Object Poly Count

Unity Engine
1

I’m curious about a discrepancy between the unity docs the wiki regarding polys per object, specifically the amount you get charged for even if you are below that threshold.

Wiki says 500 polys:
http://unify.bluegillweb.com/scriptwiki/index.php/General_Performance_Tips

Unity docs say 1500 polys:
http://unity3d.com/Documentation/Manual/Optimizing%20Graphics%20Performance.html

IMO if you’re going to pay for a number of polys anyway you might as well use them for finer model detail (provided your overall scene count is reasonable).

Does anyone know which # is more accurate?

2

It actually depends on the hardware.

The thing is, each “draw” takes some amount of CPU time (and some GPU time, but this is smaller). Add to that that most (*) video cards nowadays have really powerful vertex engines, and you’ll quickly find out that you can use pretty high poly models, while still running at the same speed (because most often you’re CPU limited or fillrate limited; but not limited by vertex processing).

(*) The exception is video cards that don’t have vertex engines at all (hence do all vertex processing on the CPU). Pretty much anything from Intel comes to mind…

In short: if your target market has video cards with vertex processing, then something like 1500 polys is more accurate. If you include Intel, keep it lower.

To be really sure: just try and see what difference it makes. There are so many factors involved (hardware, number of objects, complexity of shaders, complexity of scripts, …) that it’s hard to tell the universal answer.

To add more to the confusion, it’s actually not the number of polys that matter for the video card. It’s something between number of polys and number of triangles and how well they map to various caches on the video card :slight_smile:

3

And to confuse a little bit more also the size/position matters… :O)

4

Interesting. I think i’ll probably stay with what i was doing then until i can test on different machines (1000 tris per object max). I would of course like to support as many systems as possible but i’m not interested in sacrificing too much to support every system out there. (also - you might want to update one of the performance pages to mention this so others might not be confused as well)

i don’t understand why this would be so, could you please explain it or point me too the right docs?

Thanks ; )

5

Empirical…

Hmm i also read a post from someone from futuremark about this…but i can’t remember the link right now…cards are optimized more for handling lot’s of smaller polygons but not for dealing with few large ones for instance.

6

It’s the opposite I think. Too small (i.e. one or two pixels in size) polygons can cause some of the pixel processors being idle.

But in general polygon size on the screen does not matter (or it matters too less to be noticeable).

This all is not true if were talking about Playstation2 (there polygon size does matter) :wink:

7

Fair enough ; )
The card optimization makes sense, but i wonder small / large relative to what exactly? Each other? Game units? Screen view? (that would be my guess)

I’d ask the same question regarding position as well, though if the above answer is screen view than i guess its postion relative to the camera.

EDIT: oh i see, thanks.

8

Well, there is a difference between “too small” and normal regular sizes compared to large polygons placed behind each other…

I’ve no experience with the playstation2 so i will believe you in this case… ;O)