View Single Post
Old 03-07-04, 01:34 AM   #1
skamp
Registered User
 
Join Date: Mar 2004
Posts: 13
Default FX5900 benchmark with UT2004

Hello,

I just bought a GeForce FX5900, and I have been looking for the best compromise in terms of quality / performance, with and without Full Screen Anti-Aliasing & Anisotropic Filtering, for playing UT2004 with all details set to "max". System specs:
  • OS: LFS 5.0 (Linux From Scratch)
  • XFree86 version: 4.4.0
  • Drivers version: 1.0-5336
  • CPU: Athlon XP 2500+ (Barton)
  • RAM: 1 GiB of PC3200 DDR-SDRAM
  • Graphic card: GeForce FX5900 with 128 MiB of DDR-SDRAM
  • Display: flat panel display @ 1280x1024x32

According to NVIDIA's documentation, enabling FSAA and AF is just a matter of setting two environment variables: __GL_FSAA_MODE and __GL_DEFAULT_LOG_ANISO. As a reminder, here are their possible values:
Code:
# Full Screen Anti-Aliasing
__GL_FSAA_MODE=0 # FSAA disabled
__GL_FSAA_MODE=1 # 2x Bilinear Multisampling
__GL_FSAA_MODE=2 # 2x Quincunx Multisampling
__GL_FSAA_MODE=3 # FSAA disabled
__GL_FSAA_MODE=4 # 4x Bilinear Multisampling
__GL_FSAA_MODE=5 # 4x Gaussian Multisampling
__GL_FSAA_MODE=6 # 2x Bilinear Multisampling by 4x Supersampling
__GL_FSAA_MODE=7 # 4x Bilinear Multisampling by 4x Supersampling

# Anisotropic Filtering
__GL_DEFAULT_LOG_ANISO=0 # No anisotropic filtering
__GL_DEFAULT_LOG_ANISO=1 # Low anisotropic filtering
__GL_DEFAULT_LOG_ANISO=2 # Medium anisotropic filtering
__GL_DEFAULT_LOG_ANISO=3 # Maximum anisotropic filtering
I ran UT2004 botmatch benchmarks (?spectatoronly=true?numbots=14?quickstart=true?at tractcam=true -benchmark -seconds=80 -exec=../Benchmark/Stuff/botmatchexec.txt %1) on the following maps: Torlan, Bridge of Fate, and Rankin. The results:

Code:
# FSAA,AA          lowest / average / highest

Torlan
0,0     11 / 35 /  55 fps -- Score = 35
0,3     10 / 35 /  59 fps -- Score = 35
1,0     11 / 35 /  54 fps -- Score = 35
2,0     11 / 35 /  58 fps -- Score = 35
4,0     10 / 35 /  58 fps -- Score = 35
5,0     11 / 35 /  56 fps -- Score = 35
6,0      9 / 20 /  49 fps -- Score = 20
7,0      9 / 20 /  52 fps -- Score = 20
7,3      8 / 20 /  49 fps -- Score = 20

Bridge Of Fate
0,0     19 / 60 / 117 fps -- Score = 59
5,3     18 / 58 / 116 fps -- Score = 57

Rankin
0,0     19 / 44 / 119 fps -- Score = 43
5,3     19 / 45 / 134 fps -- Score = 45
From the beginning I noticed that AF had almost no impact on performance, whatever value was used. I then concentrated on FSAA. As you can see, given my setup, performance remains the same with FSAA set to any value from 0 to 5 included. Performance drops dramatically, though, with FSAA set to 6 or 7. Since the framerate changes so little, I assume my CPU is the limiting factor.
I didn't bother testing every possible value with the two other maps, since it was obvious that setting FSAA to 5 (4x Gaussian Multisampling) and AF to 3 (Maximum anisotropic filtering) was the best setup, since it significantly improved image quality with virtually no loss in overall performance. Note that I'm talking about FSAA here: I can't tell the difference between two images with and without AF enabled. Maybe I would if I was told what to look at :-/

The three maps are quite different from one another, and the tests confirmed that the above mentionned settings, given my hardware, were clearly the best to use. It is also worth noting that decreasing screen resolution had no impact either on performance, which confirms that my CPU is the limiting factor, as far as I know; please correct me if I'm wrong!
skamp is offline   Reply With Quote