Dual Processor PowerMacs slow Doom III.

[Little Endian]

While I was experimenting with Doom III settings and benchmarking with the timedemo I realized that if I disabled my second CPU on my Dual 2.5Ghz G5 I could get a 33% performance bump. I used the CHUD tools to disable the second processor.

With two 2.5Ghz Chips running with my Geforce 6800GT I was able to get 32.7 FPS at 1680x1050 resoultion High settings with all Advanced options on except for Vsync and FSAA. Aniso was set to the standard 8x. However upon disabling the second processor I get 43.2 FPS at the exact same settings and resolution. I can even bump up to Ultra Quality settings and still get 42 FPS in the timedemo!

I know it sounds weird and it is weird because I only get this bump from disabling the the second CPU at a resolution of 1680x1050 and with FSAA off. Having one or two proccessors on made no difference if I had FSAA running at any resolution or setting. Perhaps folks with a Dual CPU configuration machine can check for me as well. I am interested if others see this at resolutions higher than 1680x1050 like the standard 1600x1200 or at 1920x1200. I got no speed improvement from disabling the processor at lower resolutions and FSAA still kills performance but something is deffininitely not right.

[JeDiBoYTJ]

interesting observation...

my only guess, is that the game is *trying* to take advantage of the 2 CPUs in the game when its at higher resolutions, while *trying* to balance the load onto both CPUs... but since the game doesnt support it, the CPU cycles are lost.

just a quick guess, lol... but interesting observation none the less

[grabberslasher]

While I was experimenting with Doom III settings and benchmarking with the timedemo I realized that if I disabled my second CPU on my Dual 2.5Ghz G5 I could get a 33% performance bump. I used the CHUD tools to disable the second processor.

With two 2.5Ghz Chips running with my Geforce 6800GT I was able to get 32.7 FPS at 1680x1050 resoultion High settings with all Advanced options on except for Vsync and FSAA. Aniso was set to the standard 8x. However upon disabling the second processor I get 43.2 FPS at the exact same settings and resolution. I can even bump up to Ultra Quality settings and still get 42 FPS in the timedemo!

I know it sounds weird and it is weird because I only get this bump from disabling the the second CPU at a resolution of 1680x1050 and with FSAA off. Having one or two proccessors on made no difference if I had FSAA running at any resolution or setting. Perhaps folks with a Dual CPU configuration machine can check for me as well. I am interested if others see this at resolutions higher than 1680x1050 like the standard 1600x1200 or at 1920x1200. I got no speed improvement from disabling the processor at lower resolutions and FSAA still kills performance but something is deffininitely not right.

I bet it's because OS X by default is trying to split the load between the 2 processors, which the game probably doesn't like.

This way, you're getting 100% of one processor.

The deal with fps is that there's a slowdown in Apple's OpenGL so you're probably not going to get faster fps until OpenGL is changed.

[psycho bob]

This wouldn't be that suprising actually. It was discussed on another thread how poorly OS X actually handles the two CPU's and how tasks jump between them when it isn't necessary. That is why I'm so suprised people are jumping up and down at the thought of having 4 3GHz cores in their machines. Unless Tiger drastically improves the scheduler and some of the more basic functions of the OS then all that CPU power will just be wasted. Two G5's could perform so much better if apple put the time in to getting the code right.
Another question is who will actually write programs to take advantage of 4 cores, games developers can't be bothered to even enable SMP support for 2 CPU's so they won't and I can't see adobe doing it either at least not yet.
The main thing though about using both CPU's in a G5 is while playing Doom 3 you could also be working in Photoshop, perhaps on iChat maybe even listening to iTunes if you didn't want the full scary sinister dark gameplay factor. Dual machines rock but they could be so much better, for me a quad core machine would only provide bragging rights at this stage which is sad really :(

[ChrisFromCanada]

I would be happy to try that for you but I am not sure where the CHUD tools are and how to use them. Maybe you could write a small guide for those who are interested.

[grabberslasher]

I would be happy to try that for you but I am not sure where the CHUD tools are and how to use them. Maybe you could write a small guide for those who are interested.

http://www.macupdate.com/info.php/id/8506

Then there will be a little System Preference where you can turn off processors. Aptly called "Processor" ;)

[psycho bob]

I would be happy to try that for you but I am not sure where the CHUD tools are and how to use them. Maybe you could write a small guide for those who are interested.

