"How many polygons?" is the first question asked about a model and one of the least informative on its own.
Triangles, not polygons
Everything renders as triangles. A quad is two triangles; an n-gon is several. Quoting a quad count understates the render cost, and different tools count differently — so compare triangles, and say which you mean.
Why the number alone is weak
- Distribution matters more. Ten thousand triangles well distributed beats ten thousand concentrated in a hidden interior.
- Vertex count often matters more than triangle count. Vertices are what the vertex shader processes, and a vertex is duplicated at every UV seam, hard edge and material boundary — so a mesh's vertex count can be far higher than its geometry suggests.
- Draw calls usually dominate. Fifty small meshes cost more than one large one with the same total triangles.
Where the budget actually goes
On modern desktop hardware, geometry is rarely the bottleneck for a typical scene. The usual limits are, roughly in order: draw calls, texture memory, overdraw from transparency, shader complexity, and only then vertex count.
On mobile and VR the ordering tightens and geometry moves up.
Sensible ranges
Rather than absolute numbers, which age badly, think in relative terms:
- A background prop should cost a fraction of a hero prop.
- A hero prop should cost a fraction of a main character.
- Anything instanced hundreds of times should be cheap per instance.
- Anything the player never approaches can be very cheap indeed.
Set the budget from your frame time and your scene composition, not from a number someone quoted for a different project.
For generated meshes
Reconstruction density is arbitrary and typically far above any sensible budget. The count on a raw generated mesh is not a specification — it is a by-product, and the first thing retopology replaces.