PDA

View Full Version : Benchmarks: dual 1.8GHz G5 from Radeon 9600 XT to ATI FireGL X3 (aka Radeon X800 XT)




Jethryn Freyman
Mar 25, 2012, 08:16 PM
OK, so I just upgraded my graphics card from a 128MB ATI Radeon 9600 XT to a 256MB ATI FireGL X3. It's a brand new PC part flashed with a Mac ATI Radeon X800 XT ROM, as both cards are identical in specs.

Anyways, I can't find any benchmarks anywhere online for Power Mac G5 with a low clock speed like mine with a powerful graphics card in it. Any time you see high end GPUs tested in G5s, it's always in something with a much faster CPU so you can see the difference in GPU power without the CPU limiting it.

Anyway, I've got 3GB of RAM, a 1TB 7,200rpm hard disk, blah blah blah, good stable system, all tests were run several times and averaged with no other applications running.

First up: the card itself is nice and compact, the fan very rarely spins up, and only under load, I'm actually yet to hear it spin up more than about 1/4 of its' top speed, and that only happens for a few seconds just after the computer boots up. Both the DVI and dual link DVI ports work.

Tests:

OpenMark (standard benchmark)
Xbench (overall, Quartz, OpenGL, User Interface)
CorePlayer (H.264 HD video, built-in benchmark)

Halo (1280x720, built in timedemo, max details, no AA/AF)
Doom 3 (1152x864, built in timedemo high quality, no AA/AF)
Quake 3 (1280x720, timedemo "demo four", max quality, no AA/AF)

Unreal Tournament 2004 (Santaduck benchmark): flybys and botmatches, maximum and minimum details settings, 1280x800 and 1680x1050, sound disabled in botmatches.

And here we go: :)

The data:

http://i.imgur.com/a0vLP.png

OpenMark:

http://i.imgur.com/q3AVr.png

All games:

http://i.imgur.com/m7WE6.png

Game tests most representative of GPU power:

http://i.imgur.com/x0f3d.png

UT2004 tests only:

http://i.imgur.com/lV3GB.png

Xbench:

http://i.imgur.com/AWKon.png

CorePlayer (100+ means the video could be played without dropping frames):

http://i.imgur.com/Fi5Uf.png

----------

A few notes:

-All tests run with Mac OS X 10.5.8 with all updates and firmware updates installed. June 2004 base model G5.

-UT2004 botmatches are very much CPU limited.

-Doom 3 is also very CPU limited. Both it and UT2004 would totally max out one CPU, while not using the other at all.

-What's going on with the CorePlayer scores? I don't know. The results were consistent, though...

-I discarded any outlying results. This only happened a few times, for example, my first Doom 3 test scores about 30fps, the next three were all around 43fps. I kind of expected that though, I assume it's from the game loading the game data into memory the first time it's run.

-General feel of the Mac OS X UI: no change in responsiveness.

-Quake 3 scales nicely :)

Hope this is useful to somebody :apple:



monkeybagel
Mar 27, 2012, 03:33 AM
Good, interesting work.

Jethryn Freyman
Apr 30, 2012, 09:43 AM
OK, I just Googled "ATI FireGL X3 Mac OS X" and my damn thread is the first result. If I'm the only one providing this info, I think this thread deserves a bump, it sure would've helped me out three months ago. Any questions? Ask away. I'm pretty sure that I have the fastest 1.8GHz G5 going around town now.

vohdoun
Apr 30, 2012, 10:00 AM
Doom3 will probably be getting let down by ATI's modern OpenGL drivers compared to what Quake3 requires.

Even the latest AMD cards today on the PC for OpenGL are poor.

thorns
Apr 30, 2012, 11:09 AM
The fps scaling in Quake 3 is interesting. I can't recall that I reach 300+ fps in 1280x720, I reach 240fps in 1600x1200 max, however (ioquake3). I did some benchmark testing in Tiger and Leopard with a 7800GT and a X1900GT, not much of a difference (no clear winner, either).

Jethryn Freyman
Apr 30, 2012, 09:53 PM
http://i.imgur.com/UFzpE.png
Alrighty, OpenMark run with a 25% overclock on memory and core (do I win (:)?):
http://i.imgur.com/TLwHR.png
And the UT2004 flyby test on max:
http://i.imgur.com/6aPJh.png
Unfortunately, the GPU shut off a few seconds into the UT2004 botmatch benchmark, so I rebooted and decided to call it quits.