KB5068002 Updates Vitis AI EP to 1.8.25.0 on Windows 11 24H2 Copilot+ PCs

Microsoft has quietly pushed a targeted component update—KB5068002—that refreshes the Vitis AI Execution Provider (version 1.8.25.0) on eligible Copilot+ PCs running Windows 11, version 24H2, delivering behind-the-scenes optimizations to Microsoft’s on-device AI imaging and inference stack and arriving automatically through Windows Update once the latest cumulative LCU for 24H2 is installed.

Background​

The Vitis AI Execution Provider (Vitis AI EP) is AMD’s runtime and tooling that enables hardware-accelerated AI inference across AMD platforms—ranging from Ryzen AI processors with integrated NPUs to AMD Adaptable SoCs and AMD Alveo acceleration cards—and is the vendor-specific execution provider that integrates with Windows’ ONNX/ONNX Runtime-based AI plumbing.
Microsoft ships many on-device AI capabilities as modular, versioned components (Image Processing, Phi Silica, Vitis AI EP, and others) so they can be updated independently of the OS. These component-level updates are frequently scoped to Copilot+ hardware—systems that meet Microsoft’s NPU and platform requirements—and are delivered via Windows Update or managed update channels once the device meets prerequisites (notably the latest cumulative update for Windows 11, version 24H2).
Why this matters: moving AI workloads into updateable components shortens Microsoft’s iteration cycle for model and runtime improvements, but it also places more responsibility on IT teams and power users to manage drivers, firmware, and deployment rings carefully to avoid compatibility regressions.

---

What KB5068002 actually is (clear facts)​

• The package updates the Vitis AI Execution Provider AI component to version 1.8.25.0 for supported AMD Copilot+ systems running Windows 11, version 24H2.
• The update is delivered automatically through Windows Update and will appear in Settings → Windows Update → Update history once installed.
• The update does not replace any previously released update in general; it is a component refresh and requires that the device already has the latest cumulative update (LCU) for Windows 11, version 24H2.

These items are stated in Microsoft’s public-style KB notices and are consistent with prior component releases for on-device AI. The KB itself uses concise language—“includes improvements”—and does not publish an itemized changelog or CVE mappings in the public-facing entry. That brevity is intentional but reduces immediate visibility into exact fixes and performance deltas.

---

Technical context: where Vitis AI EP fits in Windows​

Vitis AI Execution Provider functions as a vendor-specific ONNX Runtime/Windows ML execution provider that enables models to offload inference to AMD NPUs, GPUs, and accelerator cards. On Windows, Microsoft exposes execution providers through its AI runtime catalog, allowing apps and OS features to select or register the best available hardware-backed provider for a given model. Windows’ supported execution providers list explicitly names AMD — Vitis AI as an available provider alongside Intel, Qualcomm, and NVIDIA options.
On the AMD side, Vitis AI is a mature software stack designed to accelerate inference on Adaptive SoCs, FPGAs, and Alveo cards. It includes model compilers, runtimes, and profiling tools used by vendors and cloud/edge developers. The Vitis AI EP variant used in Windows is a platform-tailored execution provider that bridges Microsoft’s runtime with AMD’s acceleration primitives.

---

Expected user-facing and developer-visible effects​

Because Microsoft’s KB statements are intentionally high level, the following represent realistic, evidence-based expectations derived from the role of the component, AMD’s Vitis AI responsibilities, and prior component updates:

