Compiling your maps with different settings.

While testing my map i sometimes noticed big jumps in FPS between compiles and I couldn't work out why. So i started compiling the map with different settings, taking screen shots of the same area, and changing between Normal and High GFX settings in nexuiz.

First screen shot: Full VIS, and Ambient Occlusion Lighting, Graphics level set to normal (125 fps)



Second screen shot: Full VIS, and Ambient Occlusion Lighting, Graphics level set to HIGH (112 fps) A small drop to be expected right?



Third screen shot: Fast VIS, FAST Light, Graphics set to Normal (248 FPS)
A big increase, to be expected right? The lighting is less defined, shadows are blurier but it doesnt look too bad.



Fourth screen shot: Fast VIS, FAST Light, Graphics set to HIGH (123 FPS) BIG drop in FPS between Normal and HIGH settings when using fast vis and fast light, much more than the fully compiled map above on HIGH and Normal settings.



Fith screen shot: Fast VIS, AMBIENT OCCLUSION Lighting, Graphics set to NORMAL (256 FPS) Now the map looks basically the same as the fully VISed map above but there is a HUGE jump in FPS in Normal GFX mode.



Sixth screen shot: Fast VIS, AMBIENT OCCLUSION Lighting, Graphics set to HIGH (140 FPS) Map is now being rendered in HIGH and there is that big FPS drop again.



What does this all mean exactly.... er I don't know. But it seems that it's better to compile at least THIS map with fast vis settings and full light because players that run the map on normal GFX are going to get a substantial FPS boost.

Anybody have any thoughts?

Cheers!
MintOX

errr, what the crap? Higher vis settings causes fps drops in game? Now that's some q3crap2 for you! Could you pack up the fast vis/normal vis versions (with ambient occlusion and high detail light etc) into separate pk3s so others could test? No idea why it happens otherwise from these screens...

<@divVerent> -fastvis chooses ANOTHER vis algorithm that USUALLY gives worse results <@divVerent> it can yield a better result on that particular scene

So yeah, as you said. Might be better in this case only.

I think maybe because the map is so open?

MintOX

To test and compare the performance of different map compiles, you get more accurate results if you make a demo playing around the whole map with some bots, and then run it as a timedemo (Multiplayer > Demos).

There are also some useful cvars for a closer inspection in-game:
r_speeds 1
r_showtris 1
r_showdisabledepthtest 1

02:37:10 <@LordHavoc> motorsep|linux: an unvised map in DP uses a different algorithm
02:38:01 <@LordHavoc> motorsep|linux: you can use r_useportalculling 2 to force it to do this on vised maps as well

Might be useful to always test your maps with that to see if it's faster, and add into the mapinfo if it is.

Edit: Still at least perform the fast vis, as I suppose it's needed for server-side entity culling. (and likely faster there...)

From the gtkradiant manual:

fastvis - Performs a bsp and a quick version of the vis process. The entire world is made into a single bsp partition. It's fast, but when you look in a particular direction, you see every single detail in that direction. Nothing is blocked. A light map is created. This mode was used for space maps.

Space maps.

MintOX