Recommended polygon counts / texture sizes?

[MetroidJunkie]

This isn't so much a help post as it is one of personal opinion, I'm not really good at judging this. As far as models in GL go, what would you consider generally recommended as far as polygon counts and texture sizes? Also, texture sizes for the environment. My current system is to just use 256x256 for flats/walls with generally 256x256 (512x512 rarely) for models and with a limit of 5k polygons per model for any I happen to do. What would generally be good caps for these and possibly different values for different things such as first person weapon models, decorative objects, etc? The idea is to allow as many people with hardware that can run the latest GZDoom as possible while also maximizing the fidelity. I don't want it to look like crap but I don't want to just have it be a laggy mess on most peoples' computers either.

[Zanieon]

My recommendation about limits is just keep in mind all the limits of MD3 format, even if you are going use MD2.

• 1024 animation frames
• 32 Groups
• 256 Textures per Group
• 8192 Triangles per Group
• 4096 Vertices per Group

About textures, keep in mind that textures beyond 1024 delays a bit to be processed at the first time in a map on old hardware (my old pc that have a GeForce 7400GS delays 1-2 seconds to process textures with resolutions highter than this)

As Graf said to me on DRD Team forums, the problem of the model handling on GZDoom isn't the textures, what you should care more is about the polycount since now that he implemented the model buffer, if a model contains too much polygons, or many models with many polygons at once on screen, GZDoom can crash because it will eat all your VRAM when trying to render those models.

[MetroidJunkie]

So GZDoom is more suitable to have larger texture sizes with less polygons? I could work with that.

[leileilol]

MD3 is horribly inefficient with large polygon counts. Memory usage will skyrocket if you have a lot of frames with much polygons as vertex animation cache is expensive.( FWIW, OA3 abandons the use of MD3 for player models for this very reason (and OA players typically reach the 2000-2500 polygon range - having them MD3 caused players to crash if too many joined due to memory issues. Big mistake).)

Also, having more polygons won't make a model look better if you can't even shade one normal of it. There's no benefit for pigging out on the polygons, even if the engine "supports more" of it.

MD2 also has horrible 8-bit integer vertex precision (read: blobby heavy vertex swimming), the more polygons the worse it gets.

I would recommend 1500 max triangles for MD3 and 400-700 max triangles for MD2 (depending on the context, like weapons)



What GZDoom really needs is support for a skeletal format such as this agreed upon standard format in the Qommunity.



as for large textures (~512 and up), don't use PNG. PNG is slow at loading and its "benefits" are more pointless when it's packed in a zip anyway (an offline 3d game is not a website). Try to favor uncompressed 24/32-bit TGA.

[Zanieon]

I talked with Graf about model format, but i doubt he will change the code of model handling for now, MD5 would fit a way better for what GZDoom is actually, about normals, i do recognize models can look shit without them, but Graf is already tired of say that if nobody is going to help him on improving the renderer, he will not move a finger about it anymore.

Also this IQM seems to be good to use too, but never saw it b4, i'll take a look.

[MetroidJunkie]

I used MD2 at one point but, once I started using MD3, there was no going back. 1500 is a bit on the small side but, yeah, it's not like intricate detail will be appreciated on the engine anyway. At least I can compensate for it with bigger textures, thanks for the TGA tip. I would always just save as PNG without even thinking about it.

[NeuralStunner]

What GZDoom really needs is support for a skeletal format such as this agreed upon standard format in the Qommunity.

I agree. Vertex animation is the thing that still terrifies me about using models for non-static actors.

As far as shading, I wonder if it'd be acceptable to use a fake light source above the model. I think most of the time, that's how lighting gets baked anyway.

Oh right, it should be a ModelDef option as well, so pre-lit models don't look weird.

[MetroidJunkie]

Probably a stupid question but I'm curious about something, this inefficiency with the vertexes where much higher than 1,500 polygons per model would cause it to sap resources like crazy. Let's say I make a polygon model with 6,000 polygons but I split it into 4 parts so it's just 1,500 polygons each. Would that make it any less resource hungry or would it be exactly the same with no actual gain?

[Zanieon]

That go beyond MD3 limitations yet and yeah, but it will not get much impact if the supposed model appears rarely and only this one will have all this count on the polys.