Overclocked Radeon 9600 Pro: UT2004 benchmarks

[oingoboingo]

As mentioned in a thread I posted in the Hardware section (http://forums.macrumors.com/showthre...threadid=86732), a utility called ATIcceleratorII has been released which allows simple OS X-based overclocking of ATI graphics cards, without the need for BIOS flashes or even rebooting. The software is available here:

http://thomas.perrier.name/software/...leratorII.html

I installed the software on my 1.6GHz PowerMac G5, with the standard Apple OEM 64MB Radeon 9600 Pro AGP 8x card, 1.25GB of RAM, and OS X 10.3.5. The software installed without a hitch, and adds an extra item to the System Preferences which can be used to adjust both the GPU speed and the RAM speed of the graphics card independently.

To test if the changes I was making to the Radeon 9600 were having any effect, I benchmarked the system using the Santaduck Toolpak Unreal Tournament 2004 benchmark suite. For these tests, Flyby mode on the DM-Antalus map was used, at a resolution of 1600x1200, in 32-bit colour. Details settings were at 'Maximum' (which corresponds to 'High' settings for all the UT2004 detail levels, all extra graphical effects turned on, and 'Low Sound Detail' selected). The full retail version of UT2004 was used, with the latest patches applied. All tests were run 3 times, and the average FPS scores for each run averaged out to give an overall score. No other software was running at the time of the benchmarks (except Finder).

-364.5 MHz core, 294.75 MHz VRAM (stock OEM Radeon 9600 Pro settings):38.8 FPS

- 382.5 MHz core, 294.75 MHz VRAM (4.94% core overclock):39.6 FPS

- 400.5 MHz core, 294.75 MHz VRAM (9.88% core overclock):40.2 FPS

- 400.5 MHz core, 324.00 MHz VRAM (9.88% core overclock, 9.92% VRAM overclock):43.05 FPS

- 418.5 MHz core, 337.50 MHz VRAM (14.81% core overclock, 14.50 % VRAM overclock):44.9 FPS

(*note: The FPS scores for each test are very tightly grouped, varying only by 0.1 or 0.2 FPS each run. I could add standard deviations, but at this stage, I'm too lazy )

I stopped overclocking at 15% for both GPU core and VRAM, since I wasn't sure how much Radeon 9600 Pros can be safely overclocked in general (I'll do some research), nor how well the cooling system in the G5 will cope with any increase in heat from an overclocked graphics card. However, even at 15%/15% overclocking, there was no sign of any visual distortion or artefacts during the 3 runs of the DM-Antalus Flyby benchmark (nor was there in any of the lesser overclocked runs). A 14.81% increase in GPU core coupled with a 14.5% increase in VRAM clock speed resulted in a 15.7% increase in average FPS scores, so in this part of the curve at least it looks like near to linear scaling.

When I'm feeling a little more adventurous I might try for some higher overclocks, and I'll definitely do some stress testing of the overclocked configuration (UT2004 spectator mode for a few hours at least) to see how stable this configuration is.

EDIT: PS, you can get the Santaduck Toolpak benchmarking kit here:
http://www.macgamefiles.com/detail.php?item=18189

[Mord]

it's a freaking g5, i think it cn handle an overclocked 9600, did you put some ramsinks on it?, or maybe batter cooling than the passive heatsink?

edit, i'm going to totally going to waste an ati rage on my cube with some mad voltage heatsinked 120mm fan fun

[Hoef]

Yeah would be neat to see the temperature read outs as well

[oingoboingo]

Originally Posted by Hector

it's a freaking g5, i think it cn handle an overclocked 9600, did you put some ramsinks on it?, or maybe batter cooling than the passive heatsink?

edit, i'm going to totally going to waste an ati rage on my cube with some mad voltage heatsinked 120mm fan fun

Yeah, the whole thing is perfectly stable at 15%/15%, so I'll definitely be trying to overclock it some more. I haven't added RAM sinks or anything like that yet...it's just a completely stock G5. I figured the cooling fan that takes care of the PCI bay thermal zone would wind up a bit if things got toasty in there, but who really knows.

From doing a little searching around the web, it seems that the Radeon 9600 'Pro' that Apple supplies in the G5 might not be all that 'Pro' after all. It seems that the standard GPU clock speed in PC-issue Radeon 9600 cards is 400MHz (or very close to that), not the 364.5 MHz that the Mac Radeon 9600 Pro is set to. At least the Mac VRAM speeds (294.75MHz) are close to what most PC cards are set at (around 300MHz).

I will need to pop the lid off the G5 and examine what brand and specification of RAM is being used in the OEM Radeon 9600 Pro...people are claiming VRAM overclocks of up to 350MHz and beyond with Sapphire cards. HardOCP managed to achieve a stable core GPU speed of 567MHz and a VRAM speed of 358MHz (716MHz effective speed, since it's DDR RAM). I don't expect to be able to achieve anything like that, since the OEM Radeon 9600 in the G5 does not have its own cooling fan (just a heatsink which is cooled by the shared PCI bay fan) and has no RAM sinks, but those HardOCP results are encouraging to say the least...

