Deleting unseen polys nowdays?

S_ource

I'm wondering if you delete polygons that wont be seen ingame, especially on weapons. I know it used to be done. But on cs:go weapons its not done. Is it because its not really having an impact on performance?. Or for rendering or save time?

poopipe

polygons cost very little compared to shaders on current gen hardware.

for mobile devices this isn't necessarily the case but it depends greatly on the engine you're working with as to what you need to do.

AlecMoody

Also, if you have baked normals deleting faces will affect the shading of adjacent polygons (assuming they were smoothed together).

[HP]

Yeap, what AlecMoody said. If you wanna save polies, do it before the bake.
But yeah, polies nowadays are the least of your worries.

EarthQuake

Yeah this isn't really something that has been done for years. A given scene in a current gen game is going to be rendering hundreds of thousands of triangles on screen. Deleting one side of the most important model which is closest to the player doesn't make much sense. Other than the technical issues brought up it really limits what you can do with the animations.

S_ource

Okey, thanks for the answers.

passerby

polys are pretty cheap compared to shader and textures, if you really want to save on performance, try and optimize on drawcalls, and use less textures.


actually on 2nd note, if you really want to optomize, you should actually profile the scene, and figure out what is causing the biggest slow downs, could be any number of thigns, fillrate, drawcalls, running out of vram etc.

warby

reasons against:

- effects shading and normal map baking

- might cause dynamic shadow casting artifacts

- forcing your animators to never show a certain side of your model is also quit a limitation *no spinning revolver around the finger for you sir*

- might mess up carefully planned topology and triangle strip generation (which is a much more illusive process than it should be if you ask me ! ) in a worse case scenario backface killing might actually generate more verts/memory which leads to marginaly worse performance.

- waste of time spend that time on profiling your scenes and really optimize the bottle-necks instead its probably fillrate or drawcalls not vert transform time!




bottom line: in 99.99% of the cases ---> dont do it !