Windows 11 build 29648.1000, released on 17 August, contains a hidden feature that adds dedicated Windows 11 unified memory controls for chips where the CPU and GPU draw from one shared pool of RAM. The feature is code-named IntelligentCarveout and was first identified by TechPowerUp inside the build’s code strings.
The target hardware is a class of SoC that fuses CPU and GPU cores onto one package and connects them to external LPDDR5X memory instead of giving the GPU its own VRAM. Nvidia’s upcoming RTX Spark platform works this way, as does AMD’s Ryzen Halo, and Microsoft’s build strings suggest the same controls apply to other APU-style designs built on the pattern.
How Windows 11’s unified memory limit works today
Windows currently applies one blanket ceiling to how much of a unified pool the GPU can claim, regardless of what else is running. IntelligentCarveout replaces that single limit with settings for independent partitions of the same memory, so the system can carve out separate allocations rather than treating the whole pool as one shared bucket.
Microsoft previewed the direction at Computex, writing in its RTX Spark announcement that “to realize the potential of up to 128 GB of unified memory on RTX Spark, we have focused on improving how Windows supports unified memory systems, starting with a new higher, smarter limit on total system memory accessible by the GPU.” That change, the company said, “increases the memory available to the GPU on high-memory systems, unlocking the ability to load larger local AI models or render more complex projects.” The 128GB figure is Microsoft’s own, tied specifically to RTX Spark configurations. Read Microsoft’s announcement.
A vendor feature or a broader Windows control
The two outlets that spotted IntelligentCarveout frame it differently, and the gap is worth noting. TechPowerUp’s coverage leads with the feature arriving ahead of the RTX Spark launch, tying it to Nvidia’s hardware specifically. Guru3D describes the same code as user-controlled unified memory reservations for GPUs and AI, without naming a chip at all.
Read against Microsoft’s own wording, that second framing is the more accurate one. IntelligentCarveout lives in Windows itself, not in an Nvidia driver, and Microsoft’s description of the change, a smarter limit on system memory accessible by the GPU, doesn’t mention any single vendor. That would put AMD’s Ryzen Halo chips in scope alongside RTX Spark, since both share the same CPU-GPU-LPDDR5X layout the feature is built around.
What to watch next
IntelligentCarveout is still a hidden string in a build, not a shipped setting, so there’s no confirmed date for when Microsoft turns it on for users. The two things worth tracking are whether a future Insider build exposes the partition controls directly in Settings, and whether Microsoft ties the public rollout to the RTX Spark launch specifically or ships it as a general Windows 11 update available to any unified-memory chip.
