Jump to content


Halo Performance with Universal Binary


  • Please log in to reply
73 replies to this topic

#21 a2daj

a2daj

    Uberspewer

  • IMG Pro Users
  • PipPipPipPipPipPipPip
  • 3400 posts
  • Pro Member:Yes

Posted 18 August 2006 - 04:22 PM

Lens flares shouldn't be a performance problem these days for folks running 10.3.5 or later and the latest versions of Halo.  The necessary OpenGL functionality to allow for hardware accelerated lens flares was introduced in 10.3.5 and the latest versions of Halo added that support.  With the limited testing I did when the last PPC Halo update came out I saw negligible performance differences with lens flare set to Low vs Extreme.  But considering many previous benchmarks used the Low setting, sticking to Low might be a good idea for consistencies sake with your previous benchmarks.  But I think for testing in this thread, QS's suggested test settings are pretty good since not everyone can run 1920x1200.

edit: I should note that I haven't done any comparisons with the new UB build concerning the lens flares so I might revisit those.  If the app was a significant rebuild then who knows what might have changed.

edit2:  OS update was 10.3.5.
Dual 2.5 GHz G5-RADEON X800 -4 GB RAM-Revo 7.1
MBP 2.0GHz -Mobility RADEON X1600-2 GB RAM

#22 teflon

teflon

    Bastard of the Popeye Analogy

  • Members
  • PipPipPipPipPipPipPipPipPipPipPipPip
  • 9589 posts
  • Location:London, UK

Posted 18 August 2006 - 04:53 PM

just for a quick comparison to the 1920x1200 scores here, and to show what effect the UB has on PPC performance, previous tests  from BareFeatsusing 1.5.2 got the following:

Quad-Core G5 with GeForce 6600 GT = 18 fps (No change)
Quad-Core G5 with GeForce 7800 GT = 60 fps (5FPS decrease with the upgrade)

Looking at that, im guessing it means that the game is GPU limited with the 6600GT, but the UB introduces some new CPU speed hit on the CPU. Interesting, no?
Polytetrafluoroethylene to my friends.

Macbook Pro - C2D 2.4Ghz / 4GB RAM / Samsung 830 256GB SSD / Geforce 8600M GT 256Mb / 15.4"
Cube - G4 1.7Ghz 7448 / 1.5GB RAM / Samsung Spinpoint 250GB / Geforce 6200 256Mb
Self-built PC - C2Q Q8300 2.5Ghz / 4GB RAM / Samsung 830 256GB SSD / Radeon 7850 OC 1GB / W7 x64
and a beautiful HP LP2475w 24" H-IPS monitor

#23 iRolley

iRolley

    Legendary

  • Members
  • PipPipPipPipPip
  • 752 posts
  • Location:Antibes - France

Posted 18 August 2006 - 05:09 PM

- 1280 x 800
- no FSAA
- Almost all maxed

Timedemo gives 39.37 fps

Actual game play is very good. Halo has never looked so good. Only the textures on the human characters are a bit dated, but the rest of the game is gorgeous. I am very happy with the port, Macsoft did an amazing job and once again we can realize how long Bungie's games can last. Halo is still very fun to play, the enemy AI is still entertaining.

#24 a2daj

a2daj

    Uberspewer

  • IMG Pro Users
  • PipPipPipPipPipPipPip
  • 3400 posts
  • Pro Member:Yes

Posted 18 August 2006 - 05:55 PM

I did some tests using maxed out settings @ 1024x768 with Lens Flare set to Low and Extreme.  With version 1.5.2 there's no difference. in performance.  With the UB build Low Lens Flare saw identical performance to version 1.5.2.  With LF Extreme, I saw the noticeable performance drop I saw at the lower resolution (Low 50s fps to ow 40s fps this time).  So, it would seem that the hardware accelerated lens flares code change may not have made it into the UB build, or for whatever reason, the LF gets tossed off the fast path.  That's a bummer.  Time to send in a note to MacSoft.
Dual 2.5 GHz G5-RADEON X800 -4 GB RAM-Revo 7.1
MBP 2.0GHz -Mobility RADEON X1600-2 GB RAM

#25 AV8

AV8

    Newbie

  • Members
  • 2 posts

Posted 19 August 2006 - 02:18 AM

View Postmrimac00, on August 17th 2006, 08:23 AM, said:

If you pause, you can do command-tab.

How do you pause?  If I pause by pressing ESC, I still cannot Command-Tab.  However, if I play the cam in an individual window, Command-Tab works fine.

#26 rob_ART

rob_ART

    Bare Feats

  • Members
  • PipPipPipPipPip
  • 792 posts
  • Location:Portland, OR

Posted 19 August 2006 - 06:27 AM

View Postteflon, on August 18th 2006, 10:53 PM, said:


Quad-Core G5 with GeForce 7800 GT = 60 fps (5FPS decrease with the upgrade)


That may not be the case.