• Incremental quality gains in image ops. Image scaling (super-resolution), denoising, and foreground/background segmentation used by Windows Photos, Camera, and Studio Effects are the primary surface for visible improvements. Users on qualifying hardware may notice crisper upscales, fewer segmentation artifacts, or smoother live effects. These are consistent with past image-component refreshes.
• Lower inference latency / energy per op. Micro-optimizations in model runtime, improved operator fusion, or better mapping to AMD NPUs can reduce CPU/NPU cycles and memory, yielding slightly faster responses for interactive tasks. Real-world gains are workload-dependent.
• Improved stability and input hardening. Component updates often include input sanitization, better error handling, and hardening against malformed content—important for image parsing chains which historically can be exploited or crash on malformed inputs. The KB’s succinct text does not enumerate CVE fixes; treat security-related claims as unverified until Microsoft publishes explicit CVE mappings.
• Surface area for regressions (driver/stack mismatch). Because the component sits between OS APIs, vendor runtimes, and device drivers, incompatibilities with GPU/NPU drivers or OEM imaging stacks are the most common cause of post-update regressions. Align driver versions before wide deployment.

These expectations align with AMD’s role—Vitis AI provides the kernels, compilers, and execution semantics that the Windows execution provider uses—so any runtime or operator-level change inside the Vitis stack can show up as user-visible improvements in imaging and other local inference tasks.

---

Strengths and strategic implications​

• Faster iteration cadence for on-device AI. Componentized updates let Microsoft iterate models, fixes, and optimizations more rapidly than OS feature releases. That benefits users with more frequent improvements to on-device features.
• Vendor-specific tuning at scale. By shipping AMD-targeted binaries (Vitis AI EP) Microsoft can tune inference paths to the NPU/accelerator characteristics of Ryzen AI and Alveo families—improving efficiency and quality where hardware diversity would otherwise dilute optimizations.
• Local-first privacy and responsiveness. Moving inference on-device reduces cloud transit for many scenarios and improves latency for interactive features—an advantage for privacy-conscious users and offline scenarios when Copilot+ features rely on local models.

---

Risks, unknowns, and practical concerns​

• Opaque public changelogs. Microsoft’s KB language for these component updates is brief and does not disclose line-by-line changes, fixed CVEs, or performance benchmarks. Organizations requiring forensic-level detail should request clarification through official support channels or wait for Security Update Guide entries where CVEs are published. Treat any claim of vulnerability remediation as provisional unless explicitly documented with CVEs.
• Driver/firmware coupling. The most common post-update problems stem from driver mismatches (GPU, NPU, camera ISP drivers). OEM customizations can exacerbate this. It is essential to test with vendor-recommended Adrenalin/driver stacks and to keep BIOS/firmware up to date.
• Version fragmentation. Microsoft ships separate component builds per silicon family (Intel, AMD, Qualcomm, NVIDIA in some stacks). Tracking component versions across a heterogeneous fleet increases operational overhead for enterprises. Maintain an inventory of component versions to correlate behavior with specific releases.
• Unverified performance claims. Any specific numerical performance gain (e.g., “X% faster upscaling”) is not stated by Microsoft in the public KB and must be confirmed with independent benchmarks on representative hardware. Until measured in controlled tests, treat performance gains as plausible but unverified.

---

Recommended deployment checklist (for IT admins and power users)​

• Verify prerequisites:
• Confirm devices are running Windows 11, version 24H2 and have the latest cumulative update (LCU) installed. The component will not install otherwise.
• Inventory Copilot+ endpoints:
• Identify which endpoints are eligible Copilot+ PCs with AMD NPUs (Ryzen AI, Alveo, or vendor NPU-capable SoCs). Microsoft’s Copilot+ criteria and the Windows ML execution provider list are starting references.
• Update vendor drivers in tandem:
• Install the latest AMD chipset, GPU, and NPU/firmware drivers from official vendor channels before and after applying the component update to reduce mismatch risk.
• Create a pilot ring:
• Stage the component to a small representative set of devices (laptops/desktops with varied OEM images) for 7–14 days and monitor imaging workflows, conferencing scenarios, and crash telemetry.
• Acceptance test suite:
• Run scripted checks for Photos app super-resolution/restore features, Camera capture, Windows Studio Effects in Teams/Zoom, Windows Hello biometric paths (if relevant), and any third‑party imaging software critical to workflows. Collect CPU/NPU utilization and latency metrics.
• Rollback plan:
• For managed fleets, ensure procedures to revert to prior known-good cumulative updates or implement mitigations if regressions occur. Note that some SSUs cannot be removed; validate rollback steps in a test environment.

