GPU Pro 6

1. Compute-Based Tiled Culling 455

Unreal Engine 4 Infiltrator Real-Time Demo

2.6

2.4

2.2

2.0

Reduction in GPU Execution Time (ms)

1.8

1.6

1.4

1.2

1.0

0.8

0.6

Tiled Deferred: Optimized vs. Baseline

Figure 1.10. Unreal Engine 4 tiled-culling execution time improvement for the optimized

version compared to the baseline implementation. Performance was measured over the

entire Infiltrator demo at 1920 × 1080.

of calculating the depth bounds and performing the per-tile culling. However, av-

eraged over the entire demo, tiled deferred is still faster overall. Specifically, the

average cost of standard deferred is 4.28 ms, whereas the optimized tiled-deferred

average cost ...