EQ can you explain your reason for B? Based on these tests back in 2009 by Humus it would seem C or as equilateral triangles is best for performance. http://www.humus.name/index.php?page=News&ID=228
The humus article talks about tessellation. And there are several tessellation algorithms. If the model is not going to be tessellated, any method is valid.
I use to use B in certain models (it's a way to have quads in a cylinder of parity 2), but for normal mapped models A gives me better results.
Ngons are extremely useful for terminating edge loops, if you dogmatically stick to using only quads, the only thing you are accomplishing is making your life harder than it needs to be.
Absolutely! it jes makes modeling easier!
Switching from patch modeling nurbs in poweranimator to polymodeling in mirai was probably the single most influential technical advancement I have ever experienced. Namely because I could temporarily add an ngon with complete abandon not slaved to an all quads technology restriction. Hearing 3d Application battle whining about how much Maya or Max sucks makes me wonder if they would feel the same way if winged edge freedom was just added today?
( everything else just seems like extra gravy... nice but the revolution already happened! whats there really to complain about? )
Even the ngons that will cause topology artifacts are golden!
Because they allow you to build charming localized detail when and wherever you need it at the very beginning stages of volume and silhouette building.
Seems like some modelers sweat the quads way to early in the process. ( when u r going to clean re-topology in the end anyway? )
Why waste the time and energy worrying about quads when the freedom of n-gons is a gift? ( u can even begin edge flow right away when terminating on inconspicuous areas with n-gons can be sweated over at the end when the awesome has already been hashed out )
btw thanks Earthquake...
That gun in addition to the newb tube and under over clip is my favorite weapon in any game!
Another oldie: ( sadly 14 year old CG history is getting harder to raise from the dead even on the wayback machine )
When you look at a highly tessellated model it's generally understood that it will be vertex processing heavy...
...The number along the x-axis is the vertex count around the circle and the y-axis is frames per second.
He starts with a "circle" of 48 edges and he increases the vertex count via tessellation. What i read is the impact of perfomance of the shader with one method of triangulation for a mesh with circular shapes. This is also dependant of the algorithm as far as i know.
A cylinder with 16 sides or 32 won't affect the fps if their cap is triangulated like in the A example.
According to him, the "max area" type offers the best perfomance. And he didn't say which method of tessellation was used.
For in engine game assets of course use only quads and tris. Some engine will subdivid your ngons for you, but some wont and can give you undesired smoothing/normal issues.
Surely you mean tris. Importing quads into a game engine is just begging for a quad to get divided in the other direction and give you a smoothing/normal map error.
Don't suppose anyone has actually tried to reproduce this?
I tried to benchmark it in UE4 but trying to import a 1600 vert fan just causes UE4 to hang.
Edit: Not saying it's wrong, intuitively it makes sense especially on large discs
I'm just wondering how practical it really is to actually optimize this way, to me it would seem that unless you're doing something horribly wrong with your scene you'd never actually see the benefits of tesselating a disc this way...
B is generally going to give you the most reliable baking behaivor, which is why I favor it.
The max area thing, I won't argue that using large polygons like that is technically a little faster or more efficient, however, doing benchmarks with a bajjillion cylinder caps triangulated in that way is simply not anything close to a realistic use case. Fact of the matter is no modern game engine/project is going to have that sort of thing be anything close to a bottleneck, so in real use, you're looking at performance differences that are absolutely miniscule.
So the tiny performance benefit does not outweigh the pain in the ass that it is to manually triangulate your edge caps like that, and the bake skewing issues you're likely to see with that method.
While polycount is about games stuff, I'd still like to make a few notes. Disclaimer: I'm the character modeling lead at Digic and we build assets for CG, using subdivision at render time, so a lot of this is probably irrelevant to games today but might become important in a few years.
As others have already mentioned, anything that needs to go into zbrush for sculpting should be all quads. 5+ sided polygons are broken down to random triangles and quads which is already something you don't want. Another issue is that "pole" vertices, ie. vertices with more or less than 4 edges, can create problems while sculpting, especially when using the Smooth brush.
