world impact: what users and admins are likely to see
Immediate, visible changes
Because Microsoft’s KB provides no micro-benchmarks, the visible, user-facing effects are likely to be modest but measurable in targeted scenarios:- Faster or more stable inference for apps that route ONNX workloads through the MIGraphX EP. This can affect media transforms, camera effects, and assistant tasks that run locally.
- Reduction in crashes or functional failures for workloads that previously triggered EP-level parsing or kernel bugs — if the update includes hardening fixes (unverifiable until detailed notes appear).
- First-run compilation behavior may change: some workloads may experience longer first invocations (compilation) but faster subsequent executions due to improved caching. This is typical for compiled EPs such as MIGraphX.
Operational effects for managed fleets
For IT administrators, the practical concerns are predictable:- The update is delivered via Windows Update and requires the LCU, so patch sequencing matters. Ensure endpoints are updated to the required build before expecting the component to apply.
- Because the component is shared by multiple apps and system features, a single update can affect many workloads simultaneously — a strength for rapid distribution but a risk for broad regressions. Prior community work on componentized EP updates underscores this operational trade-off.
Technical analysis — likely contents of the update (evidence-based inference)
Microsoft’s terse KB text does not enumerate exact code-level changes. However, combining the public description with known patterns for EP updates allows a reasoned, evidence-based inference about likely targets:- Performance tuning: micro-optimizations in operator kernels, threading, memory reuse, and fusion strategies to reduce latency and CPU overhead for common image and small LLM workloads. ONNX Runtime and AMD EP updates often focus here.
- Kernel and backend plugin improvements: updates that improve compatibility with recent ROCm or driver versions, or that add new optimized kernels for operator variants. Vendor-run EP updates frequently include these changes.
- Stability and parsing hardening: fixes for malformed graphs, input validation, or race conditions during graph partitioning and compilation — common hardening targets for provider updates.
Compatibility and risk factors
Driver, firmware and dependency coupling
Execution Providers like MIGraphX often assume a matched stack: specific MIGraphX binaries are validated against certain ROCm/driver/firmware revisions. Mismatches between the EP version and device driver or OEM firmware are a frequent cause of regressions. Before broad deployment:- Confirm your AMD driver/ROCm/firmware versions align with vendor guidance.
- Verify OEM camera/ISP drivers remain compatible when system-level inference touches imaging pipelines.
Rollback complexity
Component updates delivered via Windows Update can be harder to roll back than standalone installers. Managed environments should plan rollback runbooks — e.g., image-based recovery or driver/OS-level restore points — rather than relying on a simple “uninstall” path for componentized EP updates. Community guidance on component updates repeatedly warns about this operational friction.Opaque public changelogs
Microsoft’s KB text for many EP updates is intentionally minimal. Organizations that require traceable change logs, CVE mappings, or reproducible benchmark numbers should treat the KB as a packaging notice and request detailed release notes or security advisories from AMD or Microsoft support if required for compliance. Prior community analysis reinforces that KBs of this style rarely include exhaustive engineering detail.How to validate the update on your devices — practical checklist
- Confirm preconditions:
- Make sure the endpoint has the latest cumulative update (LCU) for Windows 11 24H2 or 25H2 as applicable; the KB will not apply otherwise.
- Confirm installation:
- After Windows Update installs the component, check Settings → Windows Update → Update history for an entry named Windows ML Runtime AMD MIGraphX Execution Provider (KB5072093).
- Verify runtime behavior for critical workloads:
- Run representative ONNX workloads or app scenarios that rely on AMD acceleration and capture:
- Time-to-first-inference (wall clock)
- Throughput (inferences/sec)
- CPU and GPU utilization profiles
- Compare results to pre-update baselines.
- Collect telemetry and logs if you see regressions:
- ONNX Runtime session logs, Windows Event logs, and any vendor-specific driver logs are essential. If the EP compiles models to cache, verify cache location and entry timestamps to confirm whether newly compiled kernels are being used.
- Staging and pilot:
- Test the update on a representative pilot fleet with a variety of AMD hardware families (mobile, desktop, integrated NPU if present) before broad roll-out.
- Consider imaging or snapshot-based rollback plans in the pilot phase.
Deployment recommendations for IT and developers
- Staged rollout: Apply the update first to a pilot group representing the diversity of hardware and workloads. Monitor telemetry for at least several days to detect intermittent issues caused by first-run compilation or cache growth.
- Align drivers and firmware: Coordinate AMD driver/ROCm, OEM firmware, and the Windows servicing baseline to the versions validated by your pilot. Mismatched stacks are the most common cause of post-update regressions.
- Re-run CI and model validation: For ISVs and internal ML Ops teams, re-run model regression suites and automated validation on updated endpoints. Small kernel-level changes can alter numeric rounding, operator partitioning or quantization behavior in edge cases.
- Log retention and support packaging: If you encounter problems, gather ONNX Runtime traces, Windows Update logs, driver logs and sample failing models. These artifacts accelerate vendor triage.
What we don’t know (and how to treat those gaps)
Microsoft’s KB5072093 explicitly omits technical specifics — there are no CVE identifiers, no microbenchmarks, and no operator-level changelog. That omission leaves open several unknowns:- Whether the update contains security hardenings mapped to public CVEs. Treat any security claim as unverified until official advisories appear.
- Which exact operator kernels or graph transforms were changed or optimized. Assume the update targets a combination of performance and stability hardening until AMD or Microsoft publish details.
Bottom line — strategic and tactical takeaways
- KB5072093 is a targeted, vendor-partnered component update that advances the AMD MIGraphX Execution Provider to v1.8.35.0 for Windows 11 24H2/25H2. It will be pushed automatically via Windows Update to qualifying devices that already have the latest cumulative update.
- The update is most relevant to users and organizations that rely on ONNX Runtime-based acceleration on AMD hardware; it can improve performance, stability, or compatibility for such on-device AI workloads.
- Because Microsoft’s public KB is short, treat claims about exact fixes or security contents as provisional. Plan pilot rollouts, verify driver/firmware compatibility, collect logs for any anomalies, and maintain rollback plans for managed fleets.
Quick reference — action checklist for admins (short)
- Ensure devices are on the latest cumulative update for Windows 11 24H2/25H2.
- Allow Windows Update to install KB5072093 automatically, or watch Update history for the “Windows ML Runtime AMD MIGraphX Execution Provider (KB5072093)” entry.
- Pilot the update on representative hardware, re-run ML validation tests, and monitor telemetry for regressions.
- If you need granular change detail (security or functional), contact AMD or Microsoft support and request a detailed release note or CVE mapping.
The KB is an incremental but meaningful part of the broader push to optimize on-device AI through modular runtime updates. For organizations that depend on AMD acceleration, this update is worth validating quickly — but treat it as a systems-level change and follow a staged rollout with careful functional checks and telemetry to avoid surprises.
Source: Microsoft Support KB5072093: AMD MIGraphX Execution Provider update (1.8.35.0) - Microsoft Support
