How many Triangles does a regular Part have?

Just like the question how many Triangle count does a rectangle roblox part have? Is it 12, 16 or some other number? I recently found that converting parts to Meshes can give a slight performance boost so knowing this would help me compare a model and a Mesh.

Edit: I know I said convertings parts to Meshes can give a slight performance boost but what I really meant to say was that converting Unions to Meshes can give slight performance boost.

I wanted to compare a part to a union and a union to a meshpart. Because some of objects I need have to be made out of union operations.


Parts are very optimized, and should almost never be converted to a mesh unless it is a large assembly with a lot of unnecessary faces that you intend to remove with something like blender. Converting a part to a mesh does not reduce triangle count, nor does unioning them. It only adds to the list of things the client has to download when you first join the game.


@1Joeb is correct here. The assumption that a Part is just a 12 tri mesh is only true in a geometric sense, but Roblox treats both instances very differently to the point where most comparisons are wildly inaccurate, especially in terms of performance optimization.

There’s an unsubstantiated claim going around that devs should just convert their whole map to MeshParts, but this is not helpful at best, and game-breaking at worst. Don’t do it unless you 100% know what you’re doing, and have a backup saved.

1 Like

Pretty sure parts are rendered in a way where if you put 25 parts in a square, it only renders the visible faces unless the camera is inside one of them?

idk im not a roblox engineer

(i cant draw diagrams so roll with me here, the red faces are the ones that aren’t rendered by the engine, at all, assuming the two parts are on a baseplate)

There are a lot of different kinds of rendering taking place in a game. From physics rendering to wireframe rendering and your game performance can be affected by all of those so it’s best to make sure every one of them is set up properly to avoid any unnecessary lag.

How many times did you run the comparison to account for random variation?
If you can go to the full length of testing each case 5/10 times and use (as Developer Console has been known to be fuzzy), and report on any domains of PlaceMemory that parts attribute to I might then maybe concern this stuff being valid or not to your application.

If that sounds like a lot, that’s a sign this is a preoptimization. It’s a lot more in this engine confounding the performance of just 12 tris in a default cube, to the point where even hardware variations because of software like metal on the backend may or may not be used, may or may not configure a certain way when running.

1 Like