So i'm curious as to what is cheaper on a game. Higher poly counts on character (lets say a MAXIMUM of 24k tri count) with 1 2048 texture or a lower poly count (lets say a MAXIMUM of 12k tri count) with 3-4 textures.
I've always wanted to know what is more expensive on the system. Thanks
Replies
I'm guessing normal maps costs more in performance then diffuse maps, so if you had a 512x512 diffuse and added a 512x512 normal map. How many polygons/tris could I add to the model before the performance cost would be the same as using a normal map?
I understand that this question is kinda like asking how long a rope is, but have anyone ever studied to see the difference in cost between adding a map or adding more polygons/tris?
me too
you don't want the GPU waiting on the CPU, and vice versa. But you also need to manage your memory pool
Level creation is all about balancing these two
I think the bigger issue is the number of Drawcalls, and shader instruction (which multiple texture fetches often attributes to).
That said, most DX11 pipelines will atlas textures during load.
Polycount is somewhat important, but it's largely irrelevant if the artists are smart and efficient.
It's all about balance
CryEngine and Source Engine can eat polygons for breakfast. But Texture/Material usage will eat up your ram. So models can be done away in murder with, but the way you utilize textures. Will severely cause performance problems.
In the end of my point, isn't to be said that you should make a 10k table or chair....because that would just be silly
Can also apply to OpenGL as well.
However the latter is faster on PowerVR chipsets because of the tile-based rendering structure, and the texture management (deferred higher mips unload, at least on Series1. )
Then you have the hardware in question too, and what kind of options are open for you in that regard and how they work.
With that said, somethings I found out:
If you use procedural texture, especially for games like PS2, Tribes, etc, then you can save alot of calls in more ways then one.
However, if you're working more traditional way, then texture will eat up stuff, since everything will be unique in someways.
Polygons (vertice counts) while important, will almost always be the cheaper option. You can average about 500K polies in a scene on your latest IPhone, and the performance drop will be non-existent and the space need small, however you're extremely limited when it comes to textures, to the point that even more then a couple of 1024 k's on the screen will bleed the memory to death.
Please note the stuff I mentioned aren't hard rules at all (as I'm sure you realized due to the other posts).
Edit: Someone needs to get the rights to texturecalls.com, I don't think anyone owns it, and that would be a pretty awesome name for a texturing based website.
The reason i ask is because if your in a situation where your model is used in game AND cinematic then you want to have enough polys for a really nice silhouette, but low enough to be used in game without bogging down the system. But if your model has a little higher poly count AAAND more textures then it's gonna cause problems... I was just hoping to find out if anyone knows the lesser of two evils.
I realize games now a days can handle characters that are well into the 20k range and be fine, and it really depends on the situation (ie, having 64 players in a level at once).