First you need to have the developer tools installed on your machine.
You can download the latest CHUD tools from here http://developer.apple.com/tools/performance/
The link on the page is for an FTP connection it might say it cannot connect but then just appears anyway. The latest version is 4.1.1.
You must install the developer tools first though or it errors!! Once you've installed the CHUD tools (if you install from the developer CD that came with the OS have a hunt through the developer folder until you find CHUD then you should be able to install from there) there should be a hardware icon added to system preferences. From there you can disable the CPU's and play around. Remember these are developer tools though so read up first on what they can do and if your unsure don't click on anything.

[Dont Hurt Me]

Now wouldnt you have thought Aspyr,Apple,id would have discovered this? amazing! turn off 1 cpu and presto! All 3 of those guys have work to do that is clear. this flabergasts me. Good discovery by Little Endian.

[Soulstorm]

I haven't run any benchmarks... but I have noticed an approximately 30% increase in performance.

This is remarkably...inexplicable! I think all benchmarks should be run with one processor on, and not two!

[Little Endian]

I bet it's because OS X by default is trying to split the load between the 2 processors, which the game probably doesn't like.

This way, you're getting 100% of one processor.

The deal with fps is that there's a slowdown in Apple's OpenGL so you're probably not going to get faster fps until OpenGL is changed.

Yes that is exactly my conclusion as well. I am no professional when it comes to programming but something is surely wrong with how Doom III is being handled on Dual CPU machines. I decided to run Doom III in windowed mode with both activity monitor and terminal running. When both processors are running I noticed that Doom III sets my CPU monitor usage back and forth rapidly and that threads from Doom III are being bounced back and forth from CPU to CPU.

Mac OSX is not perfect in multi-tasking and scheduling of CPU usage but it's not supposed to be this bad either. Usually there is only a small 5%-10% speed penalty if any caused from overhead when running an App or Game that is not SMP aware on an MP system. In many other cases there is no overhead at all. In fact with most Games even ones that are not SMP aware they still peform equally or better on MP systems compared to a single CPU setup.

[patseguin]

I was running Doom 3 on my 6800 Ultra and it performed like crap. I took it out and put in my x800 again and it was TONS better. I had the same result with WoW.

[applekid]

Now wouldnt you have thought Aspyr,Apple,id would have discovered this? amazing! turn off 1 cpu and presto! All 3 of those guys have work to do that is clear. this flabergasts me. Good discovery by Little Endian.

My thoughts exactly. This is a surprising find and a good one, too. :cool:

Somebody needs to bring this up to Aspyr and Apple. Also, this is definitely newsworthy and should be reported to some Mac news websites.

[Lacero]

I'm going to try and benchmark it on my Dual G5 PM and report the results. I just have to figure out this CHUD thing... :confused:

[broken_keyboard]

I think ~loserman~ explained this on another thread. When OS X is swapping between the various programs that are running at any one time, it doesn't necessarily put the prog back on the same CPU it was last running on. So therefore you don't get the same cache benefits.

When you disable one CPU, it always puts it back on the same CPU where there is stuff cached from last time.

[Dont Hurt Me]

I haven't run any benchmarks... but I have noticed an approximately 30% increase in performance.

This is remarkably...inexplicable! I think all benchmarks should be run with one processor on, and not two!
did you run it at 1280 x 1024?

[TheGimp]

did you run it at 1280 x 1024?


That's what I want to know.

[JamSoft]

My Computer:
Power Mac G5
Dual 2.5 GHz
Memory: 4 GB

Graphics Card:
GeForce 6800 Ultra
VRAM (Total): 256 MB

Mac OS X:
System Version: Mac OS X 10.3.8 (7U16)

Mac System Settings / Options
Processor Performance: Highest
ALL User programs / apps were quit. Used Activity monitor to verify.

