triangle batch size and performance

I read somewhere about the number of triangles you send to the graphics card using glDrawElements at one time made a differerence to performance, so I thought I’d test it out.

System:Celeron 600(4x75),geforce2MX, agp2X, win2K
Data Format:All vertex data (3 floats per vertex) in buffer allocated with wglAllocateMemoryNV(size,0,0,1.0), sending vertex indices over the AGP bus (it’s faster that way)
scene: one 102400 triangle teapot with a 256x256 texture, one light

batch size Time to render 1000 frames(s)
1024: 22.0
2048: 18.3
5096: 22.5
102400: 22.6

Anybody else seen results like this? Does 4 million polys/s seem about right for my system?

Update:this only seems to happen with Vsync enabled - with it disabled I just get a constant 16.7secs.

[This message has been edited by Benjy (edited 05-15-2001).]

Depends on how you set up the lighting – it makes a big difference.

Infinite lights with an infinite viewer are very fast, for example, while using a local viewer or a local light tends to slow things down a lot.

• Matt