Assignment 2 -- Due 2/14/08

Assignment 1 -- Due 2/14/08

Jonathan Hall

Problem 1:

Investigate the effects of different render modes.

Model	Vertices \ Resolution	128x128	256x256	512x512	1024x1024
Buddha	50,000	179 fps 1360 fps 247 fps 1390 fps	179 fps 1360 fps 247 fps 1390 fps	179 fps 1008 fps 247 fps 934 fps	179 fps 368 fps 247 fps 377 fps	Immediate Display List Vertex Array VBO
	250,000	39 fps 793 fps 56 fps 524 fps	39 fps 720 fps 56 fps 502 fps	39 fps 512 fps 56 fps 400 fps	39 fps 270 fps 56 fps 238 fps	Immediate Display List Vertex Array VBO
	1,087,000	8 fps 213 fps 13 fps 123 fps	8 fps 207 fps 13 fps 122 fps	8 fps 184 fps 13 fps 116 fps	8 fps 139 fps 13 fps 97 fps	Immediate Display List Vertex Array VBO
Dragon	10,000	570 fps 1400 fps 704 fps 1380 fps	570 fps 1400 fps 704 fps 1380 fps	570 fps 1240 fps 704 fps 1230 fps	440 fps 440 fps 440 fps 440 fps	Immediate Display List Vertex Array VBO
	250,000	39 fps 752 fps 56 fps 522 fps	39 fps 646 fps 56 fps 482 fps	39 fps 451 fps 56 fps 363 fps	39 fps 247 fps 56 fps 223 fps	Immediate Display List Vertex Array VBO
	871,000	10 fps 254 fps 17 fps 154 fps	10 fps 238 fps 17 fps 151 fps	10 fps 205 fps 17 fps 140 fps	10 fps 143 fps 17 fps 112 fps	Immediate Display List Vertex Array VBO
Yeah Right	?	12 fps 508 fps 19 fps 177 fps	12 fps 491 fps 19 fps 175 fps	12 fps 435 fps 19 fps 168 fps	12 fps 298 fps 19 fps 144 fps	Immediate Display List Vertex Array VBO

(Click pictures for full view)

For the most part we see that Display Lists and VBOs are much faster than Immediate mode and Vertex Arrays. A notable exception is the 10k dragon at 1024x1024; in which, all modes have the same frame rate. Such an occurrence suggests that the CPU is not bounding fps, but only the fill rate is. Another interesting thing is that in all cases, save the previously mentioned, Immediate mode and Vertex Arrays are invariant to the window size. This means that those modes must be bound by geometry traveling over the bus.

With Display Lists as well as VBOs we see that they very over both the window size and the number of triangles. However, this is a complex relation. In the 128x128 case increasing the model complexity has a proportional effect on the fps. Perhaps this is because it is geometry bound there. At other resolutions this does not hold. Most likely there is a mix in being geometry bound and fill bound. The large models weighing toward geometry bound and the high resolution samples toward being fill rate bound.

Problem 2:

Enable Blending

This was my first attempt. notice the bands along the outer edge of the sphere and the lack of Australia. This is due to the sphere not being drawn back to front. The back side is never drawn because the front side is in the depth buffer. Turning the Depth Test off will cause Australia to appear above S. America, so is not a solution.

Here I have fixed the problem by drawing the back side first then the front using culling. However, this only works if the object is convex (like our sphere). Displayed are two texture modes: Decal and Replace.

(Click pictures for full view)