Windows 11's IntelligentCarveout Lets You Partition Unified Memory for AI and Gaming

Microsoft is testing a feature called IntelligentCarveout in Windows 11 build 29648 that would let users manually partition unified memory between system, graphics, and AI workloads. If it ships publicly, it could change how builders configure local AI inference on unified-memory PCs like Nvidia's RTX Spark, which offers up to 128GB of shared memory.

What IntelligentCarveout Does

Windows 11 build 29648, released August 17, carries code strings referencing reserving memory for accelerators, graphics, and AI. Windows watcher @phantomofearth confirmed the feature to Windows Latest. Early builds show a slider for manual control alongside preset options named Custom, Recommended, High, Maximum, and Don't allow. The description string says: "Let Windows reserve additional unified memory for graphics and AI intensive games and applications. Reserved memory is not available for other applications."

The feature is not mentioned in Microsoft's official patch notes. Similar references previously appeared in Nvidia documentation, suggesting alignment with unified-memory hardware like RTX Spark.

Why It Matters for AI Builders

Unified memory pools share RAM between system, graphics, and AI tasks. Without manual control, Windows decides the split automatically, which can cause contention when running both a game and a local AI model. IntelligentCarveout would let builders reserve a portion of the pool for AI inference, potentially reducing performance penalties from memory pressure.

On RTX Spark devices, the GPU could end up with access to far more memory than even the RTX 5090's 32GB of dedicated VRAM. However, the shared pool uses slower LPDDR5x memory compared to GDDR7, so the advantage is capacity rather than speed. Nvidia's software stack (CUDA, DLSS, Reflex) gives it a built-in edge over AMD, Qualcomm, and Intel on unified-memory AI and gaming performance.

Practical Implications and Trade-offs

If IntelligentCarveout ships, users would get a settings slider or presets to reserve unified memory for graphics and AI. The reserved portion would be off-limits to other applications. This gives builders direct control over memory allocation for local AI models on unified-memory PCs.

The practical value depends on the final implementation. Early builds suggest a slider and presets, but no official UI or release date has been confirmed. Developers targeting Windows 11 unified memory platforms should plan for potential new APIs or UX patterns around memory stewardship.

What's Still Unknown

All evidence comes from an experimental Windows 11 build. There is no official release date or confirmation of public rollout. Details may change as Microsoft tests and finalizes the feature. Claims about RTX Spark's advantages rely on Nvidia's broader software ecosystem and the assumption of public rollout. Performance numbers and exact memory allocations are not provided in the source material.

For now, IntelligentCarveout is a promising signal that Microsoft is thinking about memory contention on unified-memory devices. If it ships, it could be a useful tool for builders running local AI models alongside other workloads. But it's still experimental, and the real impact will depend on how well the implementation handles the capacity-versus-speed trade-off.