The original published number was the second run.

The number I posted here for the UB version was the first run which is always slower.

I'll rerun the four examples and this time make two runs each.
rob-ART morgan
mad scientist
BareFeats.com

#27 rob_ART

rob_ART

    Bare Feats

  • Members
  • PipPipPipPipPip
  • 792 posts
  • Location:Portland, OR

Posted 19 August 2006 - 11:57 AM

View Postrob_ART, on August 18th 2006, 10:05 PM, said:

Halo UB tests at 1920x1200 and every setting at max except lens flare set to low:
Quad-Core G5 with GeForce 6600 GT = 18 fps
Quad-Core G5 with GeForce 7800 GT = 55 fps
Quad-Core G5 with GeForce 7800 GTX = 72 fps
Mac Pro 3GHz with GeForce 7300 GT = 22 fps

I reran the tests and recorded the second run instead of the first.

Halo UB tests at 1920x1200 and every setting at max except lens flare set to low:
Quad-Core G5 with GeForce 7800 GT = 60 fps (as earlier published on BareFeats.com)
Mac Pro 3GHz with GeForce 7300 GT = 23 fps

Halo UB tests at 1280x800 and every setting at max except lens flare set to low:
Quad-Core G5 with GeForce 7800 GT = 88 fps
Mac Pro 3GHz with GeForce 7300 GT = 38 fps

The Radeon X1900 XT is due to arrive in about a month I'll update you then with those numbers.
rob-ART morgan
mad scientist
BareFeats.com

#28 teflon

teflon

    Bastard of the Popeye Analogy

  • Members
  • PipPipPipPipPipPipPipPipPipPipPipPip
  • 9589 posts
  • Location:London, UK

Posted 19 August 2006 - 12:08 PM

cool, so there isnt a performance hit then...

the other good thing from those benches is that it shows that the 7300 GT is similar to a 6600 GT.

dont you get bored of the benchmark videos? ;)
Polytetrafluoroethylene to my friends.

Macbook Pro - C2D 2.4Ghz / 4GB RAM / Samsung 830 256GB SSD / Geforce 8600M GT 256Mb / 15.4"
Cube - G4 1.7Ghz 7448 / 1.5GB RAM / Samsung Spinpoint 250GB / Geforce 6200 256Mb
Self-built PC - C2Q Q8300 2.5Ghz / 4GB RAM / Samsung 830 256GB SSD / Radeon 7850 OC 1GB / W7 x64
and a beautiful HP LP2475w 24" H-IPS monitor

#29 Quicksilver

Quicksilver

    Verbal Windbag

  • IMG Writers
  • 4227 posts
  • Location:Chicago Illinois
  • Pro Member:Yes

Posted 20 August 2006 - 07:09 AM

There is a performance hit on non-Intel systems, and a huge one (I'm currently looking at a drop in timedemo performance of ~30%, which jumps to 40%+ if you use AF and FSAA).  There are also a number of huge bugs that were not present in Halo 1.5.2.  If robATI's benchmarks aren't messed up, then the question that I need to answer is whether or not it's limited to single and dual processor G5s, ATI Radeon cards, or just the X800XT graphics card.
Former Senior Hardware Editor
InsideMacGames.com

#30 teflon

teflon

    Bastard of the Popeye Analogy

  • Members
  • PipPipPipPipPipPipPipPipPipPipPipPip
  • 9589 posts
  • Location:London, UK

Posted 20 August 2006 - 07:28 AM

Rob_Art shows that the quad core with 6800 runs at the same speed. He had to do a second run to get that though, so have you taken that into account?

either way, the hit that I thought was there earlier (55FPS compared to 60 in 1.5.2) is nowhere near the 30% youre saying...

So it could be an ATI thing...
Polytetrafluoroethylene to my friends.

Macbook Pro - C2D 2.4Ghz / 4GB RAM / Samsung 830 256GB SSD / Geforce 8600M GT 256Mb / 15.4"
Cube - G4 1.7Ghz 7448 / 1.5GB RAM / Samsung Spinpoint 250GB / Geforce 6200 256Mb
Self-built PC - C2Q Q8300 2.5Ghz / 4GB RAM / Samsung 830 256GB SSD / Radeon 7850 OC 1GB / W7 x64
and a beautiful HP LP2475w 24" H-IPS monitor

#31 Quicksilver

Quicksilver

    Verbal Windbag

  • IMG Writers
  • 4227 posts
  • Location:Chicago Illinois
  • Pro Member:Yes

Posted 20 August 2006 - 07:32 AM

It could be--but, keep in mind that I've done more Halo timedemos over the years than I've seen online combined.  I did three back-to-back runs of each version of Halo (just so you know my figures are accurate), and all of my tests were performed with lens flare set to extreme--there's no reason to run the timedemos with lens flare set to low since Macsoft updated Halo to work around the lens flare performance hit (Halo 1.0.5.3 and earlier).
Former Senior Hardware Editor
InsideMacGames.com

#32 ozzy

ozzy

    Who?