Now...somehow I've managed to be relegated to the PowerBook while my soon-to-be wife is monopolising the G5. How the hell did that happen!?!

[oingoboingo]

OK, a beer or two later and I have acquired enough courage to try some more adventurous overclocking. All testing conditions are identical as the previous tests.

438.75 MHz core, 351 MHz VRAM (+20.37% core overclock, +19.08% VRAM overclock): 47.0 FPS

456.75 MHz core, 351 MHz VRAM (+25.31% core overclock, +19.08% VRAM overclock): 47.7 FPS

474.75 MHz core, 351 MHz VRAM (+30.25% core overclock, +19.08% VRAM overclock): 48.4 FPS

474.75 MHz core, 369 MHz VRAM (+30.25% core overclock, +25.19% VRAM overclock): 50.1 FPS

At the last settings (474.75/369) I started to notice visual artefacts on screen. These persisted even after exiting UT2004 (I was seeing coloured blocks of pixels appearing in Finder). I rolled the Radeon 9600 back to 474.75/351, and the artefacts disappeared. I have been testing the system now for about the last 20 minutes running a 'bot DeathMatch in Spectactor mode, and everything seems stable and free of visual artefacts.

So, from my limited testing so far, it looks like a 474.75 MHz core clock speed (up 110.25 MHz from a stock speed of 364.5 MHz) and a VRAM speed of 351 MHz (up 56.25 MHz from 294.75 MHz) results in a stable graphics card, and an approximate 24.7% improvement in UT2004 framerates, as measured in the DM-Antalus Flyby benchmark.

[supergod]

Right now I'm running my Radeon 9600 (1.6ghz G5, 1gb RAM, os 10.3.5) with processor bumped up 18.52% to 432 mhz and memory up 18.32% to 348.75mhz. Everything seems to be going well so I'm going to push it farther later today. Does anyone know if this (minor) overclocking can lower the shelf life of a system? I'm not afraid of the card getting fried or anything, but I wonder if the extra strain on the fans or whatnot has a noticeable effect.

[vraxtus]

I wonder how well this will work with my Rad9800 SE...


And to the above poster... "shelf life" is the life of a unit that's still in a store inventory, without use. I think you meant just "life."

[ChrisFromCanada]

I thought I would test it on my 9800 Pro but right now I am too lazy to do anything very time consuming so I just did the graphics tests in xbench and this is what I found:

My computer before any overclocking:
Quartz Graphics Test 154.30
Line 130.25 3.32 Klines/sec [50% alpha]
Rectangle 149.16 10.49 Krects/sec [50% alpha]
Circle 152.99 3.53 Kcircles/sec [50% alpha]
Bezier 144.30 1.57 Kbeziers/sec [50% alpha]
Text 219.48 3.58 Kchars/sec

OpenGL Graphics Test 88.67
Spinning Squares 88.67 62.05 frames/sec

And after both VRAM and processor were overclocked 10% each:
Quartz Graphics Test 164.35
Line 130.63 3.33 Klines/sec [50% alpha]
Rectangle 153.74 10.82 Krects/sec [50% alpha]
Circle 155.35 3.58 Kcircles/sec [50% alpha]
Bezier 145.54 1.58 Kbeziers/sec [50% alpha]
Text 338.31 5.51 Kchars/sec

OpenGL Graphics Test 87.03
Spinning Squares 87.03 60.90 frames/sec

So, for some reason I expirienced a slight performance loss in OpenGL but a gain in Quartz. I also realize that my bottleneck of my computer is my processor and that the video card is much more powerful than it but I thought I would give it a shot. Maybe I will do the Unreal benchmark later.

[oingoboingo]

Originally Posted by vraxtus

I wonder how well this will work with my Rad9800 SE...


And to the above poster... "shelf life" is the life of a unit that's still in a store inventory, without use. I think you meant just "life."

...unless he has it installed on a shelf, and is worried about the shelf overheating from the overclocked compuer sitting on it and...uhhh..nevermind.

BTW, have just complete a full hour of UT2004 bot DeathMatch testing with my Radeon 9600 at 474.75 / 351, and it has been totally stable and free of artefacts. I think we have a winner...

[oingoboingo]

Originally Posted by ChrisFromCanada

So, for some reason I expirienced a slight performance loss in OpenGL but a gain in Quartz. I also realize that my bottleneck of my computer is my processor and that the video card is much more powerful than it but I thought I would give it a shot. Maybe I will do the Unreal benchmark later.

XBench is a dubious benchmark. The scores don't often seem to be reproducible, and there should be no reason why overclocking your card would result in a loss of performance. Try with a 'real world' application (well, as 'real world' as Unreal Tournament is) and see if you can measure an improvement.

[vraxtus]

I think I'll try this on Halo later tonite when I get home... also oingo, what is the "dual 2.5 upgrade kit" described on your sig...?

edit: nvm I'm an idiot.

[neonart]

This is very interesting! I will try to run Halo timedemos later to see if this mod helps this supper crappy game. All my other games run amazing on my setup, while Halo sucks miserably. Hope this helps!