---

How to verify the update on a device​

• After Windows Update applies the component, open Settings → Windows Update → Update history and look for an entry that references the Vitis AI Execution Provider or the component name and version 1.8.25.0. Microsoft’s KB entries instruct users to confirm via Update history.

If the component is not visible or you need package-level detail for compliance, use managed update channels (WSUS / Microsoft Update Catalog), where component packages and package names can be examined more directly by administrators.

---

Developer and power-user notes​

• The Vitis AI EP is integrated into the ONNX/Windows ML execution provider model; developers should validate model compatibility and fallback behavior. Windows ML’s supported execution providers list shows AMD — Vitis AI as an available provider that can be dynamically registered. Applications relying on specific operator behaviors or quantization formats should be validated after the component refresh.
• AMD maintains the upstream Vitis AI project and publishes releases, model libraries, and execution provider code on GitHub and its developer portal. For deeper troubleshooting, consult AMD’s Vitis AI documentation, release notes, and the Vitis-AI open-source repositories to identify changes that might correspond to the Windows-side component change.
• For Ryzen AI platforms, AMD’s release notes and Ryzen AI runtime docs show how Vitis AI EP integrates with Ryzen AI, including session options, ONNX Runtime context caching, and hybrid execution modes—items developers should review when optimizing models for Ryzen AI hardware.

---

When to escalate and what to collect​

If post-update regressions occur (visual artifacts, camera/virtual background failures, application crashes):

• Collect:
• Windows Event logs and crash dumps from affected endpoints.
• Update history entry and component version string.
• GPU/NPU driver and chipset versions.
• Repro steps, screenshots/video of artifact, and timestamps.
• Escalate to:
• OEM or AMD driver support if issues point to driver incompatibility.
• Microsoft support if the problem appears tied to the on-device AI component behavior and cannot be mitigated by driver changes.
• Contacting Microsoft with a full package name and the Update history entry accelerates triage because Microsoft component KBs are intentionally terse and support staff may need the exact package identifier to correlate with internal release notes.

---

Strategic takeaways​

• This KB is part of an ongoing Microsoft strategy to iterate quickly on on-device AI by shipping vendor-specific execution providers and model updates as independent components. That approach delivers faster improvements but increases the operational complexity for heterogeneous fleets.
• For end users on qualifying Copilot+ AMD hardware, the update is likely to be net-positive—delivering modest UX and performance gains in imaging scenarios—provided drivers and firmware are aligned.
• For enterprises, a conservative rollout—pilot, test, then broad deployment—remains the best practice because the most common post-update issues correlate with driver/firmware mismatches and OEM customizations. Maintain an accurate inventory of Copilot+ endpoints and component versions to speed troubleshooting.

---

Final assessment and transparency note​

KB5068002 is a routine but important component update for AMD-targeted Copilot+ devices that updates the Vitis AI Execution Provider to 1.8.25.0. The most notable strengths are Microsoft’s ability to improve on-device AI functionality rapidly and AMD’s continued investment in Vitis AI to support hardware acceleration across product families. The principal limitations are opaque changelogs and the persistent risk of driver/firmware incompatibilities when a component touches cross-stack imaging and inference paths. Administrators and technical users should treat functional claims as plausible but verify them with targeted tests and, if necessary, request more granular change information from Microsoft support when compliance or security needs demand it.
This update represents a small but meaningful step in the broader shift toward modular, hardware-aware AI delivery in Windows—one that benefits users on well-managed Copilot+ hardware and imposes practical vetting responsibilities on IT teams overseeing diverse fleets.

---

Source: Microsoft Support KB5068002: Vitis AI Execution Provider AI component update (1.8.25.0) - Microsoft Support