Radeon8500 "high poly count" bug?

[Randell]

I'm surprised, but then I dont know the config script Anand was using at only 52fps at 1024x768x32 bit and 2xAA. As I said above I'm getting 60fps with 2xAA and 16x aniso in Dunes/Karnak demo's. I assume Anand is using Rev's max quality config, which I dont have.

When I get a chance (maybe tonight, maybe not) I'll run some benches on SS:SE for you/I to look at results.

The key thing in Anands review is that he has to enable aniso with AA to equal 8500 AA texture quality without Aniso, so performance draws near to each other. So with the lower hit 8500 8x & 16x aniso takes coupled with the higher hit Gf3 4x & 8x takes, then I would expect an 8500LE to perform better than a Ti200 with 2xAA and highest aniso settings.

 

[Randell]

Ok as GOA screwed up patching the Euro DAoC servers tonight, I had some spare time.

Serious Sam:SE Demo

System
Athlon XP1600
Radeon8500LE
384meg Crucial DDRAM @ CAS 2
Win98SE
9021 beta drivers

SS in-game settings

All default quality or as follows

Precached shadowmaps=no
Texture Rendering=Trilinear (I assume it falls back to bilinear when aniso forced)
Texture Anisotropy=none
Texture LOD Bias = normal
TRUFORM = High
Use TRUFORM = TRUFORM ready models only
Texture Compression=S3TC

all at 1024x768x32bit

Demo run = Sierra De Chiapas
0 AA, 0 aniso = 94.8 fps (origianlly recorded fps)
0 AA, 8x aniso = 85.9 fps
0 AA, 16x aniso = 78.7 fps
2x Quality AA, 0 aniso = 60 fps
2x Quality AA, 8x aniso = 50.2 fps
2x Quality AA, 16x aniso = 47.6 fps

So we are talking again ~ 50% hit on my system for 2xQual AA and highest aniso and approz 37% for 2xAA only in this bench.

For fun I tried D3D and got 51.2fps at 2xAA, 16x aniso, but had rendering errors on the gun etc.

Also setting aniso in game or forcing through drivers had no effect on performance.

So I still dispute any site posting 50 to 60% hit for 2xAA only in SS:FE.