Table of render speed per model per screen size.
|
Key:
128x128
256x256
512x512
1024x1024
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Observations:
The GPU used to render these models was an ATI mobile FireGL 5200. Seing these results, I am a little surprised that rendering in Vertex Array mode without Virtex Buffer Objects (VBOs) only doubles the framerate at best. It is also interesting to see that for the two faster rendering modes that there wasn't a real performance hit until the window resolution was 1024x1024. The most important piece of information from this analysis is that one should render using Display Lists whenever possible, and VBOs when it is not.
To find out when my GPU became geometry limited, I rendered the dragon model at 250k verticies at a resolution of 512x512, and judged that it became geometry limited when there was a severe drop in framerate, or when the framerate dropped to the same speed as Direct Mode rendering of this model. I found that this occurred when I was drawing the model 7 times, which is approximately 1.75M triangles.
|
|
|
|
|
|