Less polygons vs smaller polygons?

Hello. I don’t know the specifics of rendering so maybe someone can tell me, which is better in general: Less polygons or smaller polygons? By this I mean for example if I made a transparent plane (2 triangles) that I slapped a masked texture of a branch on and I got a different plane and put the the same branch texture on it but this time I added a bunch of geometry to minimize the transparent space which would perform better? Does hiding transparent polygon space have any effect at all? What about if the texture wasn’t masked and in general took up a smaller space on the screen but had more polygons, which would perform better?

there is a /v large post about this and me and my class did some experimentation with it. we took a leaf, mapped it to a single plane and sprayed at around a few thousand times, then took a copy back into maya, and cut out a rough shape to fit the leaf better, overall i believe it was around 15-20 tris. as that other post explained @ LOD one the higher geo ran much faster, because tris are the cheapest thing to render, while if you have many large masked things in front of each other, the engine has to render the same pixel lots of times until it finds a solid pixel to render. so if you have say 40 leaves that line up on a tree and 20 trees that line up (that’s a conservative number it can get much higher) in a worst case scenario you have to render a pixel 800 times, for all those pixels at varying “depths” of transparent overlap.
past LOD 2-3 the sizes are to small to get any benefit, and having so much small tris packed into single pixels will hurt performance, so cards are better. this is another reason we do LOD’s, the reduction of tris certainly helps with performance, with just the amount less you have to render, but it also prevents multiple small tris occupying the same pixels at far distances.

TLDR; less transparency is better at close distances, less tris is good at far distances.

Your question has a 2 part answer:

Smaller polygons cost more. Basically a GPU runs a shader over each 2x2 pixel grid. So if one polygon covers that it’s run once, very fast! Even if two, or three polys are there it’s not the end of the world. It’s when you get to sub pixel polygons, more than one in a single pixel, that GPUs really start to choke. That’s why we have LOD and etc.

So use that LOD when you can. Buuuut, for your example, a good optimization can be to create a close up LOD where there’s more triangles but the transparent plane has less pixels. IE use polys to cut as close to the actual “content” of your alpha cutout tree branch and/or actual visible part of your particle effect. This works close up because remember, as long as your polys are larger, in screen size, than 1x1 pixel (or 2x2 pixels ideally) you don’t totally have to care how many you have, so eliminating parts of your texture that would be empty anyway will reduce overdraw and improve your performance!

Thankyou for your replies! I’ll try incorporate this into my workflow.