OK, so I finally managed to go ahead and run some equivalence timedemo tests on my hardware. I have a C2D MacBook Pro, and a CD MacBook, so before I paste in my results, let me break down the environment and caveats. First, just a quick response to
merlijn for a matter of curiosity.
merlijn wrote:In my case though, SDL gives me no boost in speed at all in Snow Leopard - both AGL and SDL are so slow that I can literally count the frames on timedemo demos/bench1.dem.
Which makes sense, considering it's a 32 player match. How many of those do you play? I played on a GMA 950 for a year and a half, I know what it's like. It sucks. And that's entirely in the realm of Intel and Apple for making the decision to use that hardware. I rarely was able to play with a stable framerate in matches above 8v8.
And I agree that performance always was pretty bad on that machine, but in Leopard I used to get at least 20-30 FPS.
But did you receive 20-30 FPS in demos/bench1.dem on Leopard?
Anyroad, here are my results. No significant regressions on the hardware I used, see "Caveats" for further thoughts.
Hardware:
9400M and 9600MGT:
2.4GHz Intel Core 2 Duo
2GB DDR3 PC3-8500
256MB 9400M integrated GPU
and 256MB 9600M GT discrete GPU
7200RPM SATA2 250gb hard drive.
Tested drive: 7200RPM PATA drive booted over USB 2.0
GMA950:
2GHz Intel Core Duo
1GB DDR2 PC2-6400
64MB Intel GMA 950
7200RPM SATA hard drive (I don't even remember its capacity or anything else).
Environment:
1. For the higher resolutions, I tested both medium and low presets. For the lower resolutions(800x600 and 640x480), I simply tested low.cfg.
2. I tested native resolution, default (1024x768), and lowest resolutions to have a pragmatic test for laptop gaming.
3. A clean copy of nexuiz-251.zip on a fresh user.
Caveats:
1. For all the tests, sans the GMA 950, these were conducted on clean installs of Mac OS X 10.5 and 10.6 running off a
external hard drive. 10.5 was upgraded to 10.5.8 with all associated software updates installed. As I was uncertain if
divVerent still used the MacBook, I didn't want to touch it. Maybe
merlijn can run similar tests on his SL install and see how they jive with my numbers.
I had to boot off an external because my MBP is my vital machine, and I had no time nor will to dink around with my current workflow. I might invest in a swappable 2.5 sata drive in the near future, which would help immensely for this.
2. timedemo and darkplaces as a whole, I think, is not really meant to be a strict system-wide benchmark. Plus bench1.dem seems like a worst case scenario of a 32 person match. But it does provide, at the very least, a snapshot of how actual gaming experience should be.
3. I haven't made a pretty graph on these. Only ascii table for now.
4. I'd like to know more about Phoronix Test Suite, since its numbers vary so wildly from Darkplaces.
(L and SL stand for Leopard and Snow Leopard)
(/m stands for med.cfg and /l stands for low.cfg)
- Code: Select all
---------------------------------------------------------------------------------------------
| 9400M | 9600M GT |
| AGL | SDL | AGL | SDL |
| L | SL | +/- | L | SL | +/- | L | SL | +/- | L | SL | +/- |
1440/m | 25 | 23 | -2 | 34 | 29 | -5 | 26 | 27 | +1 | 37 | 37 | +0 |
1440/l | 37 | 28 | -9 | 43 | 39 | -4 | 41 | 42 | +1 | 52 | 51 | -1 |
1024/m | 27 | 25 | -2 | 37 | 33 | -4 | 27 | 28 | +1 | 39 | 37 | -2 |
1024/l | 40 | 38 | -2 | 49 | 46 | -3 | 43 | 44 | +1 | 55 | 55 | +0 |
800/l | 43 | 41 | -2 | 49 | 52 | +3 | 43 | 45 | +2 | 57 | 56 | -1 |
640/l | 44 | 43 | -1 | 53 | 54 | +1 | 44 | 45 | +1 | 56 | 57 | -1 |
----------------------------------------------------------------------------------------------------
Avg | 36 | 33 | -3 | 44 | 42 | -2 | 37 | 39 | +1 | 49 | 48 | -1 |
- Code: Select all
-----------------------------------------------
| GMA950 |
| AGL | SDL |
| L | SL | +/- | L | SL | +/- |
1280/m | 4 | | | 5 | | |
1280/l | 9 | | | 7 | | |
1024/m | 6 | | | 5 | | |
1024/l | 11 | | | 11 | | |
800/l | 13 | | | 11 | | |
640/l | 13 | | | 19 | | |
------------------------------------------------------
Avg | 9 | | | 10 | | |
I was slightly surprised by these numbers, to say the least, as AGL has always
felt slower than SDL. Likewise, I could have sworn that, by just feel, Snow Leopard did feel slightly slower. I do plan on getting around to doing equivalent timedemos on Windows SDL and WGL builds and, if I can, some SDL and GLX timedemos in Linux on my MBP once I get an internal drive.
I would expect that these numbers will be higher in Windows across the board -- I'm not certain about Linux, that will probably depend upon a few factors. Namely proprietary vs open source drivers.
For reference here are all my captures from each test instance:
composite list:
http://pastie.org/private/47oowugwaed7n9ziubkq
Leopard GMA950 AGL
http://pastie.org/private/xskq6iwaoqa1nrbzfgqsa
Leopard GMA950 SDL
http://pastie.org/private/xkltr72kkxcgoowbygkwuw
Leopard 9400M AGL
http://pastie.org/private/mrnmutvstwleugeznhdea
Leopard 9400M SDL
http://pastie.org/private/kxd2d8opyiotrpogex0pq
Leopard 9600M GT AGL
http://pastie.org/private/flmjfnh4kbfiigr4iy0q
Leopard 9600M GT SDL
http://pastie.org/private/naulcuinu1thhfs9dnjk5w
Snow Leopard 9400M AGL
http://pastie.org/private/zfwmipcgtjhwve8vqxr6g
Snow Leopard 9400M SDL
http://pastie.org/private/bsr0ztoirsnrzmkkhzza9q
Snow Leopard 9600M GT AGL
http://pastie.org/private/uzxo5ppvs08fqk9b7hlhmw
Snow Leopard 9600M GT SDL
http://pastie.org/private/hljamx4cteymqodgqevkjw