VBO Performance with Cat 3.10

Here’s a quick benchmark of VBO performance using the latest Catalyst 3.10 drivers.

Benchmark System Specs:
WindowsXP Pro SP1
Athlon XP 2700+
512MB Ram
Radeon 9700 Pro
Catalyst 3.10

The scene being rendered:
1280 x 1024 x 32bpp (fullscreen)
1 cubemap (each face 512 x 512)
1 sky box (display list)
1 sphere (128880 triangles; VBO; reflection map)

Using brute force rendering for everything.
Sphere uses 2 buffer objects - one for vertex data, and the other for index data.

Results:
Using VBO 180 FPS
Using standard vertex/index arrays 34 FPS

5.3 times speed increase using VBO

[This message has been edited by dpoon (edited 12-30-2003).]

Any perf numbers for vertex locking in the same situation? (glLockArrays) or when placing the sphere geometry in a displaylist?

Here are some updated results.

VBO: 180 FPS
Display list: 80 FPS
CVA + Vertex Array: 36 FPS
Vertex Array: 34 FPS

BTW. The sphere’s geometry is organized such that triangles around the poles are in triangle fan order, and the triangles making up the body of the sphere are in triangle strip order. With the exception of the display list, the sphere was rendered as an indexed triangle list with a single call to glDrawRangeElements().