Slow Sauerbraten

by hfdsjkconorhfdjs on 03/01/2008 02:07, 59 messages, last message: 06/11/2008 18:06, 26514 views, last view: 04/19/2024 00:38

For some reason, sauerbraten runs really slow.

I run on these settings:

720x480x32

0XAA

Shaderdetail 3

grass 1

waterrefract 1

waterrflect 1

shadowmap 1

shadowmapsize 7


I get very low fps whenever I play.

Vsync is off, maxfps is 500.

Now, you may say I need a new CPU/GPU/more ram, but here are my specs...


MacBook Pro
nVidia GeForce 8600M (128MB) maybe 256MB
2GB ram
Intel Core 2 Duo 2.2Ghz
Mac OS X 'Tiger' 10.4.11

I have the latest System Software, and sauerbraten is up to date. I also made sure that it wasn't using rosetta to emulate PowerPC.


Cheers,
Conorkirk

   Board Index   

#40: ..

by absinth_grrr on 05/27/2008 16:57

>using the default settings, shaders etc,
> what fps do you get in metl4

start position 40-55fps
whole map: 20-130fps

the same map gives nearly 500fps in win32 (but this is last release, not cvs)

> vs what fps do you get in an empty /newmap ?
500.
but just adding a few cubes using the scroll-wheel instantly lowers the fps. down to 120 after adding cubes for one minute.

reply to this message

#41: ..

by absinth_grrr on 05/27/2008 17:19

ok i mailed you the shark analysis.

the weird thing is also that the cpu usage is quite high, 95-99%.

on a iMac with same processor speed but with the "much slower" ATI Radeon 2600 Pro the same map gives 100-200 fps, with a cpu usage of ~85%.

still quite high - i remember sauerbraten only using 50% at most.

reply to this message

#42: ..

by absinth_grrr on 05/27/2008 17:22

>Er, sorry, I meant "/floatvtx 1".

yes, much, thx.

100-240fps on metl4 instead of 20-130.

not quite the windows scores, but definitely acceptable.

should we force floatvtx 1 for the geforce 8800 on osx for the next release? what about other nvidias?

reply to this message

#43: Re: ..

by absinth_grrr on 05/27/2008 17:36, refers to #43

no disabling VBOs + floatvtx 1 is worse than floatvtx 1 alone.

reply to this message

#44: ..

by absinth_grrr on 05/27/2008 18:38

baby rabbit testes macbook pros which contain 8600Ms while i have a 8800GT.

according to aspyr (the mac game porting house) the 7300 and 8800 have particular bad driver. that's why those are unsupported in neverwinternights 2, while the 7600 and 8600 are supported.

reply to this message

#45: Re: ..

by a~baby~rabbit on 05/27/2008 23:59, refers to #44

so the question is whether conors issue with his 8600 is the same, he seemed to have it far worse... hfdsjkconorhfdjs??

reply to this message

#46: Apple update, speedups

by fhdshfkjconorfhsdkj on 05/28/2008 04:34

There was an Apple update that sped it up, I can't right now, but I will give you some numbers in a little when I am at the macboo pro.

reply to this message

#47: ..

by absinth_grrr on 05/28/2008 19:55

note, 10.5.3 doesn't change anything here.

reply to this message

#48: Re: ..

by absinth_grrr on 05/28/2008 20:09, refers to #47

well, it changes something after all.

performance with "floatvtx 1" is much better, 400-500fps at metl4.

without "floatvtx 1" it is still complete crap.

reply to this message

#49: ..

by absinth_grrr on 05/28/2008 20:15

diff -r1.444 rendergl.cpp
280a281,285
>
> #ifdef __APPLE__
> if(strstr(renderer, "8800") )
> floatvtx = 1;
> #endif

reply to this message

#50: Better

by kdfjshsdhkfjconorfhdsjk on 05/29/2008 05:44

It seems way better, but still not near Windows

http://conorkirkpatrick.com/uploads/perf.zip

reply to this message

#51: ..

by absinth_grrr on 05/29/2008 13:54

>struct vertex : vertexff { bvec n; short reserved; };

doesn't make much of a difference :-<

>It seems way better, but still not near Windows

so did you test if "floatvtx 1" does speed things up for you too?

reply to this message

#52: Re: ..

by kdfjshsdhkfjconorfhdsjk on 05/29/2008 15:39, refers to #51

floatvtx 1 does speed things up, and those screenshots were from it

reply to this message

#53: Re: ..

by absinth_grrr on 05/29/2008 18:49, refers to #52

hm 30fps on metl4 is already faster? how much do you get without floatvtx 1?

reply to this message

#54: ..

by absinth_grrr on 05/29/2008 22:48

hard to say since "floatvtx 1" in 10.5.3 already maxes out at 500fps (@default / metl4). still i think it provides a minor additional boost. btw i wished for a benchmark mode in sauerbraten more than once, this would greatly simplify this - especially since the fps vary quite much with minor position&rotation changes. should be easy now that there is a demo-replay system?

reply to this message

#55: ..

by absinth_grrr on 06/03/2008 15:18

>Try going into a map like hades or k_rpg1

ok i don't think float UVs make an additional difference - one of the spawnpoints in hades gives exactly 297 fps either way.

reply to this message

#56: ..

by absinth_grrr on 06/03/2008 23:50

that has done the trick!

it is the same speed as "floatvtx 1" now!

great work ;)

reply to this message

#57: ..

by absinth_grrr on 06/04/2008 12:47

>And you are sure it was set to 0 when you tested? :)

yes, i usually triple-check things ;-)

>Please update source from CVS (and remove any of your changed files),
>and check if the "floatvtx 0" stuff works by default now.

yes it does. speed is fine now with floatvtx 0, i.e. default.

thanks very much for taking care of this issue.

reply to this message

#58: ..

by absinth_grrr on 06/05/2008 02:35

btw. whats to learn from this? VBOs of shorts need to be 16-byte aligned for current mac nvidia drivers?


maybe you'll have more luck reporting this problem to apple ;)

reply to this message

#59: Speed

by prophile on 06/11/2008 18:06

The majority of renderers on Mac OS X work exclusively in floats apart from colours, so for tex coordinates or vertices you should always be using floating-point data for maximum efficiency. Particularly with VBOs, since they'll need to be converted to floats on the fly if they are not already in that format, which is done by the CPU and basically defeats the purpose of using VBOs in the first place.

reply to this message

   Board Index