Polycount

[svenskmand]

As we are currently working close with OGA to get lots of great art an important question has reemerged, namely polygon count.

As Open Dungeons is a long term project I think we could raise the bar regarding poly count for the creatures. We have talked about 1000-2000 polygons, but since GPU's and CPU's are improving dramatically over a few years - just look at the development in the GTA series over the past 6 years - I think we should reconsider this polygon limit of 1000-2000. For the human faction it has been suggested to use this Dwarf from OGA, it is fully rigged and textured, and as far as I could see in Blender it had a poly count of 6500 (Blender says Ve: 7079 and Fa: 6381).

So as a proof of concept I think we should try and put the Dwarf in game and see what happens if we have say 25 of him on screen at a time, and similarly try with 50 versions of him.

[oln]

Yeah, should test this. (I can't though, I have a crap computer that is struggling with the fps in the game already.)
Though, isn't this what normal maps are for? Like taking a high poly model, reducing it and generate a normal map from the high poly count. (If that makes sense.)

[StefanP.MUC]

I have absolutly no "feeling" for these numbers. How much does, for example, doubling the poly count make a difference compared to, for example, a better texture with good normal mapping?

If we can achieve the same results with different technologies we should use the one that needs the lowest gpu specs (only if it doesn't mean to have the artists redo everything).

[Danimal]

Remapping and retopo takes a lot of work, you should try first the current version, and if things get too choppy, it can be done, but given the current lack of 3d modellers is nothing that is going to happen soon.

[svenskmand]

but given the current lack of 3d modellers is nothing that is going to happen soon.

Even though we do not have any models yet, we still need a guide line on poly count.

Also CruZ from OGA wants to make OD the "Topic of the Week" which guess will bring in allot of artwork. So models might not be that long in the future anyway. But I think we should try and get the Dwarf into the game and make some tests to see how well it runs. I can run the tests, but I do not know how to the the Dwarf into the game.

[svenskmand]

A new guys is asking about the poly limit. I think it is very urgent that we make some tests of this. If someone with the knowhow could make a few levels with 25, 50, 75 and 100 versions of the Dwarf from OGA fighting each other then we could run it on our hardware and report back our FPS count.

I think just displaying the model should be fine for this performance measure as I assume that animations does not have a say in the rendering speed anyway, please correct me if I am wrong.

[Julius]

Unless you want to include a FPS mode like in dungeon keeper as a significant part of the gameplay I would strongly recommend against using such high polycounts. It's just a waste of resources, the models take longer to make, and on screen the details will not be noticable anyways.
Even 2000 polygons almost seem too much already for a game with a camera perspective of a RTS game.

[oln]

Ogre supports Level of detail. (I don't know if we are using it or not.) So higher polycounts should be okay, even though I agree that the models shouldn't be overly complex.

[svenskmand]

We are going to have a fps mode too, so in that respect it would also be nice with higher details on the models, also the hardware is growing stronger all the time (the GTA video speaks for it self), and as oln says we can also use a lod system so we might as well use it. Also OD is not your standard RTS like StarCraft 2 or Command & Conquer where you can easily have 200-500 units on screen at a time. When goes highest I think we will have 100 units, so we can afford more detail for each model.

[p0ss]

poly counts are the wrong way to dictate asset quality, huge poly counts are not an issue even on old hardware if the models and scene have been properly optomised.

Understanding how a GPU renders vertexes and textures is more important than watching your poly count in blender, since that poly count barely relates at all to the number of polys that get rendered at run time. Knowing wether you are fill or transform bound on a given model, and optomizing accordingly will result in far greater performance benefits than an arbitrary poly count limit.

Check out this great guide for more info: http://dungeonhack.sourceforge.net/Model_Optimization

[svenskmand]

Nice link, thanks