Fast mobile shaders or rather Mobile! Fast!

It’s very hard (i.e. practically impossible) to get GPU timing for a draw call on TBDR, makingprofiling hard. Often easier to profile on Tegra or Adreno.Tiled Deferred Caveats••Scene splitsMore transistors for smarts• Less “raw power”•No GPU time per draw call• Easier to profile on non-TBDR: Tegra orAdrenoSo it seems that TBDR architecture is all awesome, but of course it has some drawbacks aswell.If vertex data per frame (number of vertices * storage required after vertex shader) exceedsthe internal buffers allocated by the driver, the scene has to be “split” which costsperformance. The driver might allocate a larger buffer after this point, or you might need toreduce your vertex count. We ran into this on iPad2 (iOS 4.3) at around 100 thousand verticeswith quite complex shaders.TBDR needs more transistors allocated for the tiling & deferred parts, leaving conceptuallyless transistors for “raw performance”.