Windows 11 prepares unified memory controls ahead of NVIDIA RTX Spark PCs
VideoCardz reports that Microsoft is testing a new Windows 11 control for unified-memory systems. The experimental feature, internally called IntelligentCarveout, would let users reserve part of the shared memory pool for graphics and AI acceleration. Memory assigned to that reservation would no longer be available to ordinary Windows applications, giving users a direct way to prioritize workloads that depend on GPU or AI resources.
The setting was spotted in Windows 11 build 29648.1000, released on August 17 through Microsoft's Experimental Future Platforms channel, after Windows Latest found references to it in NVIDIA documentation. The unfinished interface includes separate controls for memory reserved for accelerators and for graphics and AI acceleration, but it is still hidden and cannot be enabled through the normal Settings app. Microsoft warns that features in this channel can change or disappear before reaching a retail Windows release. The work follows Microsoft's earlier RTX Spark optimization effort, which increased the amount of system memory available to the platform's GPU so larger local AI models and more complex creative projects could fit in memory.
RTX Spark is a likely early beneficiary because NVIDIA's Arm-based processors combine Blackwell graphics and AI acceleration with up to 128 GB of unified memory. AMD already offers a comparable user-facing control through Variable Graphics Memory on Ryzen AI Max systems; a 128 GB Ryzen AI Max+ 395 machine can reserve up to 96 GB for graphics. For Canadian buyers considering RTX Spark PCs this fall, the proposed Windows control could make the large shared memory pool more practical by allowing creators, developers, and AI users to tune the balance between GPU capacity and general-purpose applications. It remains a preview feature, however, and should not be treated as a confirmed launch capability.
Source: VideoCardz.com. This article summarizes third-party reporting. Follow the source link for the full original article.