With the release of "sam440ep_setup" fix for Sam440s, which should speed-up graphics performance, I thought I'd runs some tests to see how it affects Quake 3. (This will repeat some stuff I posted on AW.net ages ago.)
First, how do we benchmark Quake 3?
* Open the console of Quake 3 (typically by pressing Tab).
* Type "timedemo 1" (without the quotes!).
* Type "demo four"
* Wait for the demo to complete.
* Return to the console to see the results (fps).
The following are the "best" settings I'd previously found (where I aimed for good looks without much speed penalty)! Here's what I use:
Quote:
GL Extentions=On (Off is no faster)
Video Mode = 800x600 (1024x768 is signficiantly slower)
Colour Depth = 16-bit (32-bit only makes it slightly slower, but I can't see the difference)
Lighting = Vertex (Lightmap is massively slower, and I can't see the difference)
Geometric Detail = Medium (Low gives no speed benefit, while High is significantly slower)
Texture Quality = 4 (max) (Worse settings are almost no faster)
Texture Depth = Default (This doesn't make much difference)
Texture Filter = Trilinear (Bilinear is not any faster)
On OS4.1 beta with Debug Kernel I got 18.2 fps for "demo four". The next post will contain what I currently get (with OS4.1 update 2) before/after using the "sam440ep_setup" fix.
For posterity (and to save me writing it down elsewhere!) here are the results of changing only one setting & then running the same demo:
Quote:
1280x1024 = -2.9fps (-16%), i.e. 15.3 fps
Lightmap Lighting = -2.6fps (-14%)
1024x768 = -1.6fps (-9%)
High Geo Detail = -1.4fps (-8%)
32-bit Depth = -0.8fps (-4%)
32-bit Textures = -0.2fps (-1%)
Low Geo Detail = +0.0fps (+0%)
GL Extensions Off = +0.0fps (+0%)
Bilinear Filtering= +0.0fps (+0%)
Texture Detail 3 = +0.1fps (+0%)
Texture Detail 2 = +0.2fps (+1%)
16-bit Textures = +0.2fps (+1%)
640x480 = +0.5fps (+3%)
Note that fps varied by at least 0.1 between tests, and so anything between -0.1 & +0.1 fps can be considered "no change", while -0.2 or +0.2 fps (1%) should not be construed as being significant due to the high chance of errors.