__________ Game Settings __________
• Video Quality: High Quality
• Screen Size: 1280 x 1024
• Fullscreen: Yes
• High Quality Effects: Yes
• Enable Shadows: Yes
• Enable Specular: Yes
• Enable Bump Maps: Yes
• Vertical Sync: No
• Antialiasing: Off

__________ My Custom Tweaks __________
– all .pak files were 'unzipped' / decompressed
image_useCache 1
image_cacheMegs 512
imagecacheMink 3072
____________________

Results:

Both CPU's On:
• First Run: 33.7 FPS
• Second Run: 37.1 FPS

Only 1 CPU On:
• First Run: 41.8 FPS
• Second Run: 46.7 FPS
____________________________________________________________


As you can see, by turning the 2nd CPU off, the results definitely increased. I still can't believe that I only average MAX 46.7 FPS with my computer. Such a shame.

I don't want to sacrifice video quality, so if there are any other tweaks / suggestions for boosting FPS or loading that DO NOT sacrifice quality — let me know. :cool:

Also, if anyone would like me to do any other new/different tests, just ask.

Thanks,
Nate

[Lacero]

As you can see, by turning the 2nd CPU off, the results definitely increased. I still can't believe that I only average MAX 46.7 FPS with my computer. Such a shame.

No kidding. On such a top of the line PM/Video card, I would expect at least 80 FPS at 1600x1200 and 55 fps with 4xAA.

[Dont Hurt Me]

My Computer:
Power Mac G5
Dual 2.5 GHz
Memory: 4 GB

Graphics Card:
GeForce 6800 Ultra
VRAM (Total): 256 MB

Mac OS X:
System Version: Mac OS X 10.3.8 (7U16)

__________ Game Settings __________
• Video Quality: High Quality
• Screen Size: 1280 x 1024
• Fullscreen: Yes
• High Quality Effects: Yes
• Enable Shadows: Yes
• Enable Specular: Yes
• Enable Bump Maps: Yes
• Vertical Sync: No
• Antialiasing: Off

__________ My Custom Tweaks __________
– all .pak files were 'unzipped' / decompressed
image_useCache 1
image_cacheMegs 512
imagecacheMink 3072
____________________

Results:

Both CPU's On:
• First Run: 33.7 FPS
• Second Run: 37.1 FPS

Only 1 CPU On:
• First Run: 41.8 FPS
• Second Run: 46.7 FPS
____________________________________________________________


As you can see, by turning the 2nd CPU off, the results definitely increased. I still can't believe that I only average MAX 46.7 FPS with my computer. Such a shame.

I don't want to sacrifice video quality, so if there are any other tweaks / suggestions for boosting FPS or loading that DO NOT sacrifice quality — let me know. :cool:

Also, if anyone would like me to do any other new/different tests, just ask.

Thanks,
NateThats a very big improvement. You should send this to Aspyr,id,& Apple. Still who would have thought to think different? ;)

[Little Endian]

I sent a message to Glenda Adams of Aspyr Media regarding this Dual Processor slow down and have yet to wait for a reply. I have also contacted Rob_Art Morgan of Barefeats.com and hopefully he will look into it and post his findings on his Mac benchmarking website.

[MacsRgr8]

Cool :cool:

I'm gonna test my setup too!

Dual 1.8
2 GB
6800 Ultra

Get back tonite (CET that is)

Nice one, Little Endian

[TheGimp]

Cool :cool:

I'm gonna test my setup too!

Dual 1.8
2 GB
6800 Ultra

Get back tonite (CET that is)

Nice one, Little Endian

Help! Chud won't install on my computer. It reports that it "can't be installed on this computer". How do I get around this?

[Soulstorm]

did you run it at 1280 x 1024?
Yes, I did, and the performance increase is 30%.

EDIT: No, I didn't I run it 1024x768 at Ultra High and all setting enabled except antialiasing and vertical sync.

Someone please, bring this up to Aspyr and Apple!!
Macologist already knows that issue...

[MacsRgr8]

It's the same for me!!

Here we go:

PowerMac Dual 1.8 GHz G5
2 GB RAM
nVidia GeForce 6800 Ultra
2 x 160 GB S-ATA HD RAID0

Doom 3 settings:
1600 x 1200
High Quality
Advanced default (= Shadows on, AA off etc)

"Timedemo demo1":

Dual CPU:
1st: 27.7 fps
2nd: 32.7

Single CPU:
1st: 28.3
2nd: 36.1

as these settings are a real burden on the grfx card, I'd say this is a huge increase... :cool:

[Lacero]

BUNK. I tried it on my G5 PM and noticed no increase in fps performance. It actually decreased. Sorry, but I think 30% increase is not possible from the results of my test.

