Tutorial: Vertex count is what counts!

Tutorial: Vertex count is what counts!

Postby Julius » 11 Feb 2013, 17:09

http://distantsoulsdev.blogspot.com.ar/ ... count.html

Albeit nowadays it is usually rather the pixel shader fillrate that limits the rendering speed... with unified vertex and pixel shaders you can free up some shader cores for that by following that tutorial though.
User avatar
Julius
Community Moderator
 
Posts: 2450
Joined: 06 Dec 2009, 14:02

Re: Tutorial: Vertex count is what counts!

Postby riidom » 11 Feb 2013, 17:24

Interesting read, I didn't realize that Vertices get duplicated along UV seams as well. Even if it sounds pretty logical after reading this :)
User avatar
riidom
 
Posts: 218
Joined: 25 Jan 2010, 02:02
Location: irgendwo an der Elbe

Re: Tutorial: Vertex count is what counts!

Postby ctdabomb » 12 Feb 2013, 15:40

that was quite interesting.

uncompleted thoughts that might not be thought out the best:
though I maybe wouldn't say poly count is almost useless. It may not be the most accurate(right word?) way to tell how "complex" your scene/model is, but if that is the way most people are used to, and how they mesure how "complex" their scene can be, it would be the correct way.....
Some people are like slinkies... not really good for anything, but you still can't help smiling when you shove them down the stairs.
ctdabomb
SuperTux Moderator
 
Posts: 1075
Joined: 13 Dec 2011, 21:21
Location: halfway there

Re: Tutorial: Vertex count is what counts!

Postby amuzen » 12 Feb 2013, 18:21

This still isn't fully accurate because the batch count (number of different material+texture+shader combinations) of the model can have a much bigger impact on the performance than the vertex count, especially on modern GPUs and certain graphics engines. This is the case because every batch requires the CPU to bind different material properties/textures/shaders and issue a separate drawing call to the GPU.

Hence, the rendering time of a model can sometimes be decreased without touching the geometry at all. For example, if you have a model that uses 10 different diffuse textures, 3 different shaders, and 5 different specularity and shininess values (at least 10 batches), it will render slower than one in which the diffuse textures have been baked into 1 texture, the shader count has been reduced to 1 and the materials have been reduced to 1 by using a specular+shininess texture (1 batch total).
User avatar
amuzen
LoS Moderator
 
Posts: 327
Joined: 05 Dec 2009, 02:49

Who is online

Users browsing this forum: No registered users and 1 guest