6 replies to this topic

[thefire]

hi,
i made a game engine in c# and directx. my designer is asking me how much polygon count does it support so that it can model accordingly. how can i determine that. is there a test for polygon count. and doesnt it depend on the system specifications like processor memory graphics card etc.

[geon]

Of course it is dependant on the system. Chose your target system and to tests with that. Use some dummy models and dummy textures, shades, landscape etc. and test it to see how much it can handle.

[TheNut]

The fewer polygons you use, the better. Even if your system can handle millions of triangles per second, that doesn't mean you should get carried away. There are a lot of other influences on your final frame rate other than geometry. Remember, a good artist will create good looking models. A great artist can do it with half the number of polygons.

In almost all games I've played, fill rate is the number one killer on frame rate, followed by texture memory and then geometry. For a geometry test, construct X number of tessellated primitives such as spheres or tori. They can be constructed mathematically in software and all you need to provide is a resolution value. If you use vertex buffers, make sure to not exceed 65K indices per geometry, otherwise you'll hit a bottleneck.

[thefire]

thanku geon and thenut.

[kariem2k]

Hi
AFAIK modern cards works better when having much polygons sent to it in a fewer batches.
So use models whith quality between High polygon and low polygons.

[Almos]

I might be wrong about it, but I believe that it's done the following way - you create two models, first one lowpoly, the other one highpoly, and then map the geometry of the highpoly model onto the normal map you then assign to lowpoly model. One way to do it is to use Open Render Bump (http://engineering.s...tools/177.aspx), but you can as well use Nvidia Melody or Zbrush. What you get is an appearance of a complex, highpoly model while preserving low polygon count.

... or am I wrong?

[geon]

Almos:

That's a technique to reduce triangle count while preserving the visual impression of the details. The high resolution input models for this would typically have a tri-count in the order of millions...

...wich needs to be reduced to something more manageble. This is where the performance of the engine comes in. Naturally, a model will always look better with 10k triangles than just 1k. But normal mapping like that requires shaders (Fill rate) and a lot of texture memmory, wich then becomes more performance limiting, like The Nut mentioned.