Microsoft Tests IntelligentCarveout in Windows 11 to Manage Unified Memory Allocation.

Microsoft Tests "IntelligentCarveout" Feature in Windows 11 to Customize Unified Memory Allocation

Microsoft is testing an advanced system memory management feature in preview builds of Windows 11, enabling users to manually allocate dedicated portions of Unified Memory specifically for graphics, gaming, and local AI workloads. Discovered within recent Windows 11 Insider preview builds under the internal codename IntelligentCarveout, the feature integrates direct system controls into the native Settings app to optimize performance on unified memory architectures.

Granular Allocation Mechanics and System Settings Integration

The proposed system menu allows users to reserve dedicated memory pools that general background applications cannot access or pre-empt:

• Dedicated Hardware Reserves: Provides direct controls to reserve specific blocks of system RAM exclusively for GPU rendering and NPU/AI acceleration, preventing general desktop applications from consuming critical VRAM pools.

• Targeted Hardware Architecture: Designed specifically for systems utilizing unified memory designs—where the Central Processing Unit (CPU), Graphics Processing Unit (GPU), and Neural Processing Unit (NPU) draw from a single shared pool of physical memory, unlike traditional desktop PCs with dedicated VRAM.

• System Settings Path: Preliminary build data indicates the interface is located under Settings > System > Advanced > Unified Memory, offering preset allocation profiles along with a custom slider control.

• Operational Trade-Offs: Allocating larger memory blocks for GPU and AI execution proportionally reduces the remaining system RAM available for the Windows operating system and background software. For instance, on a 64GB unified memory configuration, a user could lock 32GB exclusively for local AI model inference, leaving the remaining 32GB for system tasks.

Functional Purpose and Feature Status

Microsoft emphasizes that memory carving reorganizes existing hardware resources rather than expanding capacity:

• Allocation vs. Expansion: The feature does not convert standard system RAM into discrete VRAM for dedicated graphics cards, nor does it increase physical memory capacity; it simply locks existing unified RAM to prioritized computing engines.

• Target Workloads: Reserving static memory blocks enhances system stability when running memory-intensive local Large Language Models (LLMs), Generative AI diffusion models, and high-resolution gaming titles.

• Deployment Timeline: Microsoft has not officially announced target device compatibility lists or a final public release schedule, noting that menu interfaces and feature names remain subject to change before general availability.

Running large language models locally requires significant VRAM to keep model weights in active memory. Without dedicated memory reservation, operating system background processes can fragment available RAM, leading to operational failures or severe latency in token generation. Allowing users to lock static memory blocks ensures that large parameters remain cached in high-speed unified RAM without interruption.

Unified memory architectures—popularized by high-performance silicon such as the Qualcomm Snapdragon X platform, Apple M-series chips, and AMD APUs—blur the line between system RAM and traditional video memory. Native OS-level memory partitioning enables Windows 11 to better leverage these integrated chipsets, offering desktop-level control over shared memory architectures.

While locking 50% or more of system RAM for GPU/AI tasks boosts performance for demanding applications, it increases the risk of system memory shortages for general multitasking. Windows 11 requires robust dynamic safeguards to prevent critical system processes from running out of non-reserved RAM, ensuring the desktop interface remains responsive even under heavy AI processing loads.

 

 

Source: neowin 

💬 AI Content Assistant

Ask me anything about this article. No data is stored for your question.

Microsoft Tests "IntelligentCarveout" Feature in Windows 11 to Customize Unified Memory Allocation

Microsoft is testing an advanced system memory management feature in preview builds of Windows 11, enabling users to manually allocate dedicated portions of Unified Memory specifically for graphics, gaming, and local AI workloads. Discovered within recent Windows 11 Insider preview builds under the internal codename IntelligentCarveout, the feature integrates direct system controls into the native Settings app to optimize performance on unified memory architectures.

Granular Allocation Mechanics and System Settings Integration

The proposed system menu allows users to reserve dedicated memory pools that general background applications cannot access or pre-empt:

• Dedicated Hardware Reserves: Provides direct controls to reserve specific blocks of system RAM exclusively for GPU rendering and NPU/AI acceleration, preventing general desktop applications from consuming critical VRAM pools.

• Targeted Hardware Architecture: Designed specifically for systems utilizing unified memory designs—where the Central Processing Unit (CPU), Graphics Processing Unit (GPU), and Neural Processing Unit (NPU) draw from a single shared pool of physical memory, unlike traditional desktop PCs with dedicated VRAM.

• System Settings Path: Preliminary build data indicates the interface is located under Settings > System > Advanced > Unified Memory, offering preset allocation profiles along with a custom slider control.

• Operational Trade-Offs: Allocating larger memory blocks for GPU and AI execution proportionally reduces the remaining system RAM available for the Windows operating system and background software. For instance, on a 64GB unified memory configuration, a user could lock 32GB exclusively for local AI model inference, leaving the remaining 32GB for system tasks.

Functional Purpose and Feature Status

Microsoft emphasizes that memory carving reorganizes existing hardware resources rather than expanding capacity:

• Allocation vs. Expansion: The feature does not convert standard system RAM into discrete VRAM for dedicated graphics cards, nor does it increase physical memory capacity; it simply locks existing unified RAM to prioritized computing engines.

• Target Workloads: Reserving static memory blocks enhances system stability when running memory-intensive local Large Language Models (LLMs), Generative AI diffusion models, and high-resolution gaming titles.

• Deployment Timeline: Microsoft has not officially announced target device compatibility lists or a final public release schedule, noting that menu interfaces and feature names remain subject to change before general availability.

Running large language models locally requires significant VRAM to keep model weights in active memory. Without dedicated memory reservation, operating system background processes can fragment available RAM, leading to operational failures or severe latency in token generation. Allowing users to lock static memory blocks ensures that large parameters remain cached in high-speed unified RAM without interruption.

Unified memory architectures—popularized by high-performance silicon such as the Qualcomm Snapdragon X platform, Apple M-series chips, and AMD APUs—blur the line between system RAM and traditional video memory. Native OS-level memory partitioning enables Windows 11 to better leverage these integrated chipsets, offering desktop-level control over shared memory architectures.

While locking 50% or more of system RAM for GPU/AI tasks boosts performance for demanding applications, it increases the risk of system memory shortages for general multitasking. Windows 11 requires robust dynamic safeguards to prevent critical system processes from running out of non-reserved RAM, ensuring the desktop interface remains responsive even under heavy AI processing loads.

 

 

Source: neowin