Microsoft is quietly testing a new Windows 11 feature that would give users direct control over how much unified memory gets reserved for graphics and AI workloads, a capability that could become essential as Nvidia prepares to ship its first RTX Spark laptops later this year.
The feature, internally called “IntelligentCarveout,” surfaced in Windows 11 build 29648.1000 released on August 17. It is not mentioned anywhere in Microsoft’s official patch notes, but code strings discovered in the experimental build reference a new system component labeled “SettingsHandlers_UnifiedMemory.dll,” according to Windows Latest, which first reported the finding with confirmation from Windows watcher @phantomofearth.
The description embedded in the build reads: “Let Windows reserve additional unified memory for graphics and AI intensive games and applications. Reserved memory is not available for other applications.” In practice, that means users would be able to manually set aside a portion of system RAM for GPU-accelerated tasks, with the rest of the system locked out from touching that allocation.
Similar references have previously appeared in Nvidia documentation, suggesting the feature is being developed in coordination with the GPU maker ahead of the RTX Spark launch. The chip, unveiled earlier this year, supports up to 128GB of LPDDR5X unified memory shared between the CPU and its integrated Blackwell GPU.
The timing is hardly coincidental. Microsoft has already released a Windows 11 developer driver for RTX Spark, a strong signal that notebooks powered by the new SoC are expected to arrive before the end of 2025. When Nvidia announced the platform, it emphasized the ability to run 120-billion-parameter AI models locally, which would require reserving a massive share of that 128GB pool for GPU memory while leaving a minimum of roughly 8GB for system tasks.
Unified memory architectures are not new to Windows. AMD’s Ryzen AI Max+ 395 processor, which also supports up to 128GB of memory, already offers a similar capability through its graphics drivers, allowing users to reserve up to 96GB for the GPU. What makes IntelligentCarveout notable is that it would bring memory allocation controls directly into Windows 11’s settings interface rather than leaving the task to vendor-specific drivers.
The broader industry context explains why Microsoft is investing in this area. Apple’s Mac lineup has long used unified memory, and the local AI boom has driven demand for Macs configured with 24GB or more of unified RAM to the point of shortages. Qualcomm and Intel are also building AI-focused PCs around shared memory pools that can stretch into the hundreds of gigabytes.
For Nvidia, the software story matters as much as the silicon. The company’s dominance in discrete GPUs has been built on more than raw hardware performance. CUDA programming, DLSS upscaling, Reflex latency reduction, and the broader RTX software ecosystem have become genuine selling points. Extending that same stack into the integrated-graphics space gives RTX Spark a built-in advantage over AMD, Qualcomm, and Intel, none of which can yet match Nvidia’s software ecosystem for AI or gaming performance.
IntelligentCarveout would amplify that advantage. If users can route a large share of the 128GB pool toward graphics and AI tasks, the RTX Spark’s GPU could end up with access to far more memory than even the RTX 5090’s 32GB of dedicated GDDR7 VRAM. There is an important caveat, however: LPDDR5X memory is considerably slower than GDDR7, so the advantage is one of capacity rather than raw bandwidth.
Microsoft has not publicly commented on IntelligentCarveout, and it remains unclear whether the feature will ship in the next major Windows 11 update or arrive later. What the experimental build makes plain is that the company is preparing its operating system for a wave of unified-memory devices that will blur the traditional line between system RAM and dedicated video memory.
For users, the shift could fundamentally change how they configure performance on these systems. Instead of Windows deciding autonomously how much memory to allocate to graphics, gamers and AI enthusiasts would gain a direct dial, letting them prioritize one workload over another depending on what they are running at any given moment.