Hello there.
When it comes about render speed it's a good idea to keep the objects geometry and textures as low as possible -- well known. Still, it's of humongous use to know this a little in detail. It can considerably boost FPS's to follow simple rules like: which of the following matter most, which less and which are irrelevant?

1. vertices number
2. polygons number
3. objects number
4. texture size

The whole idea behind this is to have a balance between them so to get zero bottleneck.

You're probably asking what does texture size have to do with vertices? Well, I could split a polygon of an object in more several polygons, by extra verts, and use a smaller texture that otherwise would need to be larger to hold the initial single polygon.

I know some of you will say it depends on each particular project but still, there are general rules to start from and then one would bend them as seen fit for a case.

What does your experience or theory say about it?

object count -> more than 2k Objects rendered at once will already cause your CPU to be the bottleneck

vertex count -> a higher vertex count means more vertices to be pusht to your graphicscard, which could make you bandwidth limited

texture size -> basicly like the vertex count + more expensive sampling. On the other hand, mipmaps help a lot to minimize the cost. Maybe this should be before vertex count, I am not sure.

polygon count -> a little more data has to be pusht to your graphicscard and your graphicscard has to draw more polygons, which could become a problem, especially with complex vertex shaders.


In the end, it is hard to give them an order. If you try to render 10million polygons (which most probably also means, pushing a few million vertices to your graphicscard), this will most probably cause your rendering speed to be not THAT great. But sampling an extremely highres texture several times for each pixel, or even different ones, will have the same effect. Same goes for an extremely complex vertexshader (which could also sample highres textures) in combination with a high vertex count.
It all also depends a lot on your hardware. Modern graphicscards basicly don´t care about the number of polygons to render (today I for example rendered a watersurface with a bit over 1million polygons at about 150fps with my gtx 460), here the problem tends to get the amount of data, which is the reason for creating geometry on the fly on your graphicscard. In contrast to this, older hardware has for example easier restricted by textures and polygon count.
And again, shader complexity can cause everything to be slow as hell, too.
So in the end you should always check the number of objects being rendered first, and if your cpu is not the problem, all the other points.

it also depends greatly on the platform your shooting for. mobile... polygons will have the biggest impact, casual, I'd guess texture size. Hardcore, then its probably a mix between texture and shaders. Newer graphics cards are really not impacted by polygons like they used to be. (not that you can have infinite, but it can still be fairly high).

also as slin said object count is a big issue with a lot of engines as well.

Thanks. So generally it's O V T P

Not really, it is more O VTP .