Windows 11 may let you control AI vs graphics memory
Hidden feature ‘IntelligentCarveout’ targets unified-memory PCs like Nvidia's RTX Spark
PCs without a dedicated GPU already share a single memory pool between graphics, AI, and general system tasks, with Windows currently making that split entirely on its own.
A hidden feature spotted in a recent experimental Windows 11 build suggests that's about to change, as Microsoft prepares to hand users direct control over how unified memory gets divided.
Windows 11 build 29648.1000, which was launched on August 17th, introduces an unknown feature, known as 'IntelligentCarveout', which has been given feature ID 61121285, along with a new DLL file named 'SettingsHandlers_UnifiedMemory.dll'. It has not been mentioned in Microsoft's official list of updates.
A Windows Insider, phantomofearth, has told Windows Latest that the build code strings mention allocating memory exclusively for accelerators, graphic processes, and AI, while other information gathered by Pureinfotech reveals the location of the setting to be set in the future. Settings > System > Advanced > Unified memory.
Based on strings found in the build, the setting would let users choose between preset levels, including "Custom", "Recommended", "High", "Maximum", and "Don't allow", to reserve a portion of unified memory exclusively for graphics and AI-intensive applications.
This section would be inaccessible to everything else on the machine, ensuring that games and AI processes could use the memory without having to compete with the background processes for it.
The idea stands apart from what Task Manager's current display of shared GPU memory is since it looks like a protected memory allocation rather than the current display of post-hoc information.
RTX Spark is the name given by Nvidia to the technology of using a GPU and CPU combination, where the GPU in question is a Blackwell GPU, while the CPU is a Grace Arm CPU and can have up to 128GB of shared memory.
On these systems, letting users manually route a larger share of that pool toward graphics or AI tasks could meaningfully change performance tuning, though it comes with an obvious trade-off: a heavily reserved allocation could make everyday multitasking far less forgiving when several large applications are open at once.
