Here are some benchmarks from two NVIDIA T&L benchmark programs and my app.
Benmark5 - Direct X, no lighting
Balls - OpenGL, lighting, display lists
Mars3D(My app) - OpenGL, lighting, vertex arrays (glDrawElements)
The figures represent million polys/sec
My P.C. spec is PIII 700 + ELSA GF2
Original ELSA driver
Benmark5 - 17.9
Balls - 7.2
Mars3D - 3.3
Detonator 6.72
Benmark5 - 17.9
Balls - 12.1
Mars3D - 3.8
Detonator 10.5 (leaked)
Benmark5 - 17.9
Balls - 12.1
Mars3D - 7.1 !!
Even when I implemented the VAR extension to avoid the AGP bottleneck (I ditched this version afterwards for various reasons) my app never went much past 5M Tris/Sec because of memory bandwidth bottleneck on the GF2(I think). The NVIDIA VAR demo acheives its high T&L figure by using calculated vertices and doesnt need to get them from memory.
So this is quite amazing, almost a 90% speed increase!. As you can see from the figures display list performance is unchanged but it looks like there has been some driver optimisation to reduce bandwidth requirements for vertex array apps? (I’m guessing thats what theyve done )
Finally I can say that my app uses the T&L chip fully…well almost
If you want to try out Mars3D its here www.mars3d.co.uk
Benchmark figures were taken using the viewer software running the Mars Globe dataset at the highest resolution(not automatic).
I’d be interested to hear if anyone else has T&L benchmarked the 10.5 drivers. If AGP is not a bottleneck in your app you may not see any improvement.
[This message has been edited by Adrian (edited 02-18-2001).]