Microsoft is testing a hidden Windows 11 feature that lets users manually reserve unified memory for graphics and AI-heavy workloads, rather than leaving Windows to manage it dynamically. The discovery, internally called IntelligentCarveout, was spotted in Windows 11 Insider build 29648.1000 and confirmed by Windows Latest, with code strings referencing "reserved memory for accelerators" and "memory for graphics and AI acceleration."
For context, unified memory refers to a shared pool of high-bandwidth memory accessible by the CPU, GPU, and other processors, rather than the dedicated VRAM found on traditional discrete graphics setups. Windows currently manages that pool dynamically, shifting resources between workloads as needed.
That works fine for everyday multitasking, but it can become a problem for heavier workloads such as local AI inference or demanding games, where background tasks and foreground applications compete for the same memory.
- Read more: Check out Bosgame's new monster Mini-PC powered by AMD's new Strix Halo APU, costs $1699
- Read more: New Intel driver lets you dedicate 93% of system memory to the iGPU for VRAM, enabling support for larger AI models
- Read more: ASUS ProArt P16 and P14 are the first ASUS laptops built around NVIDIA's RTX Spark superchip
The new feature would let users reserve a portion of unified memory specifically for graphics and AI, keeping it off-limits to other workloads so those tasks get priority. Reports suggest it could eventually live under Settings > System > Advanced, possibly as a slider or a set of preset profiles.
The timing is particularly interesting given NVIDIA's upcoming RTX Spark platform, which pairs an Arm CPU with Blackwell graphics and supports up to 128GB of unified memory. Microsoft confirmed in May that it had already adjusted Windows to allow RTX Spark's GPU to access more of that shared pool.
Since that shared memory uses LPDDR5X rather than the faster GDDR7 found on dedicated GPUs, the advantage is primarily capacity rather than raw bandwidth. In theory, that could give an RTX Spark GPU access to far more memory than even NVIDIA's RTX 5090, although at considerably lower memory bandwidth.
AMD already offers something similar through its own software. Variable Graphics Memory in AMD's Adrenalin drivers lets Ryzen AI Max users manually adjust how much of their system's unified memory is allocated to the integrated GPU. A 128GB Ryzen AI Max+ 395 system, for example, can reserve up to 96GB for graphics.
Frequently Asked Questions
Open a question for an answer from TweakTown's coverage of this news, or ask your own below.
What is IntelligentCarveout and how does it change Windows 11's handling of unified memory?
Which Windows 11 Insider build includes the IntelligentCarveout strings and where is the feature being tested?
Where in Settings is Microsoft reportedly planning to place the memory reservation control (slider or presets)?
How does Windows' unified memory (LPDDR5X) compare to dedicated GPU memory like GDDR7 in bandwidth and capacity?
Have a question about this content not listed here? Ask below and TweakBot will answer it.
That said, the Windows feature is still confined to Microsoft's Experimental Future Platforms channel, a testing branch the company uses well before changes reach general Windows builds. It also isn't accessible through the normal Settings interface yet.
