The great debate in rendering.
>>108718899strip for being the most general and no pathologically small triangles
>>108718899Now rotate them left 90 degrees
>>108718899fan is ideal because it's the most useful in modeling. you can loop cut around the inside and create a smaller circle within it. it's actually one of the most common things you will do when working with circular faces. you can't easily or cleanly do that with the other two which makes them junk in comparison.
>>108718899Why triangles, though?
>>108721365a triangle is always a perfectly flat plane
>>108718899you should post the source: https://www.humus.name/index.php?page=News&ID=228tldr: max area is optimal for performance from rasterization standpoint, even more so when MSAA is used.What he doesn't mention though is that this can only really work for static meshes, as max area triangles topology really doesn't go well with animation.The rule of thumb is to avoid thin triangles, because with many vertexes the whole circle would be densely covered with triangle edges, while max area still retains the big gaps in the middle and is only dense along the circle edges.
>>108721400which plane? a square is just as flat tangentially. likewise any polygon.
>>108722379any plane. three points in space decribe a plane on which they all reside because two points describe a line, adding another makes them two dimensional (and strictly two dimensional relative to each other). adding a fourth point allows you to create geometries that are not coplanar relative to all other points in that set. you can bisect these geometries into triangles with coplanar faces, but the entire set of n>3 points itself is not inherently coplanar; its very very very rare for them to describe a plane.https://en.wikipedia.org/wiki/Coplanarity
>>108722379move one point individually, it is no longer flat, move one point of a triangle and it is always flat
>>108722213Difference only starts showing after your circle is mad up of 100 triangles, you have more important problems at that point than pixel coverage.
>>108718899Left shape has 12 triangles, not 10
>>108722379Like anon alluded to, you can assume a triangle is flat, but you can't do that for >3 points, so you would have to check or do other slow retarded shit. Better to keep things simple(x).
>>108721365Any polygon can be decomposed exactly as a set of triangles. Since we can always use a polygon to approximate a general shape regardless of complexity, then triangles allow a generalized rendering pipeline of that..
>>108718899>>108721180The true answer is whatever the ai decides to spit out.
>>108722827I see where you're coming from and that's a common beginner mistake. One of those triangles is an anti-triangle (-1) and that cancels out a neighbouring triangle to get the final value of 10.
>>108722852what about a digon and monogon?
>>108718899> 10 triangles on each> first has 12 trianglesoh ok
>>108722213The post says that fragments are processed in 2x2 quads, is that true in modern GPUs as well? or has the area increased
>>108723761I believe it's still true
>>108723770I see interesting, that makes dFdx and dFdy essentially cost free (register access) as the gpu needs to calculate that anyways for mipmapping and anisotropic filtering
>>108722213So what you're saying is if I'm making static props or modeling static elements on a vehicle, I should say "fuck you" to clean topology and UV's and go with "whatever uses the least and largest triangles" and make exceptions only in areas where a model is rigged for skeletal animation, such as the suspension or flaps on a car as opposed to the roll cage or frame (assuming no deformation from collisions) if the game is forward rendered and uses MSAA?
>>108718899quads
>>108721365simplest 2d shape (duh), always flat irrespective of translation of points.
>>108721180best shape for modeling doesn't necessarily mean best shape for rendering
>>108721365it's the 2d simplex. it has a lot of nice mathematical properties that no other shape has that makes it convenient for rendering.>>108724489quads, for instance, suffer from not being able to guarantee the verts are coplanar (and they rarely are) which can lead to strange output in the worst cases
>>108718899triangles are for niggercattleSDFs + raymarching are the way and the lightInigo Quilez(PBUH) will make lsd real.
>>108718899A single triangle and one petaflop of DLSS 5.
>>108724268This is why retopo tools has existed for decades.So you don't have to deal with the fucked up modelling, and you can bake to bypass the work.
>>108721365Asking the real questions.>Saturn's use of quadrilaterals as its basic geometric primitive
>>108718899what about pic-related with all triangles from a single vertex?t. geometry teacher who has to teach kids that any polygon can be divided into (n-2) triangles.
>>108725636this is a fan
>>108725636>>108725863yeah, what the OP calls "fan" is more like a wagon wheel.Also the OP "fan" only works with polygons with an even number of sides
>>108723638Non euclidean polys need not apply
>>108724780Based and sdf pilled
all this 3D graph bullshit is overcomplicatedgive me raw asses to GPU hardware with 4kb "driver" an i will draw perfect sphere faster than the billshit triangles
>>108724842that was to save vram? to make a quead you need 6 vertices indices while with quads you only need four
if you were to measure circle path in hex in model and calculate pi value in hex how would it look
>>108724778quad overdraw
>>108727562>raw assesl-lewd!
>>108724489Sega lost.
>>108727773
>>108718899max area will have least amount of garbage in shader group executions, and as such will be the most performant.
>>108722449>>108722508>>108722828TY Anons. The comments in the original article posted here>>108722213 is great as well.
>>108727711It's literally just because 3D was a last minute thing after Sega saw the PSX reveal and realized acceptable 3D was going to be necessary, so Saturn polygons are basically sprites
Just draw one equilateral triangle and discard any fragments outside of the radius from the center in a shader. It's not 1997 anymore
>>108729451How is this better?
>>108724489>Image shows triangles>>108718899Obsolete. Mesh-based micropolygons are the future. The vertex stage is a dying breed.
