2.3 is touted to have some big performance improvements in rendering. Just to see if the 10-500% is realistic, why not take some timedemos (suggested demo1 as it's pretty intensive) for 2.2.3 and 2.3 and post the results? The mean result should be used as the min and max are too variable. Settings are arbitary but need to be equivalent across the different versions. Post your graphics card model too so people have some idea what they may get from their hardware.

I've got a 6600GT AGP 128Mb.
2.2.3: 28fps
2.3: 41fps
Giving 46% improvement

Anyone getting +500% I wonder?

stand on a certain map and looking at a certain direction. Then its very possible that its 500-1000% more fps
"tdem demo1" is an other matter me thinks. I get about 50% more fps

Ed wrote:Anyone getting +500% I wonder?

Don't expect to get +500% on average... But nevertheless on some maps this is realistic in certain spots. Loof (v1) is pretty much the worst case:


As you can see it's 14fps with 2.2.3 vs. 186fps with 2.3 with the same settings, that's roughly +1200%

Momentary increases in framerate may be enormous but you can't get any very accurate numbers for them and they don't give a good reflection on improvement during normal gameplay, this is what timedemos are for.

I've now done a comparison on a laptop with Dell integrated graphics running the free drivers under Linux. This system is pretty poor for Nexuiz so graphics settings are pretty low. Even so, 2.3 gives a worthwhile performance improvement:

2.2.3: 9.2895218
2.3: 14.3845882
That's a 55% improvement in average framerate.

I also found during testing that this system performed slightly better (statistically significantly) with gl_vbo 0.

Well it just depends on the map.. with demo1 i get 50% more in 2.3. From around 90 to 145fps.. demo5 (when i copy it into 2.2.3) raises from 100 to 200fps so 100%.. the biggest improvement i found in a single spot was in a map from the q3mappack from 35 to 310fps.
And thats at my usual LOW play settings. At normal or higher settings the raise can be move extreme in %

Have the notoriously high-polygon models been fixed? Is that off the list of things I should gripe about.

Since around 1.5 I guess, I'd say the performance increase is very signifigant. I can get playable FPS on pretty much any map, with everything turned on. I generall just stick with the default settings tho.

Workaphobia wrote:Have the notoriously high-polygon models been fixed? Is that off the list of things I should gripe about.

No, its hard to find moddelers.. its still 'planned' though

7950GT512
Everything on & at max except for HDR, Bloom & realtime shadows:
2.2.3: 111.6956327 fps
2.3: 132.2766175 fps,

... So that's about 18% better, which is well appreciated. This has allowed me to enable realtime shadows while still experiencing better frame rates than previously.
The big performance difference is the reduced slowdowns, I read somewhere that multithreading was enabled for this release, is this correct?

zenwalker wrote:The big performance difference is the reduced slowdowns, I read somewhere that multithreading was enabled for this release, is this correct?

Only for osx and only OpenGL calls are handled in a seperate thread by the driver if this option is enabled. On any other OS it's up to the driver whether it uses multithreading or not.

zenwalker wrote:The big performance difference is the reduced slowdowns

If you mean the min/max output from the timedemos.. the measurement has changed.. the old 2.2.3 output was 'absolute' min/max values.. like look at a wall for a split second and get 800fps or if some shader has to compile in the beginning and you get 30fps because of that. The new are 1-second averages for min/max like other games do.. those are higher for min and lower for max but kinda more useful..

esteel wrote:If you mean the min/max output from the timedemos.. the measurement has changed.. the old 2.2.3 output was 'absolute' min/max values.. like look at a wall for a split second and get 800fps or if some shader has to compile in the beginning and you get 30fps because of that. The new are 1-second averages for min/max like other games do.. those are higher for min and lower for max but kinda more useful..

The old measurements were very inaccurate, I did come up with a way of calculating uncertainties, this is now out of date:
http://www.forums.alientrap.local/viewtopic.php?t=952

esteel wrote:

Workaphobia wrote:Have the notoriously high-polygon models been fixed? Is that off the list of things I should gripe about.

No, its hard to find moddelers.. its still 'planned' though

I think you should try to get at least one model fully optimized, then make it the default model, and add a client option to force all models to show up as the optimized one. Not the best fix but it would work great until more/all models can be fixed.

esteel wrote:

zenwalker wrote:The big performance difference is the reduced slowdowns

If you mean the min/max output from the timedemos.. the measurement has changed.. the old 2.2.3 output was 'absolute' min/max values.. like look at a wall for a split second and get 800fps or if some shader has to compile in the beginning and you get 30fps because of that. The new are 1-second averages for min/max like other games do.. those are higher for min and lower for max but kinda more useful..

That's not exactly what I meant, though I did think it backed up my experience. Things used to get really bogged down on a few maps & fps could drop to single figures when things got hectic. I've not come across this so far with 2.3 & that's with a higher level of effects

That part of your experience comes from the general engine improvements, it handles maps much better now. What i wrote is really just about timedemos.

The general look and framerate in 2.3 does seem better, but I'm experiencing a *long* load time compared to 2.2.3.

Any ideas as to why this would be the case?

Do you mean the first startup? Or from map to map?
For the first startup its the menu that should be reworked to be less of a burden at startup

After more play I've found that a couple of maps still kill performance. I took screenshots so you know which ones & their frame rate. That's playing at ultra -bloom which produces a timedemo result of:

Code: Select all

result 1910 frames 15.7278011 seconds 121.4410069 fps, one-second min/avg/max: 73 122 153

Turning off realtime shadows has little/no impact on these two & when the action hots up frame rates drops further, forcing me to reduce quality. While playing, XP users often say performance is fine, I've only heard fellow Linux users mention this hit. Some people on the ubuntu forums have compared the performance between linux & windows, showing Linux to be faster, this doesn't make sense.

http://ubuntuforums.org/showpost.php?p= ... stcount=55

esteel wrote:Do you mean the first startup? Or from map to map?
For the first startup its the menu that should be reworked to be less of a burden at startup

Not from map to map, it's from when I click, say, "Instant Action" after starting nexuiz. I then sit, and wait, and sit, and make a cup of coffee, and then I get my "instant" action

After that, map to map is fine.

Seems like its the bots that create their waypoint system.. What kind of sytem do you have? Its basicly less then 2 seconds for me..