Also, when you're working with blendshapes for facial animation, a lot of the deformation is about sliding the skin around on top of the underlying bone structure. This works much better if your mesh is all quads (and has a good density) as you can easily use brush based tools like Nudge in Zbrush, or Edge constraints in 3ds max.
Also, with CG work, most studios don't try to build rigs that work 100% of the time, as that'd be far too time consuming. Instead, the animated character is usually sculpted as a final pass, by baking all the deformations into a point cache; some studios use custom tools, for others there's Chronosculpt (although it is a very primitive tool unfortunately and it seems to be discontinued). A full quad mesh is preferred for this as well.
And yeah there's Sullivan from Pixar, but let me point out that his entire body was covered with fur... And also, that model was made ages ago, right after Pixar introduced subdivision surfaces and they had very little experience with the tech at that time.
Now on the other hand, ngons still have a place in the modeling workflow. With today's computers we can work with relatively high res models, for example our heads are about 30-50k quads and we model all the facial wrinkles so that we can control them with blendshapes. This means that the retopo workflow can and should rely on subdividing the mesh once (and then apply some final tweaks). Basically you build a model with about 1/4th of the target poly count and on that mesh, anything goes.
But for hard surface stuff that you don't want to sculpt in zbrush, ngons, or rather, 5-sided faces, are of course perfectly fine.
B is generally going to give you the most reliable baking behaivor, which is why I favor it.
The max area thing, I won't argue that using large polygons like that is technically a little faster or more efficient, however, doing benchmarks with a bajjillion cylinder caps triangulated in that way is simply not anything close to a realistic use case. Fact of the matter is no modern game engine/project is going to have that sort of thing be anything close to a bottleneck, so in real use, you're looking at performance differences that are absolutely miniscule.
So the tiny performance benefit does not outweigh the pain in the ass that it is to manually triangulate your edge caps like that, and the bake skewing issues you're likely to see with that method.
Wasn't all that stuff written back when hardware was still experimenting with coming up with the best techniques? I remember all sorts of stuff about vertex cache optimization, tri stripping, type streaming, but never saw anything else after that. Now that the GPUS have grown bigtime, I am a little curious what tricks they use.
Another issue is that "pole" vertices, ie. vertices with more or less than 4 edges, can create problems while sculpting, especially when using the Smooth brush.
while this is true, you can't avoid poles so where to put them is key, with 50k heads they are thankfully pretty small ^^
However i hope you do know the alternative smooth mode that works much much better with poles in zBrush.
The thing with ngons is, that they can always be easily removed by creating triangles and quads, and yes rerouting topology is the proper way of fixing ngons, but the point is that it's not hard to simply not have ngons in portfolio pieces, so just don't.
The thing with ngons is, that they can always be easily removed by creating triangles and quads, and yes rerouting topology is the proper way of fixing ngons, but the point is that it's not hard to simply not have ngons in portfolio pieces, so just don't.
too dogmatic, a well used Ngon workflow is way better than a shitty quad only attempt.
The thing with ngons is, that they can always be easily removed by creating triangles and quads, and yes rerouting topology is the proper way of fixing ngons, but the point is that it's not hard to simply not have ngons in portfolio pieces, so just don't.
too dogmatic, a well used Ngon workflow is way better than a shitty quad only attempt.
That quads vs ngon is a thing from sub-d modelling . Honestly I don't understand why people still use it when every shading issue is perfectly and quickly fixable by editing vertex normals . Or just by using CAD modelling from the start . Even Zbrush re-mesher does a decent job now. The whole thing looks like a sport still being taught for some uncertain reason.
I already forgot when I did Sub-D last time. All those extra edge loops the shading is relying on just make you extra problems with LODs .
For hard surfaces you do hard edges and edit normals . Who cares quads or n-gons since you put triangulate modifier in the and anyway.
Honestly I don't understand why people still use it when every shading issue is perfectly and quickly fixable by editing vertex normals . Or just by using CAD modelling from the start . Even Zbrush re-mesher does a decent job now. The whole thing looks like a sport still being taught for some uncertain reason.