[Little Endian]

BUNK. I tried it on my G5 PM and noticed no increase in fps performance. It actually decreased. Sorry, but I think 30% increase is not possible from the results of my test.

What is the exact specifications of your machine? Leaving FSAA on hurts performance either way and if I leave it on it makes no difference if one or two processors are on the results will be the same. Also the diffence is amplified at higher resolutions and settings. If you tested at something like 1027x768 and below and or in conjunction with medium or low settings the scores will be the same.

[broken_keyboard]

BUNK. I tried it on my G5 PM and noticed no increase in fps performance. It actually decreased. Sorry, but I think 30% increase is not possible from the results of my test.

Don't forget the 1 cpu speedup comes from better utilization of the cache, so you would need to run it twice to get a clear idea.

[Soulstorm]

BUNK. I tried it on my G5 PM and noticed no increase in fps performance. It actually decreased. Sorry, but I think 30% increase is not possible from the results of my test.
I think I mentioned that I didn't run any timedemo... This is my personnal "measurement". But 30% performance increase IS possible... and when I say performance I don't necessarily mean fps, but hickups also, and loading times. FPS increased by 20-25% and loading times decreased by 10%, hickups decreased by 10% also.

Either way, I'm talking about the general gaming experience the game leaves me with, and not just plain numbers. And I'm sure the performance has substantialy increased when I disabled the second CPU...

Anyway I would like to see your specs...

[Dont Hurt Me]

It's the same for me!!

Here we go:

PowerMac Dual 1.8 GHz G5
2 GB RAM
nVidia GeForce 6800 Ultra
2 x 160 GB S-ATA HD RAID0

Doom 3 settings:
1600 x 1200
High Quality
Advanced default (= Shadows on, AA off etc)

"Timedemo demo1":

Dual CPU:
1st: 27.7 fps
2nd: 32.7

Single CPU:
1st: 28.3
2nd: 36.1

as these settings are a real burden on the grfx card, I'd say this is a huge increase... :cool:you should run this again at 1280 x 1024 since thats native for most Lcds. This will give us a better comparison of the Cpu's vs JamSoft 2.5

[TheGimp]

I think I mentioned that I didn't run any timedemo... This is my personnal "measurement". But 30% performance increase IS possible... and when I say performance I don't necessarily mean fps, but hickups also, and loading times. FPS increased by 20-25% and loading times decreased by 10%, hickups decreased by 10% also.

Either way, I'm talking about the general gaming experience the game leaves me with, and not just plain numbers. And I'm sure the performance has substantialy increased when I disabled the second CPU...

Anyway I would like to see your specs...

Unless the performance improvement is reflected in timedemo results, it's doubtful that any improvement in average fps has been gained. "Subjective" could be a placebo effect.

[whooleytoo]

Also the diffence is amplified at higher resolutions and settings.

I'd find this surprising, I'd expect better CPU performance to improve framerates at all resolutions (or conversely, maintain the same framerates with higher poly counts and/or settings).

[Lacero]

I did the timedemo twice. There was a decrease in fps when switching to 1 processor. This whole topic is BUNK. Don't bother installing CHUD if you haven't already. It won't help in the slightest.

See my old thread (http://forums.macrumors.com/showthread.php?t=117242) on this.

[Little Endian]

I did the timedemo twice. There was a decrease in fps when switching to 1 processor. This whole topic is BUNK. Don't bother installing CHUD if you haven't already. It won't help in the slightest.

See my old thread (http://forums.macrumors.com/showthread.php?t=117242) on this.

Well Glenda Adams of Aspyr Media has observed the same thing. Maybe you should ask her since she is the Head of Development at Aspyr. You still have not posted your exact specifications on your machine. I have noted that those runnning with 9600 series cards won't see a difference anyhow since the GPU is a bigger bottleneck. If you are using a 9600 series card then I doubt you will see any difference.

[Soulstorm]

Well Glenda Adams of Aspyr Media has observed the same thing. Maybe you should ask her since she is the Head of Development at Aspyr. You still have not posted your exact specifications on your machine. I have noted that those runnning with 9600 series cards won't see a difference anyhow since the GPU is a bigger bottleneck. If you are using a 9600 series card then I doubt you will see any difference.

