Optimizing rendering for Mobile

What the talk is about

Vladyslav Aleksandrov explains how mobile GPU architecture differs from PCs and consoles and what that means for rendering. His main point: the vast majority of mobile chips (about 99% of his audience's devices) are tile-based, splitting the screen into small regions held in fast on-chip memory. That makes overdraw and MSAA cheaper, but switching render targets expensive, since data travels through slow system memory. He compares the Mali, Adreno and PowerVR families, their vector and scalar instruction sets, shows results of his own synthetic test, and gives shader advice: medium precision, register usage, vectorization, moving computations into the vertex shader, and caution with transcendental functions. He closes with dynamic geometry handled without buffers, and driver fragmentation on Android. Useful for graphics and rendering programmers working on mobile games.