Microsoft is preparing Windows 11 to give users direct control over how much unified memory their PC reserves for AI and graphics workloads, a feature discovered in recent preview builds. This development arrives as NVIDIA's RTX Spark processors, which combine CPU, GPU, and AI acceleration on a single chip with shared memory, begin shipping in laptops from major manufacturers this fall.

What Is Unified Memory and Why Does It Matter?

Unified memory represents a fundamental shift in how computers manage processing power. Instead of splitting memory into separate buckets for the CPU (system RAM) and GPU (dedicated video memory), unified memory creates a single, high-bandwidth pool shared across all processors. Apple popularized this approach with its M-series chips, and now Windows PCs are catching up.

The practical benefit is significant: a laptop with 128GB of unified memory can run large AI models locally, handle demanding graphics tasks, and perform everyday computing all without the usual memory bottlenecks. However, this flexibility creates a new problem. Running a large language model might need 80GB of memory reserved, leaving only 48GB for everything else. Gaming might need the opposite allocation. Windows needs a way to let users decide.

How Will Windows 11 Let You Control Unified Memory?

• IntelligentCarveout Feature: Microsoft is testing a hidden feature called IntelligentCarveout in Windows 11 build 29648.1000, released August 17, which will allow users to reserve specific amounts of unified memory for graphics and AI workloads.
• Reserved Memory Allocation: Unlike the current "Shared GPU Memory" shown in Task Manager, this new feature creates a hard reservation, meaning memory set aside for AI or graphics will not be available to other applications.
• Settings Location: Evidence in the build suggests the control will appear under Settings > System > Advanced, though Microsoft has not yet revealed whether it will be a slider, fixed presets, or workload profiles.

The new feature includes a file called SettingsHandlers_UnifiedMemory.dll and contains strings stating: "Let Windows reserve additional unified memory for graphics and AI intensive games and applications. Reserved memory is not available for other applications".

Why Does This Matter for RTX Spark and Beyond?

NVIDIA's RTX Spark represents the first mainstream Windows laptop processor built around unified memory at scale. The chip packs a 20-core Grace Arm CPU, a Blackwell GPU with up to 6,144 CUDA cores (specialized processors for AI and graphics), and up to 128GB of unified memory onto a single processor. This is a departure from traditional laptop design, where a separate GPU with its own memory sits alongside the CPU.

RTX Spark laptops and mini PCs are scheduled to ship this fall from Microsoft, Lenovo, ASUS, Dell, HP, and MSI, with Acer and GIGABYTE to follow. Microsoft is also preparing the Surface Laptop Ultra, its RTX Spark answer to the MacBook Pro, and a developer-focused Surface RTX Spark Dev Box with 128GB of unified memory.

Without memory carve-out controls, users would face constant trade-offs. Reserve too much for AI, and gaming performance suffers. Reserve too much for graphics, and running a local AI model becomes impossible. The new Windows feature solves this by letting users adjust allocation based on their current task.

How Is Windows Preparing for RTX Spark Beyond Memory Control?

Microsoft is making deeper changes to Windows 11 specifically for RTX Spark's unique architecture. The company is testing new scheduling, memory management, Prism tuning, and power management under the name Workload Profile Scheduling. Regular Windows scheduling was not designed for a chip that combines CPU, GPU, and AI acceleration on a single processor with shared memory.

NVIDIA CEO Jensen Huang stated that RTX Spark runs every Windows app NVIDIA and Microsoft have ever built, through native Arm64 support and Prism emulation for x86 software. NVIDIA has already shipped a developer-preview GeForce driver, version 616.00, for Windows 11 on Arm, confirming two RTX Spark configurations with 6,144 and 5,120 CUDA cores respectively.

Leaked Geekbench 7 scores for the RTX Spark N1X chip already beat AMD's Ryzen AI Max+ 395 and Intel's Core Ultra X9 388H in multi-core performance, on hardware that is not yet final.

What About Apple's Approach to Unified Memory?

Apple does not give MacBook owners a direct way to adjust how much unified memory is reserved for graphics or AI workloads. Users get a fixed allocation determined by the operating system. While this simplicity works for Apple's controlled ecosystem, Windows faces a different challenge: the operating system must support vastly different workloads and hardware configurations.

A YouTuber recently praised a MacBook's 128GB of unified memory while criticizing a Windows gaming laptop's RTX 5090 for having "only" 24GB of dedicated video memory. However, this comparison misses the point. The RTX 5090 is built specifically for gaming with high-bandwidth memory optimized for that task. RTX Spark, by contrast, is designed to handle gaming, local AI models, and creative work all on the same chip, which is why memory carve-out controls are essential.

When Will This Feature Become Available?

The IntelligentCarveout feature is currently experimental and hidden in Windows 11 preview builds. Microsoft has not announced an official release date or confirmed the feature will ship in a public build. However, the presence of the feature in recent builds, combined with RTX Spark's fall launch, suggests Microsoft is working to have memory controls ready when the new hardware arrives.

Users on the Experimental build can attempt to enable the feature using ViVeTool with feature ID 61121285, though this is not recommended for non-technical users, as it may cause system instability.

The broader story here is that Windows is evolving to support a new class of hardware that blurs the lines between CPU, GPU, and AI accelerator. As unified memory becomes more common in consumer laptops, giving users control over how that memory is allocated will become a standard expectation, not a novelty feature.