Windows 11 is getting a new setting to control Unified Memory for AI and graphics

• Microsoft is testing a Unified Memory setting for Windows 11 that reserves memory for graphics and AI acceleration.
• It’s currently hidden in Windows 11 build 29648 and requires a supported device with an NPU, such as a Copilot+ PC.
• The company has not announced a release date or confirmed which Windows 11 version will receive it.

Microsoft is testing a new feature for Windows 11 that could let users decide how much Unified Memory to reserve for graphics and AI workloads. The feature appeared with the release of Windows 11 build 29648 (Feature Platforms), released on August 17, 2026, but there is no confirmed release date for the final version.

The feature is particularly interesting for modern computers where the CPU, GPU, and AI accelerators share the same pool of memory.

What is Unified Memory?

Unified memory is a shared pool of high-bandwidth memory that can be accessed by the CPU, GPU, and other accelerators. That’s different from a traditional computer with a discrete graphics card, which relies on its own dedicated video memory (VRAM) separate from the main system memory.

The approach is becoming more important as processors increasingly combine the CPU, GPU, and AI hardware into a single silicon. It can also be useful for local AI because large models can require more memory than many dedicated laptop GPUs provide.

What is Microsoft testing?

Windows 11 build 29648.1000 (via Windows Latest) contains an unreleased feature called IntelligentCarveout, with ViveTool code of 61121285, as well as a new SettingsHandlers_UnifiedMemory.dll file (via @XenoPanther).

Once the feature is available on the device, Settings > System > Advanced includes a new “Unified memory” section with the “Memory for graphics and AI acceleration” settings.

The setting lets you change the reserved memory for accelerators, with options including “Custom,” “Recommended,” “High,” “Maximum,” and “Don’t allow.”

This doesn’t appear to be simply another way of displaying the existing Shared GPU Memory information in Task Manager. Microsoft appears to be testing an actual memory reservation that would make part of the Unified Memory pool unavailable to normal applications.

How would the setting work?

The idea is to let users reserve more or less Unified Memory for graphics and AI workloads as necessary.

Reserving more memory could help demanding games or local AI applications, while reserving less would leave more memory available to the operating system and other applications.

Microsoft hasn’t officially revealed the final interface, but early bits of the feature show a slider to control the memory manually and preset options (see image above).

Can you enable it now?

Microsoft hasn’t publicly announced the features, but an early preview has already been spotted in the Experimental channel of the Feature Platforms version, which can be exposed using this ViveTool:

vivetool /enable /id:61121285

However, this isn’t a finished feature, and Microsoft warns that experimental features can change or never ship. Also, remember that the Unified Memory feature is available on supported devices equipped with a Neural Processing Unit (NPU), such as Copilot+ PCs.

Why this matters for AI and gaming

Unified memory gives computers a larger pool that can be shared between traditional applications, graphics, and AI workloads. However, that also means those workloads are competing for the same resource.

That’s where Microsoft’s proposed memory reservation could become useful. A user running a large local AI model might want to reserve more memory for the accelerator. Someone primarily using the workstation for productivity might prefer to leave more memory available to the operating system.

Gaming could benefit too, particularly on systems where graphics and AI workloads increasingly run together.

This is also why RTX Spark is an interesting example. NVIDIA is building a Windows platform around a large Unified Memory pool, and the company is already working on changes to system scheduling, memory management, Prism, and hardware power management.

When will it be released?

Microsoft has not announced the feature, so there’s no release date yet.

For now, IntelligentCarveout remains an experimental feature on Windows 11 build 29648. Microsoft has not confirmed which version of the operating system will receive it, or whether it’ll eventually ship to everyone.

What are your thoughts about Microsoft’s Unified Memory setting for Windows 11? Let me know in the comments.