[invaLPsion]

I'm not surprised that people are having such good luck with OC'ing their video cards, seeing as how Apple underclocks their OEM cards in the first place...

For example, my 9600XT is supposed to run at 500MHz CPU, 600MHz memory. It only runs at 400.5MHz CPU and for some strange reason, 621MHz memory.

I'll try turning my 9600XT into an actual 9600XT later... I will inform everyone of the results...

[supergod]

Thanks for taking the time to make fun of my using the word shelf life on a forum. Just so you know, the definition of shelf life is not limited to use in a store (or on a shelf). It simply means the length of time in which a product remains suitable for use... or at least that's what the dictionary says. Now that that's out of the way, while a computer with a overclocked video card may not experience artifacts or overheat, is there any other kind of long term damage? I'm not particularly worried, just thought I should know.

[vraxtus]

Originally Posted by supergod

Thanks for taking the time to make fun of my using the word shelf life on a forum. Just so you know, the definition of shelf life is not limited to use in a store (or on a shelf). It simply means the length of time in which a product remains suitable for use... or at least that's what the dictionary says. Now that that's out of the way, while a computer with a overclocked video card may not experience artifacts or overheat, is there any other kind of long term damage? I'm not particularly worried, just thought I should know.

shelf life
n.
The length of time a product may be stored without becoming unsuitable for use or consumption.

tr.v. stored, stor·ing, stores
To reserve or put away for future use.

Sooooooooo... not really!

Anyhow, the long term effects of overclocking remain to be seen, at least for most Mac users, who typically (with the exception of the eMac) don't really OC processors... especially GPUs, so time will tell.

And lighten up dude, I'm just trying to show you that you didn't really use the term correctly, that's all.

[invaLPsion]

Originally Posted by oingoboingo

So, from my limited testing so far, it looks like a 474.75 MHz core clock speed (up 110.25 MHz from a stock speed of 364.5 MHz) and a VRAM speed of 351 MHz (up 56.25 MHz from 294.75 MHz) results in a stable graphics card, and an approximate 24.7% improvement in UT2004 framerates, as measured in the DM-Antalus Flyby benchmark.

Where can I find this DM-Antalus Flyby benchmark?

[Demon Hunter]

Originally Posted by invaLPsion

I'm not surprised that people are having such good luck with OC'ing their video cards, seeing as how Apple underclocks their OEM cards in the first place...

Does this include Mobilty chipsets in the Powerbooks? That would make more sense than underclocking a desktop card.

Has anyone tried this with one of the new Powerbooks? Or is that GPU suicide?

[supergod]

Don't worry I take no offense. I'm just trying to show you that online dictionaries have one definition whereas the complete 20 set Oxford English dictionary has many many definitions for even the most common words. Well, at least I can be confident that overclocking my machine won't melt it (poor overclocked imacs!). Danke.

[broken_keyboard]

Originally Posted by invaLPsion

I'm not surprised that people are having such good luck with OC'ing their video cards, seeing as how Apple underclocks their OEM cards in the first place...

For example, my 9600XT is supposed to run at 500MHz CPU, 600MHz memory. It only runs at 400.5MHz CPU and for some strange reason, 621MHz memory.

I'll try turning my 9600XT into an actual 9600XT later... I will inform everyone of the results...

It's the same with the 9800 Pro. The reference frequencies for this card are Proc: 378MHz, Mem: 337.5MHz but I just ran this program on my Apple OEM version and it shows Proc: 351MHz, Mem: 324MHz

[vraxtus]

Originally Posted by broken_keyboard

It's the same with the 9800 Pro. The reference frequencies for this card are Proc: 378MHz, Mem: 337.5MHz but I just ran this program on my Apple OEM version and it shows Proc: 351MHz, Mem: 324MHz

Can you get me a link to these specs? I need to spec my 9800 Pro SE before I go and OC it. Thanks!

[Converted2Truth]

I get artifacts @ 9.88%/9.92% Oh well.

[invaLPsion]

Originally Posted by dferrara

Does this include Mobilty chipsets in the Powerbooks? That would make more sense than underclocking a desktop card.

Has anyone tried this with one of the new Powerbooks? Or is that GPU suicide?

Speeds on the Mobility Radeon 9700 CPU can safely range from 390MHz to 450MHz. Apple has the ones in the powerbooks running at 392MHz...

It may not be GPU suicide, but you could risk overheating your powerbook...

[dabirdwell]

In a Dual 867 MDD? It's older, but still ATI.?

[broken_keyboard]

Originally Posted by vraxtus

Can you get me a link to these specs? I need to spec my 9800 Pro SE before I go and OC it. Thanks!

These are the official figures, but they round them slightly.
http://www.atitech.com/products/rade...o/compare.html

For the exact values check out sites like ocfaq.com

[vraxtus]

Originally Posted by dabirdwell

In a Dual 867 MDD? It's older, but still ATI.?

Again, if you read the release notes page, it says all ATI cards are supported, even Rage128 cards...

bk - thx