Mobile GPUs and Tile-Based Rendering
6 days ago
- #tile-based-rendering
- #mobile-gpu
- #graphics-architecture
- Mobile GPUs use tile-based deferred rendering (TBDR) to address power and bandwidth constraints.
- Desktop GPUs traditionally use immediate mode rendering (IMR), which requires high memory bandwidth.
- TBDR splits rendering into tiling (geometry processing) and rendering (fragment processing) phases.
- Tiling phase divides the screen into tiles and associates geometry with each tile (binning).
- Rendering phase processes each tile independently, keeping data in on-chip memory to reduce bandwidth.
- Apple's AGX architecture, inspired by PowerVR, implements TBDR with unique optimizations.
- AGX handles sample shading and blending differently than desktop GPUs, favoring software flexibility.
- Vulkan API is optimized for TBDR with features like render passes, subpasses, and transient attachments.
- TBDR architectures influence software design, favoring moderate geometry complexity and bandwidth-conscious algorithms.
- Future trends include scaling TBDR for ray tracing and compute, with desktop GPUs adopting tile-based optimizations.