  • Members+
  • PipPipPipPipPip
  • 509 posts
  • Steam Name:ozzy
  • Location:London, UK

Posted 20 August 2006 - 08:43 AM

I think someone mentioned earlier in this thread that the lens flares might be broken again in the UB version.  Try turning off Lens flares in the UB version and see if your performance problem goes away.

#33 Quicksilver

Quicksilver

    Verbal Windbag

  • IMG Writers
  • 4227 posts
  • Location:Chicago Illinois
  • Pro Member:Yes

Posted 20 August 2006 - 09:36 AM

Lens flare is not broken.  Switching from "Extreme" to "Low" lens flare increases timedemo performance on my primary system by about 3%, which is as it should be.
Former Senior Hardware Editor
InsideMacGames.com

#34 a2daj

a2daj

    Uberspewer

  • IMG Pro Users
  • PipPipPipPipPipPipPip
  • 3400 posts
  • Pro Member:Yes

Posted 20 August 2006 - 11:40 AM

My initial 1.5.2 tests were flawed somehow.  The were slightly slower than they should have been.  I still can't figure out how exactly.  Maybe it was too hot in my room at the time the tests were run.  That's actually a reoccuring theme this summer.  I can't spend much time on my G5 for temp reasons.  My room is much cooler now and I'm getting higher numbers with 1.5.2.  So I'm seeing a performance hit with the UB version now, but not the 30% QS is seeing.  More like 13% performance hit.

I can also confirm my findings with Lens Flares.  They do affect performance.  About an 18% performance penalty on my G5 and a 21% hit on my MBP.  So I still stand by my claims that Lens Flares is causing more of an impact than it should be.
Dual 2.5 GHz G5-RADEON X800 -4 GB RAM-Revo 7.1
MBP 2.0GHz -Mobility RADEON X1600-2 GB RAM

#35 bobbob

bobbob

    Uberspewer

  • Members
  • PipPipPipPipPipPipPip
  • 3367 posts

Posted 20 August 2006 - 02:00 PM

Are you guys using the same settings?

#36 iRolley

iRolley

    Legendary

  • Members
  • PipPipPipPipPip
  • 752 posts
  • Location:Antibes - France

Posted 20 August 2006 - 03:18 PM

[from the left field] 

Halo's fov is irritating. After playing Quake 4 and COD2, Halo FOV seems very small. But still I am glad it supports widescreen. Very nice.

As for performance, on my iMac CD 1024 - 640 all maxed +FSAA gives a very nice performance.

#37 Gloops

Gloops

    Heroic

  • Members
  • PipPipPipPip
  • 435 posts
  • Location:Herts, UK

Posted 21 August 2006 - 01:49 PM

Runs acceptably at 1024x768 on my 2 GHz Intel iMac, but there seems to be a bug with grenades. Explosions turn everything into a slideshow until the dust settles. Plasma frags are fine though.
What, Me worry?

#38 Mister Mumbles

Mister Mumbles

    Uberspewer

  • Members
  • PipPipPipPipPipPipPip
  • 2511 posts
  • Location:Not here; not there; not anywhere!

Posted 21 August 2006 - 02:33 PM

It plays just fine for me on the same iMac. Played it with a setting of 1280x800, though.
Formerly known as a Mac gamer.

#39 Gloops

Gloops

    Heroic

  • Members
  • PipPipPipPip
  • 435 posts
  • Location:Herts, UK

Posted 21 August 2006 - 03:56 PM

View PostPegasus, on August 21st 2006, 09:33 PM, said:

It plays just fine for me on the same iMac. Played it with a setting of 1280x800, though.
I had 4x FSAA turned on. No FSAA makes explosions run a lot better! Strange, because nothing else was causing slowdowns.
What, Me worry?

#40 Eric5h5

Eric5h5

    Minion Tormentor

  • Members
  • PipPipPipPipPipPipPipPipPipPipPipPip
  • 7160 posts

Posted 21 August 2006 - 09:51 PM

View PostGloops, on August 21st 2006, 05:56 PM, said:

I had 4x FSAA turned on. No FSAA makes explosions run a lot better!

That's what used to happen to me with an older version of Halo/OS X/my video card.  All three of those got upgraded, and at some point one or more of those items meant I could turn on FSAA and the explosions would not go all slideshowy.  Shame the slideshowyness made a comeback...that's progress for ya....

--Eric