How does a polygon budget work in a game?

Ok so recently I have just starting an obsession with games, specifically the creation side and I was reading about polygons. Obvious over the years games have been able to use more and more polygons meaning the shape of objects can be much more realistic, but how do game developers know when they can't use any more polygons. I guess developing for a console is much easier (only guessing) because everyone has the same equipment but PC gamers have hundreds of thuosands of different configurations with completely different graphic performance and every couple months there is a better GPU or CPU allowing developers to import more polygons into their models.

What I'm trying to say is I have made a 3D head in Blender (which I am incredibly proud of as its my first model I have ever made on PC) how do I budget how many polygons can be used on each object in a scene so my computer doesn't explode when I try and bundle it all together? Or is that not how its done?

And also, something else related to polygons is, do polygons behind the player affect performance of the game? For example if I made a tree this billions of polygons in it and stuck it out of view of a player in a game, will that tree affect performance even though it cant be seen?

Thanks in advance, Matt

AS far as I am aware polygons not in view arent drawn due to z buffering/stencilling, although not being drawn I would imagine a model consisting of billions of polygons to consume quite a lot of memory and impact performance that way. edit: culling, knew it was called something just couldnt remember which

most gaming models would never be anywhere that high and if we are talking about a simple tree, game creators use techniques like normal mapping for the trunk and alpha transparencies on the leaves to approximate the detail and then of course instancing to make that one tree into a forest.

in terms of target it comes down to the minimum spec, after all as a developer you are doing this for money so you need to make sure it can work on most pc's out there.

Take world of warcraft for instance, will happily run at low settings on even the lowest intergrated graphics.

 

I know what you mean about the head modelling, when I made my first one that didnt suck I was over the moon :P

well basically polygons have vertices that need calculating, the more you have to calculate the more power you need and you only have so much power

as far as polygons behind the player those are still rendered, its just "offscreen Rendering"

now there is a thing which I believe is called culling which when something is behind a building far away it won't render it.

some light reading

http://en.wikipedia.org/wiki/Hidden_surface_determination#Culling_and_VSD

Thanks that has made things a little more clearer to me although I now need to do some research on terms I have never heard of! Haha

I am going to be making a street in Blender and then importing all the models into UDK and build a scene, roughly how many polygons should I aim at to be my limit, if that makes sense?