When did she said that? Is there an article or what? Links plz!

[MacsRgr8]

you should run this again at 1280 x 1024 since thats native for most Lcds. This will give us a better comparison of the Cpu's vs JamSoft 2.5

Well, I ran the timedemo demo 1 twice each time (Single vs. Dual) running at 1280 x 1024. (man... I can dream this timedemo :D )
Both timedemo's gave almost exactly the same results.

But these are surprising IMHO:

Dual CPU:

1st: 29.5 fps
2nd: 36.3 fps

Single CPU:

1st: 28.6 fps
2nd: 35.8 fps

So, hardly a difference this time... actually Single CPU slightly slower!

I ran the tests twice, both after rebooting.

In my case: Single CPU is faster than Dual enabled at 1600 x 1200, but no real difference running it at 1280 x 1024. Rest of settings are the same. All options "on" except AA.

[ZildjianKX]

Wow, Doom 3 runs bad enough on my SP 1.8 GHz G5 w/ a Radeon 9800 Pro, if it runs any slower with a 2nd processor I can't imagine how bad it would be.

[Santaduck]

There are apparently contradictory and varying results with the second processor. According to Glenda the second processor should handle audio. Some setups show a slowdown, whereas other setups show a speedup, by disabling the second processor. On a test with the DP 2.5 with the best cards (x800/6800U), it seems that a single processor is slower, not faster, in practically every case.

And hints from testers with Tiger seem to indicate that the DP/SP difference may change to nil.

Indeed, it will be interesting to see what comes of the next Doom3 patch, since Glenda has now heard of the issue. To me, I'm not too concerned with performance with their first release-- people are discovering interesting things, and it will only get better.

At any rate, the biggest benefits seem to be using: Medium quality, shadows disabled, and using the ATI Displays utility to manually disable vertical sync.



MacRgr8: You shouldn't really be listing the first run of the timedemo: it's caching information, and you can actually see it lag. After the second run, the numbers stabilize and are usually exactly the same, to the tenth of a frame per second.

To save time and get within 0.2 fps of the max value on the first run, you can also use: timedemo demo1 cache

[MacsRgr8]

MacRgr8: You shouldn't really be listing the first run of the timedemo: it's caching information, and you can actually see it lag. After the second run, the numbers stabilize and are usually exactly the same, to the tenth of a frame per second.

To save time and get within 0.2 fps of the max value on the first run, you can also use: timedemo demo1 cache

Cheers. Will do.

I find it interesting that at higher resolutions disabling the 2nd CPU seems to speed the game up (10 to 15%), whereas at medium resolutions enabling the 2nd CPU seems to speed it up a little in my case.

BTW, I have also tried unpacking the .pk4 files, but that had no influence. Maybe slower disks show better performance increases when these files are extracted?

My setup:
Dual 1.8 GHz G5
2.0 GB RAM
nVidia GeForce 6800 Ultra
2 x 160 GB S-ATA HD RAID0 (Apple SoftRAID)
LaCie photon20vision (1600 x 1200 rez native)
Disk Journaling Disabled (still trying to figure out when this makes any difference ;) )

[Santaduck]

Forgot to mention...

Also Glenda has told Rob-ART that the second processor is used for audio. Why is this significant?

Remember, the audio is disabled in the timedemo benchmarks.

Your actual in-game framerate may indeed suffer if you disable that second processor. You can try viewing your continuous framerate in-game to see if you can notice any difference (I would think it would be difficult to reliably get an average framerate from the running in-game display though).

[MacsRgr8]

Forgot to mention...

Also Glenda has told Rob-ART that the second processor is used for audio. Why is this significant?

Remember, the audio is disabled in the timedemo benchmarks.

Your actual in-game framerate may indeed suffer if you disable that second processor. You can try viewing your continuous framerate in-game to see if you can notice any difference (I would think it would be difficult to reliably get an average framerate from the running in-game display though).

Agreed.... but that should mean no difference, shouldn't it?
Comparing the timedemo with actual game play is not a good idea. But it still the best way to benchmark the Mac against other Macs.

Maybe we should all play the level "Hell" (WOW, that really is so cool!!), in god-mode, and post the average fps... after completing it?

