Microsoft began rolling out an automatic Windows 11 update this week that upgrades the Nvidia TensorRT-RTX Execution Provider to version 1.8.22.0, promising improved performance for on-device AI workloads on consumer RTX graphics cards. The update, identified as KB5077528, replaces an older runtime and targets Windows 11 versions 24H2 and 25H2. While the change may seem quiet, it quietly reshapes how AI models run locally on millions of PCs.

What KB5077528 Actually Delivers

The patch is a component update delivered through Windows Update, not a traditional driver or user-installed package. According to Microsoft’s brief advisory, it “includes improvements to the execution provider component” and positions the Nvidia TensorRT-RTX Execution Provider as the preferred GPU backend for consumer RTX hardware—officially superseding the older KB5068004 parcel, which carried version 1.8.14.0.

Key facts from the KB article:

  • Version: 1.8.22.0 of the TensorRT-RTX Execution Provider
  • Prerequisites: The device must already have the latest cumulative update (LCU) for Windows 11, version 24H2 or 25H2
  • Delivery: Automatic via Windows Update; no manual intervention required
  • Visibility: After install, it appears under Settings → Windows Update → Update history as “Windows ML Runtime Nvidia TensorRT-RTX Execution Provider Update (KB5077528)”

Microsoft’s note is intentionally sparse—no detailed changelog, benchmark data, or security impact statement. That’s common for silicon-specific runtime components, but it does place the onus of validation on power users and admins.

Why This Update Matters to You

The TensorRT-RTX Execution Provider is a specialized plugin for ONNX Runtime and Windows ML that optimizes AI model inference on Nvidia’s consumer GPUs (GeForce RTX 30 series and newer). It replaces the broader CUDA Execution Provider with a smaller, faster-loading runtime purpose-built for RTX silicon. For end users, that translates to apps and system features that use local AI—think Copilot-like assistants, photo editing tools, or background processing—feeling snappier with less latency.

Home Users

If you have an RTX graphics card and keep Windows Update on automatic, KB5077528 will install itself once your system meets the LCU requirement. No action is necessary. You may notice modest improvements in AI-enabled applications, though the difference will be subtle in everyday use. The main benefit is laying the groundwork for future AI features that depend on a fast, lightweight inference engine.

Power Users and Enthusiasts

For those who tinker with local AI models, experiment with Stable Diffusion, or run tools like ONNX-based upscalers, this update is worth verifying actively. A post-install checklist:

  • Confirm the update appears in Update history.
  • Ensure your Nvidia driver is recent—while the KB doesn’t specify a minimum, ONNX Runtime build documentation suggests driver 555.85 or newer for full TensorRT-RTX support, so updating to the latest Game Ready or Studio driver is prudent.
  • Test your most-used models before and after the update; measure first-inference latency and sustained throughput. Many users report faster engine compilation and shorter load times with TensorRT-RTX.

Developers

KB5077528 formalizes TensorRT-RTX as the default EP for RTX-class consumer devices. If you ship Windows apps that use ONNX Runtime, you should:

  • Validate your models against TensorRT-RTX EP, using CUDA EP as a fallback for unsupported operators.
  • Leverage ONNX Runtime’s compilation APIs to generate EP context models, which exploit ahead-of-time (AOT) compilation to nearly eliminate first-run delays.
  • Update your build environments: ONNX Runtime’s TensorRT-RTX provider requires CUDA 12.9 and driver 555.85 or above for compilation. Your CI pipelines and dev machines should align with these versions.

IT Administrators

Because this is an OS component update, it’s manageable through standard deployment tools (WSUS, Windows Update for Business, Microsoft Intune). Organizations with strict update policies should:

  • Test KB5077528 in a staging environment alongside line-of-business apps that use local AI or Windows ML.
  • Request detailed change information from Microsoft or Nvidia if security or stability is a concern—the public KB lacks CVE mappings.
  • Plan a phased rollout, blocking automatic installation if necessary until validation is complete. Rollback options include uninstalling via Update history or restoring a system checkpoint.

How We Got Here: The Evolution of AI Runtimes on Windows

Windows ML and ONNX Runtime rely on Execution Providers to target different hardware backends. The CUDA EP has long been the go-to for Nvidia GPUs, while the legacy TensorRT EP catered to datacenter workflows. Nvidia introduced TensorRT-RTX to bridge the gap for consumers, delivering a runtime with a fraction of the disk footprint (around 200 MB vs much larger datacenter packages) and faster engine compilation through JIT and AOT caching.

Microsoft first shipped TensorRT-RTX via KB5068004 (version 1.8.14.0) earlier this year, making it an official Windows component. The new KB5077528 refines that foundation, signaling Microsoft’s commitment to integrating AI runtime updates directly into the OS rather than relying on third-party installers. This approach ensures that even casual users get performance enhancements automatically, while developers gain a predictable target platform.

What To Do Right Now

Here’s a concise action plan to make the most of this update:

  1. Check installation: Go to Settings > Windows Update > Update history and look for KB5077528. If it’s missing, manually trigger a scan for updates after confirming your system has the latest LCU.
  2. Update your GPU driver: Download the newest driver from Nvidia’s website or GeForce Experience. The recommended minimum from ONNX Runtime is 555.85, but newer drivers generally improve compatibility and performance.
  3. Test AI workloads (power users/devs): Run a representative set of models—vision transformers, text encoders, or quantized LLMs—and capture metrics like load time, memory usage, and inference speed. Compare against previous results to quantify the benefit.
  4. Fallback plan (if issues arise): Uninstall the update from Update history, or use System Restore. For enterprise-managed devices, block deployment through your update management platform until you’ve completed internal testing.
  5. Provide feedback: If you encounter regressions, report them via the Feedback Hub or directly to Microsoft/Nvidia channels. The sparse KB documentation makes community reports vital for identifying edge cases.

Outlook: The Quiet Expansion of AI on Windows

KB5077528 is a small but telling piece of a larger picture. Microsoft is steadily embedding AI runtimes into the OS fabric, with Windows Update as the delivery vehicle. Nvidia’s TensorRT-RTX, Intel’s OpenVINO EP, and AMD’s ROCm-based providers are all vying for slots in Windows ML. Expect more silent updates like this one, each tightening the integration between hardware and the AI features that increasingly define the user experience. For now, RTX owners can enjoy a faster, more efficient local AI engine—one that requires no extra clicks to activate.