This is a very subjective take, based on your own experience only. Now of course everyone is entitled to their opinion, but such a bold claim here is bordering on misinformation.
" I already forgot when I did Sub-D last time. All those extra edge loops the shading is relying on just make you extra problems with LODs .
For hard surfaces you do hard edges and edit normals . Who cares quads or n-gons since you put triangulate modifier in the and anyway. "
I wish more artists i work\ed with didnt, because it sure its a slap of reality when they have to find out they lack basic subd knowledge on workflows where clients rely on those still.
I use n-gons (pentagons) and triangles. A proper subdivision modeller has to use them when they are needed, including edge weights.
To rely on zbrush remesher because it gives us a good but dense mesh is a fail in all senses. Subdvision modelling is all about topology and clean meshes.
Don't be lazy, learn proper subdiv modeling... and you'll make models faster. Don't ignore the basics, just an advice!
Bout zbrush, i got tired of the "spindle" like topology (infinite loops if you select rings, hated it) it created, it doesn't help remove rings(+) guess people got away with clicking remesher and calling it a day? Does it still spindle loop?
Bout zbrush, i got tired of the "spindle" like topology (infinite loops if you select rings, hated it) it created, it doesn't help remove rings(+) guess people got away with clicking remesher and calling it a day? Does it still spindle loop?
it does. not as often but it does. it gives you sculptable topology, but i wouldnt call it production ready by any means.
I use n-gons (pentagons) and triangles. A proper subdivision modeller has to use them when they are needed, including edge weights.
To rely on zbrush remesher because it gives us a good but dense mesh is a fail in all senses. Subdvision modelling is all about topology and clean meshes.
Don't be lazy, learn proper subdiv modeling... and you'll make models faster. Don't ignore the basics, just an advice!
I see ngons as being valuable, but mostly as an intermediate step. Ultimately, models are just going to get triangulate or converted to quads if for no other reasons than to promote clean topology and avoid non-manifold geometry. Nobody wants their models or textures distorting in unpredictable ways.
Ngons are good for showing a model off, especially archvis or CAD stuff. However, for everything else - as a matter of personal preference - I try to avoid them for anything outside of the WIP stage.
Replies
Would love to hear your thoughts.
....although A is probably the best if I need to make an edit later
That 18-gon can be optimally triangulated like this:
There's more information here:
http://www.polycount.com/forum/showthread.php?p=2213096#post2213096
I use to use B in certain models (it's a way to have quads in a cylinder of parity 2), but for normal mapped models A gives me better results.
Switching from patch modeling nurbs in poweranimator to polymodeling in mirai was probably the single most influential technical advancement I have ever experienced. Namely because I could temporarily add an ngon with complete abandon not slaved to an all quads technology restriction. Hearing 3d Application battle whining about how much Maya or Max sucks makes me wonder if they would feel the same way if winged edge freedom was just added today?
( everything else just seems like extra gravy... nice but the revolution already happened! whats there really to complain about? )
Even the ngons that will cause topology artifacts are golden!
Because they allow you to build charming localized detail when and wherever you need it at the very beginning stages of volume and silhouette building.
Seems like some modelers sweat the quads way to early in the process. ( when u r going to clean re-topology in the end anyway? )
Why waste the time and energy worrying about quads when the freedom of n-gons is a gift? ( u can even begin edge flow right away when terminating on inconspicuous areas with n-gons can be sweated over at the end when the awesome has already been hashed out )
btw thanks Earthquake...
That gun in addition to the newb tube and under over clip is my favorite weapon in any game!
Another oldie: ( sadly 14 year old CG history is getting harder to raise from the dead even on the wayback machine )
In praise of 5 sided polys ( Don't worry be happy )
He starts with a "circle" of 48 edges and he increases the vertex count via tessellation. What i read is the impact of perfomance of the shader with one method of triangulation for a mesh with circular shapes. This is also dependant of the algorithm as far as i know.
A cylinder with 16 sides or 32 won't affect the fps if their cap is triangulated like in the A example.
According to him, the "max area" type offers the best perfomance. And he didn't say which method of tessellation was used.