[Jigglelicious]

Don't forget that when you want to remove the GPU from the equation and simply test CPU performance, you need to LOWER the resolution to the lowest possible (in this case, 640x480), NOT increase it. By increasing the resolution you further strain the GPU and would quickly negate any performance increase/decrease you might notice by disabling the 2nd CPU.

In other words, you want to keep this test CPU bound, not GPU bound. Lower the resolution, then test!

[Dont Hurt Me]

Well, I ran the timedemo demo 1 twice each time (Single vs. Dual) running at 1280 x 1024. (man... I can dream this timedemo :D )
Both timedemo's gave almost exactly the same results.

But these are surprising IMHO:

Dual CPU:

1st: 29.5 fps
2nd: 36.3 fps

Single CPU:

1st: 28.6 fps
2nd: 35.8 fps

So, hardly a difference this time... actually Single CPU slightly slower!

I ran the tests twice, both after rebooting.

In my case: Single CPU is faster than Dual enabled at 1600 x 1200, but no real difference running it at 1280 x 1024. Rest of settings are the same. All options "on" except AA.
Thanks, must be something wrong with the drivers. Its either the OS or its the Drivers. Maybe both.

[Santaduck]

Don't forget that when you want to remove the GPU from the equation and simply test CPU performance, you need to LOWER the resolution to the lowest possible (in this case, 640x480), NOT increase it. By increasing the resolution you further strain the GPU and would quickly negate any performance increase/decrease you might notice by disabling the 2nd CPU.

In other words, you want to keep this test CPU bound, not GPU bound. Lower the resolution, then test!

Better yet (just read the various reviews out there for mac) is to use the r_skiprender 1 console command. Visible 3D rendering is disabled. Also for some reason, disabling shadows increases framerate when skiprender is enabled.

However, with any resolution, this is a really really odd result. I wouldn't ignore testing both high and low resolutions (and skiprender), as well as both highend (x800/6800U) and low end cards.

Also remember that these results may or may not hold after Tiger.

Still I'd never turn off the second processor when actually playing the game, the framerate can sometimes fall to 50% when audio PDA logs are playing.

[v-i-c-]

Don't forget that when you want to remove the GPU from the equation and simply test CPU performance, you need to LOWER the resolution to the lowest possible (in this case, 640x480), NOT increase it. By increasing the resolution you further strain the GPU and would quickly negate any performance increase/decrease you might notice by disabling the 2nd CPU.

In other words, you want to keep this test CPU bound, not GPU bound. Lower the resolution, then test!

this makes no sense - my system runs BETTER or EQUAL at 1680x1050 in most cases. i never saw any gain in lowering resolution in most games that i play. (didn't tried another resolution in doom yet because it runs smooth most of the time at the 60 fps limit)

G5 dual 2.5, geforce 6800 ultra, 1.5GB RAM

i made both tests (2x) with

1680x1050
high settings
shadows off
AA off
anisotropic off
cache activated

dual: 57fps

single: 53fps

[Santaduck]

this makes no sense - my system runs BETTER or EQUAL at 1680x1050 in most cases

That's because you have a 6800Ultra. However that's not the point.

If you wish to contribute to this thread, which concerns CPU usage, please consider trying the r _skiprender 1 test with shadows both enabled and disabled.


____________________

QUESTION:

I still don't see reliable evidence from *anyone*. Please provide it if you are claiming a benefit. (e.g. Little Endian, Soulstorm, Jamsoft, etc.)

If you are claiming huge 20-50% benefits,
Please provide:
*Exact system specs of your Mac
*Exact game settings in Doom (resolution, quality, and all advanced settings)
*Confirm you are using results from the highest of *multiple* runs of the timedemo. (Don't ever use the first run).
*Confirm you are not using any other tweaks such as ATI Displays Utility or CFG/console tweaks, nor unpacking PAK files.
* If you claim a benefit, test other settings to see if the benefit holds for all settings: 640x480, r_skiprender "1", antialiasing maxed, etc.


If you provide the above, it would also be interesting to provide:
* Screenshot at a part of the single-player game where animation is taking place (background machines in the map), while you are playing audio from your PDA, with framerate enabled (com_showfps 1). Walk to a place where framerate doesn't fluctuate wildly. Repeat the same screenshot with 2nd CPU disabled.

