Help, theory, experience

[rocneasta]

Last year i've spoken to Alex Sandri - who told me that the render time (he uses mental ray) depends on the geometry in terms if there are all quads it will be faster VS. trianglulated geometry (other software/booleans etc)

Now i'm in need to write another science paper and only topic from 3D that's coming to mind is this - render time in terms of scene geometry

Googling for this type of information seemed impossible and it is as it turned out.

So if anyone has any idea, info, theory or has done some testing i would like to hear it.

Thanks!

[poopipe]

The first thing I'd do is establish whether that's true or not and what context it's in because I find it very hard to believe that if all other things are equal and both objects are plain geometry that would be the case.

I can see there being a difference if we're comparing eg. a quad poly/subdivision surface and triangulated mesh as there may be optimisations based on assumptions you can make about the poly/surface

[sean_rodrigues]

I believe Poopie is correct. I'm pretty sure most render engines break down all geometry into triangle upon render. Render engines define the model by a triangles.

Sean

[ClassicGamer-3DT]

It could also boil down to the model's complexity I guess, there's also the possibility of intersecting polys or points that can occur and I know some renders take a while over 5 sided or more polys, but why i'm not sure.

Calculation in most renderer's take in to account the model's complexity but also scale, i've found scale to be a fundamental thing to get right..... too big and the renderer will stumble & struggle to start. There's also the system itself that you are using that can greatly enhance or hinder your workflow.

I've seen 12 core machines render lighting fast but still hesitate on compiling the geometry at the start, it seems most renderer's have a period of time when they have to work out how much geometry is being used in the scene before anything else happens....... to my knowledge the difference between the polys being triangulated or quads is irrelevant as the same amount of time passes for a compile.... however this can also be software specific and as Poopipe & sean_rodriques correctly say most (but not all) render engines will break down the geometry into triangles before finally rendering.

The only real test would be to bench-test with a simple scene such as the cornell box but that would need to be tested amongst all the latest softwares for a proper comparison.

You'd need 3DS Max Standard Render Engine, Mental Ray within all it's supported softwares, Maya, Vray within all it's supported softwares, Modo, Cinema 4D AR3, Lightwave, Blender, Vue, Fry Render, Maxwell Render, Final Render etc etc.

[astroAJ]

Interesting that u mention this. I was watching a Gnomon tut today (Mentalray Fundamentals) and this topic came up in relation to displacement maps. It depends on the type of tessellation methods you are using. In the tut he states that the Catmull Clark subdivions will render a lot faster then the standard Maya displacement tessellation method. I'm no expert on this but you can get different speed renders depending on the tessellation methods you are using, probably because the render is calculating in a different way??

Everything would be converted to triangles by the renderer but how it goes about the process would affect render times??

Correct me if I'm wrong.

[Perversonality]

I seem to recall reading something about quad strips being the most efficient thing for most graphics engines to calculate despite everything ultimately being rendered as triangles but I can't remember when, where or why...

The only thing I can think of here is maybe, and this is a wild guess, with quads there is less time spend on calculating surface smoothing and normals when averaging. A quad is still two triangles but with a soft edge maybe only four vertex calculations are carried out to smooth the two inner triangles whereas for two hard edged triangles you have to calculate based on 6 vertices; 3 per triangle... That's about the only thing I can think of that would make sense right now

[rocneasta]

Well today i got an answer from NVIDIA.

If anyone is interested I'm pasting it here.

Dear Mr. Suka,

Question being: Is the render time faster if inside a software models
are all made out of quads, when compared to triangulated mesh
imported from other software, or even made in the same software
only end result being a model with triangles (topology wise of course).

let me give an answer of this question from perspective of mental ray,
our established CPU based software rendering product.

Triangles have been used in the past rendering engines primarily to
achieve best speed. They are considered the simplest form of a "face".
Because they are always "flat" geometrically, the intersection test (for
ray tracing) is most simple compared to polygons with more than 3
vertices. Triangle meshes may use more memory because 2 triangles
are needed to represent 1 quad, for example.
Even rasterization engines, like on a GPU, support triangles for similar
reasons. Modern rasterizer's may render quads at (almost) the same
speed as triangles, though.

On the other hand, most modeling tools are based on quads, which is
a better match with typical geometry, like regular meshes or freeform
surface patches. They are also most suitable for subdivision, for example.
This is an important aspect, like for level-of-detail. When it comes to
rendering, those quads are often triangulated on-the-fly (which is only
exact if the quad is "flat" !) just because the engine expects it like this.
This extra step seems like wasted.

It's hard to measure an overall speed advantage if you switch from triangle
to quads. In mental ray, we always tessellate into triangles from quad input,
for the reasons mentioned above.

Hope this helps a bit.
Regards,
Steffen Roemer

Now back to finding another topic for my paper

[poopipe]

Originally Posted by Perversonality

I seem to recall reading something about quad strips being the most efficient thing for most graphics engines to calculate despite everything ultimately being rendered as triangles but I can't remember when, where or why...

There'll be a trade off between stored information eating up available memory and calculations done on the fly adding to the amount of processing required.

on relatively simple objects you'd be far better off rendering explicit triangles but on massively high res objects you'd be better off saving the memory and tessellating on the fly as it would mean the amount of memory used by the object could be smaller

[rocneasta]

Well i'm stuck and I need to do something...

Managing memory usage in 3ds max workflow, viewport and render wise

new working title

so far things to cover that fall to mind

- one object or a dozen smaller ones (the fact that max handles badly large number of models and handles with greater easy one or few larger ones (larger in scale of polys))
- optimized modeling (low poly count plus Turbosmooth) compared to non optimized modeling (high poly count plus Turboosmooth (lower iteration)
- poly count in general
- modifier stack, collapsing
- collapsing into editable mesh or editable poly (diffence in file size)
- mr proxy
- substitute, xref
- maps size (effect to the memory of unused maps still in mat editor)

This is of the top of my head - please if you remember anything else, know of an example or have something on paper (recall a segment in a book etc) please share. If it's published i can quote, if it's not i can say thank you

Gone to process***