GPU Pro 4

2. Introducing the Programmable Vertex Pulling Rendering Pipeline 31

256,000128,00064,00032,00016,000

x-axis: Number of Draws

y-axis: Processing Time in ms

8,0004,0002,000

Discarded

Rendered

1,000

0

20

40

60

80

100

Figure 2.9. Comparing performance between discarded draws and draws rendering a

single quad per draw producing 16 pixels on the framebuffer.

indirect buffer. For this chapter we used a brute force OpenCL kernel to check

every object in the scene, but nothing prevents us from using space partitioning

techniques to continue scaling up the scene complexity.

Taking a closer look at the new draw call functions, it is both interest-

ing and annoying to acknowledge ...