[Jigglelicious]

this makes no sense - my system runs BETTER or EQUAL at 1680x1050 in most cases. i never saw any gain in lowering resolution in most games that i play. (didn't tried another resolution in doom yet because it runs smooth most of the time at the 60 fps limit)

G5 dual 2.5, geforce 6800 ultra, 1.5GB RAM

i made both tests (2x) with

1680x1050
high settings
shadows off
AA off
anisotropic off
cache activated

dual: 57fps

single: 53fps

It makes perfect sense, and has always worked this way. Generally speaking, if you don't notice an increase in performance when you lower the resolution, that means you are CPU bound - in other words, the CPU isn't fast enough to keep up with the video card, regardless of the resolution. However with the Mac, no one really knows what is screwed up, except to know that something is screwed up.

[v-i-c-]

looks like it's true (but only in some cases) :confused:

here my results:

PowerMac Dual 2.5 GHz G5
1.5 GB RAM
nVidia GeForce 6800 Ultra

all results taken from the 2nd run
AA is off, high settings, no vsynch:


at 640x480 the dual is faster:

118,2 fps • single - 640x480 shadows off, r_skiprender "1"
123,8 fps • dual - 640x480 shadows off, r_skiprender "1"
45,9 fps • single - 640x480 shadows off,
48,6 fps • dual - 640x480 shadows off,
100,8 fps • single - 640x480 r_skiprender "1"
110,5 fps • dual - 640x480 r_skiprender "1"
53,8 fps • single - 640x480
57,1 fps • dual - 640x480


at 1680x1050 the single is faster (when skiprender is not activated)

101,4 fps • single - 1680x1050 shadows off, r_skiprender "1"
108,4 fps • dual - 1680x1050 shadows off, r_skiprender "1"
53,7 fps • single - 1680x1050 shadows off,
44,7, fps • dual - 1680x1050 shadows off,
115,3 fps • single - 1680x1050 r_skiprender "1"
123,4 fps • dual - 1680x1050 r_skiprender "1"
44,4 fps • single - 1680x1050
36,0 fps • dual -1680x1050


i found that really amazing so i tested using my own "tweaked" settings again:

at my lower settings (anisotropic off, cache on, shadows off) the dual is faster again at both resolutions.

115,5 fps • single - 640x480 r_skiprender "1"
122,1 fps • dual - 640x480 r_skiprender "1"
54,2 fps • single - 640x480
55,3 fps • dual - 640x480

113,6 fps • single - 1680x1050 r_skiprender "1"
123,3 fps • dual - 1680x1050 r_skiprender "1"
53,9 fps • single - 1680x1050
57,2 fps • dual - 1680x1050


i don't think that this will help when sound is on (or how really to test that) anyway it would be interesting to find out which function is bad at higher resolution on dual machines. so here are my "tweaked" settings:


