DirectX 11.2 Tiled Resources Memory Features

DirectX 11.2 introduced Tiled Resources, a sparse texturing technology designed to significantly reduce graphics memory consumption while enabling higher-resolution rendering. By decoupling the logical dimensions of a texture from its physical allocation in video RAM, the API allows systems to stream and render only the specific portions of a surface that are currently visible to the camera, drastically lowering overall memory footprints.

Sparse Virtual Allocation

Before DirectX 11.2, allocating a texture required reserving contiguous physical GPU memory for its entire dimension and all associated mip levels. Tiled Resources introduced a virtual memory model for textures, dividing surfaces into standardized 64 KB tiles. Developers can allocate massive logical textures (up to the hardware’s maximum texture dimensions) without committing physical memory to every tile. Regions that are occluded, out of view, or not yet needed consume zero physical VRAM.

Memory Aliasing and Tile Sharing

DirectX 11.2 allows multiple logical tiles within the same resource—or across different resources entirely—to point to a single physical 64 KB tile of GPU memory. This capability, known as tile aliasing, allows developers to map repetitive regions, such as uniform terrain textures, flat skyboxes, or completely transparent areas, to a single shared block of memory. This drastically reduces data duplication in high-resolution textures.

Tier 1 and Tier 2 Capabilities

DirectX 11.2 categorized hardware support for tiled resources into two distinct tiers:

  • Tier 1: Provides fundamental sparse mapping capabilities. In Tier 1, applications must manually manage tile residency and handle texture filtering boundaries carefully, as filtering between resident and non-resident tiles is not natively resolved by hardware.
  • Tier 2: Introduces hardware-level residency management and shader-driven optimizations. Operations accessing unmapped (NULL) tiles automatically return predefined values (such as zero) without crashing or causing visual artifacts. Tier 2 also adds minimum Level of Detail (LOD) clamping, allowing shaders to clamp sampling to resident mip levels dynamically to prevent sampling missing data.

Shader-Driven Residency Feedback

Under Tier 2 implementations, DirectX 11.2 added HLSL shader instructions, such as CheckAccessFullyMapped. This intrinsic allows pixel and compute shaders to determine whether a texture fetch successfully accessed mapped physical memory or hit an unmapped tile. When an unmapped tile is hit, the shader can return feedback to the application's streaming engine, signaling that the tile must be loaded from system memory or disk for upcoming frames. This mechanism ensures that high-resolution data is only loaded on demand, preventing premature out-of-memory errors on constrained hardware.