Surely you mean tris. Importing quads into a game engine is just begging for a quad to get divided in the other direction and give you a smoothing/normal map error.
I tried to benchmark it in UE4 but trying to import a 1600 vert fan just causes UE4 to hang.
Edit: Not saying it's wrong, intuitively it makes sense especially on large discs
I'm just wondering how practical it really is to actually optimize this way, to me it would seem that unless you're doing something horribly wrong with your scene you'd never actually see the benefits of tesselating a disc this way...
The max area thing, I won't argue that using large polygons like that is technically a little faster or more efficient, however, doing benchmarks with a bajjillion cylinder caps triangulated in that way is simply not anything close to a realistic use case. Fact of the matter is no modern game engine/project is going to have that sort of thing be anything close to a bottleneck, so in real use, you're looking at performance differences that are absolutely miniscule.
So the tiny performance benefit does not outweigh the pain in the ass that it is to manually triangulate your edge caps like that, and the bake skewing issues you're likely to see with that method.
As others have already mentioned, anything that needs to go into zbrush for sculpting should be all quads. 5+ sided polygons are broken down to random triangles and quads which is already something you don't want. Another issue is that "pole" vertices, ie. vertices with more or less than 4 edges, can create problems while sculpting, especially when using the Smooth brush.
Also, when you're working with blendshapes for facial animation, a lot of the deformation is about sliding the skin around on top of the underlying bone structure. This works much better if your mesh is all quads (and has a good density) as you can easily use brush based tools like Nudge in Zbrush, or Edge constraints in 3ds max.
Also, with CG work, most studios don't try to build rigs that work 100% of the time, as that'd be far too time consuming. Instead, the animated character is usually sculpted as a final pass, by baking all the deformations into a point cache; some studios use custom tools, for others there's Chronosculpt (although it is a very primitive tool unfortunately and it seems to be discontinued). A full quad mesh is preferred for this as well.
And yeah there's Sullivan from Pixar, but let me point out that his entire body was covered with fur... And also, that model was made ages ago, right after Pixar introduced subdivision surfaces and they had very little experience with the tech at that time.
Now on the other hand, ngons still have a place in the modeling workflow. With today's computers we can work with relatively high res models, for example our heads are about 30-50k quads and we model all the facial wrinkles so that we can control them with blendshapes. This means that the retopo workflow can and should rely on subdividing the mesh once (and then apply some final tweaks). Basically you build a model with about 1/4th of the target poly count and on that mesh, anything goes.
But for hard surface stuff that you don't want to sculpt in zbrush, ngons, or rather, 5-sided faces, are of course perfectly fine.
Wasn't all that stuff written back when hardware was still experimenting with coming up with the best techniques? I remember all sorts of stuff about vertex cache optimization, tri stripping, type streaming, but never saw anything else after that. Now that the GPUS have grown bigtime, I am a little curious what tricks they use.
while this is true, you can't avoid poles so where to put them is key, with 50k heads they are thankfully pretty small ^^
However i hope you do know the alternative smooth mode that works much much better with poles in zBrush.
I'd go with B for Baking.
3ds Max/Zbrush: Proboolean + Dynamesh hardsurface workflow tutorial
too dogmatic, a well used Ngon workflow is way better than a shitty quad only attempt.
This is a very subjective take, based on your own experience only. Now of course everyone is entitled to their opinion, but such a bold claim here is bordering on misinformation.
I wish more artists i work\ed with didnt, because it sure its a slap of reality when they have to find out they lack basic subd knowledge on workflows where clients rely on those still.
I use n-gons (pentagons) and triangles. A proper subdivision modeller has to use them when they are needed, including edge weights.
To rely on zbrush remesher because it gives us a good but dense mesh is a fail in all senses. Subdvision modelling is all about topology and clean meshes.
Don't be lazy, learn proper subdiv modeling... and you'll make models faster. Don't ignore the basics, just an advice!
it does. not as often but it does. it gives you sculptable topology, but i wouldnt call it production ready by any means.
Ngons are good for showing a model off, especially archvis or CAD stuff. However, for everything else - as a matter of personal preference - I try to avoid them for anything outside of the WIP stage.