seta r_stretched "0"
seta s_device "-1"
seta r_screen "-1"
seta r_maxDisplayRefresh "0"
seta r_minDisplayRefresh "0"
seta sys_lang "english"
seta s_numberOfSpeakers "2"
seta s_doorDistanceAdd "150"
seta s_globalFraction "0.8"
seta s_subFraction "0.75"
seta s_playDefaultSound "1"
seta s_volume_dB "0"
seta s_meterTopTime "2000"
seta s_reverse "0"
seta s_spatializationDecay "2"
seta s_maxSoundsPerShader "0"
seta r_debugArrowStep "120"
seta r_debugLineWidth "1"
seta r_debugLineDepthTest "0"
seta r_cgFragmentProfile "best"
seta r_cgVertexProfile "best"
seta r_forceLoadImages "0"
seta r_shadows "0"
seta r_skipBump "0"
seta r_skipSpecular "0"
seta r_skipNewAmbient "0"
seta r_renderer "best"
seta r_brightness "1.642857"
seta r_gamma "1"
seta r_swapInterval "0"
seta r_useIndexBuffers "0"
seta r_customHeight "1050"
seta r_customWidth "1680"
seta r_fullscreen "1"
seta r_mode "-1"
seta r_multiSamples "0"
seta image_downSizeLimit "256"
seta image_ignoreHighQuality "0"
seta image_downSizeBumpLimit "512"
seta image_downSizeSpecularLimit "64"
seta image_downSizeBump "0"
seta image_downSizeSpecular "0"
seta image_useCache "1"
seta image_cacheMegs "375"
seta image_cacheMinK "2048"
seta image_usePrecompressedTextures "1"
seta image_useNormalCompression "0"
seta image_useAllFormats "1"
seta image_useCompression "1"
seta image_preload "1"
seta image_roundDown "1"
seta image_forceDownSize "0"
seta image_downSize "0"
seta image_lodbias "0"
seta image_anisotropy "0"
seta image_filter "GL_LINEAR_MIPMAP_NEAREST"
seta m_strafeSmooth "4"
seta m_smooth "1"
seta m_strafeScale "6.25"
seta m_yaw "0.022"
seta m_pitch "0.022"
seta sensitivity "5"
seta in_toggleZoom "0"
seta in_toggleCrouch "0"
seta in_toggleRun "0"
seta in_alwaysRun "0"
seta in_freeLook "1"
seta in_anglespeedkey "1.5"
seta in_pitchspeed "140"
seta in_yawspeed "140"
seta gui_configServerRate "0"
seta com_guid ""
seta com_videoRam "256"
seta com_showFPS "1"
seta com_purgeAll "0"
seta com_machineSpec "2"
seta gui_filter_idle "0"
seta gui_filter_gameType "0"
seta gui_filter_players "0"
seta gui_filter_password "0"
seta net_master4 ""
seta net_master3 ""
seta net_master2 ""
seta net_master1 ""
seta net_clientMaxRate "16000"
seta net_serverMaxClientRate "16000"
seta gui_mediumFontLimit "0.60"
seta gui_smallFontLimit "0.30"
seta g_decals "1"
seta g_projectileLights "1"
seta g_doubleVision "1"
seta g_muzzleFlash "1"
seta g_spectatorChat "0"
seta mod_validSkins "skins/characters/player/marine_mp;skins/characters/player/marine_mp_green;skins/characters/player/marine_mp_blue;skins/characters/player/marine_mp_red;skins/characters/player/marine_mp_yellow"
seta g_mapCycle "mapcycle"
seta g_voteFlags "0"
seta g_gameReviewPause "10"
seta g_countDown "10"
seta g_password ""
seta g_showBrass "1"
seta g_showProjectilePct "0"
seta g_showHud "1"
seta g_showPlayerShadow "0"
seta g_showcamerainfo "0"
seta g_healthTakeLimit "25"
seta g_healthTakeAmt "5"
seta g_healthTakeTime "5"
seta g_useDynamicProtection "1"
seta g_armorProtectionMP "0.6"
seta g_armorProtection "0.3"
seta g_damageScale "1"
seta g_nightmare "0"
seta g_bloodEffects "1"
seta r_aspectRatio "2"
seta ui_showGun "1"
seta ui_autoReload "1"
seta ui_autoSwitch "1"
seta ui_team "Blue"
seta ui_skin "skins/characters/player/marine_mp"
seta ui_name "Player"
seta si_spectators "1"
seta si_usePass "0"
seta si_warmup "0"
seta si_teamDamage "0"
seta si_timeLimit "10"
seta si_fragLimit "10"
seta si_maxPlayers "4"
seta si_map "game/mp/d3dm1"
seta si_gameType "singleplayer"
seta si_name "DOOM Server"

[v-i-c-]

uups in my last results (my own tweaked settings) i exchanged 640x480 and 1680x1050! - anyway still wondering why my mac was faster this morning :confused: (see above)

[v-i-c-]

:eek: and in the first 640x480 i exchanged the shadows off/on results - sorry 3:37 AM here :o

[Santaduck]

Very nice post Vic!

Looks like 1680x1050 is the key on your machine, but the benefit disappears when 3D rendering is disabled (using r_skiprender 1).


Now I'm curious!
what monitor are you using, (apple CD 20"?), and is 1680x1050 the native resolution of that monitor?

Theories:
1) GF6800Ultra at higher resolutions does odd things with 2nd CPU (prediction: should see same effect at 1600x1200)

2) Doom3 at Monitor's Native LCD resolution does odd thing with 2nd CPU (prediction: should not see the same effect at 1600x1200)

[v-i-c-]

yes it's the 20" cinema display aluminium and 1680x1050 is the native resolution.