"I got my Mac Pro this week, replacing a Quad G5. Of course I ran some game benchmarks (I submitted benchmarks using the Quad G5 in May 2006). Mac Pro, 2.66 GHz, 3 GB RAM, Radeon X1900 XT 512 MB, Mac OS X 10.4.8

Quake 4 v1.3, default settings (fresh config), id_demo001.netdemo. The very first timedemo result was discarded (as usual).

First, I set the game to "high quality" (optimal for 256 MB video memory), and manually set the resolution:

640x480, high quality, noAA, SMP OFF, with Shadows: 56.1 fps

640x480, high quality, noAA, SMP ON, with Shadows: 88.1 fps (+57% vs SMP OFF)

640x480, high quality, noAA, SMP ON, no Shadows: 113.5 fps (+29% vs w/Shadows)

1920x1200, high quality, noAA, SMP OFF, with Shadows: 49.5 fps

1920x1200, high quality, noAA, SMP ON, with Shadows: 51.9 fps (+ 5% vs SMP OFF)

1920x1200, high quality, noAA, SMP ON, no Shadows: 96.0 fps (+85% vs w/Shadows)

1920x1200, high quality, 2xAA, SMP ON, no Shadows: 90.5 fps (- 6% vs noAA)

1920x1200, high quality, 4xAA, SMP ON, no Shadows: 74.7 fps (-22% vs noAA)

Indeed, AA seems to work now - it didn't with Q4 v1.2 on Quad G5 with GF 7800 GT. I don't know if the v1.3 of Q4 fixed this, or if AA works only on ATI cards. 8xAA and 16xAA still don't work properly, as their timedemo results are higher than the 4xAA one (both ~78 fps).
(last week's news also had Quake 4 FPS tests comparing the 7300GT card to the X1900XT (stock and OC'd))

After this I set the game to "ultra quality", which I thought should work fine with 512 MB video memory.

640x480, ultra quality, noAA, SMP OFF, with Shadows: 51.4 fps (- 8% vs high quality)

640x480, ultra quality, noAA, SMP ON, with Shadows: 75.5 fps (-14% vs high quality)

640x480, ultra quality, noAA, SMP ON, no Shadows: 91.8 fps (-19% vs high quality)

1920x1200, ultra quality, noAA, SMP OFF, with Shadows: 42.3 fps (-15% vs high quality)

1920x1200, ultra quality, noAA, SMP ON, with Shadows: 42.9 fps (-17% vs high quality)

1920x1200, ultra quality, noAA, SMP ON, no Shadows: 68.7 fps (-28% vs high quality)

As you can see, the performance difference compared to "high quality" is considerable. I noticed that Quake 4 detected only 256 MB of video memory ("com_videoram 256"), so I guess it doesn't make use of the full 512 MB, which is probably why the results are so much lower than the high quality ones. So this is a bug that needs fixing I imagine. (Only a single display was connected to the video card BTW)

For fun I compared the above to "medium quality", which is optimal for 128 MB of video memory IIRC. You'll note that the difference to high quality isn't as much.

1920x1200, medium quality, noAA, SMP ON, with Shadows: 53.4 fps (+ 3% vs high quality)

1920x1200, medium quality, noAA, SMP ON, no Shadows: 101.1 fps (+ 5% vs high quality)

So I'm probably going to play at 1920x1200, high quality, 2xAA, no Shadows (in Multiplayer) or with Shadows (in Singleplayer) for now.

BTW, I also put an Accelero X2 cooler on my Radeon X1900 XT to keep its noise down. I could hear the original cooler even with headphones on while gaming, the Accelero on the other hand is whisper quiet. On a sidenote, the Raptor 150 I was using in my Quad G5 previously is *much* more audible now that it's in the Mac Pro (I'm talking about seek noise, there's nearly no idle noise). Of course the Mac Pro is quieter than the G5 so the hard disks are easier to hear. More importantly, I think the way the hard disks are mounted in the Mac Pro (directly on metal sleds, vs plastic rails in the G5) amplifies their seek noise considerably.

Anyway, I'm quite happy with the Mac Pro so far. WinXp/BootCamp also runs fine (yay for Half Life 2). I'll submit some other benchmarks later... I might also try to overclock the Radeon X1900 XT.

Unreal Tournament 2004 v3369.2, Asbestos FlyBy & BotMatch using Santa Standard Bench.app (stock Mac X1900 XT clock speeds - see below for results with OC'd X1900 XT)

Asbestos FlyBy (Minimum FPS / Average FPS / Maximum FPS, score):

1920x1200, minimum settings: 109.3 / 530.3 / 1748.3, score = 80.04

1920x1200, maximum settings: 89.1 / 359.9 / 1075.7, score = 80.04

Asbestos Botmatch (Minimum FPS / Average FPS / Maximum FPS, score):

1920x1200, minimum settings: 56.4 / 158.3 / 347.0, score = 79.94

1920x1200, maximum settings: 19.5 / 99.1 / 309.7, score = 77.63

(He later wrote)

I reflashed my (Apple OEM) ATI Radeon X1900 XT from 600 MHz core & 650 MHz memory clock to 675 MHz core & 800 MHz memory clock (AFAIK regular WinPC X1900 XTs have 625 & 725, while XTXs have 650 & 775).

Prior to setting these rather high clockrates I ran the ATI catalyst tool in Windows to gradually overclock it to these settings, while running 3DMark06 in a loop to make sure these settings work properly (with 700 MHz core clockrate it did hang some times).

Then I extracted the ROM from my card while booted from DOS, set the new clocks using Graphiccelerator (ROM editor, see Graphiccelerator home page) in Mac OS X, and reflashed the card while booted from DOS again. (The exact how-to can be found in the strangedogs.com forum, where they say how to flash a WinPC X1900 XT to a Mac version).

Here are some Q4 results at high quality (w/675MHz core/800MHz memory) - same settings as my earlier report:

1920x1200, high quality, noAA, SMP ON, with Shadows: 59.7 fps (+15% vs non-overclocked)

1920x1200, high quality, noAA, SMP ON, no Shadows: 104.3 fps (+ 9% vs non-overclocked)

1920x1200, high quality, 2xAA, SMP ON, no Shadows: 99.4 fps (+10% vs non-overclocked)

As I wrote in my first mail today, the Mac OS X version of Quake 4 only recognizes 256 MB of the 512 MB on my X1900 XT video card. The same is true for the Mac OS X version of Doom 3 apparently.
In Windows however, Doom 3 did recognize the full 512 MB, so I guess this is a bug (?) specific to the Mac port of either game.

More UT2004 results with the overclocked Radeon X1900 XT. X1900 XT running at 675 MHz core clock and 800 MHz memory
UT 2004 v3369.2, Asbestos FlyBy & BotMatch using Santa Standard Bench.app

Asbestos FlyBy (Minimum FPS / Average FPS / Maximum FPS, score):
- 1920x1200, maximum settings: 92.2 / 394.6 / 1481.2, score = 80.04
(+10% average FPS vs. non-overclocked)

Asbestos BotMatch (Minimum FPS / Average FPS / Maximum FPS, score):
- 1920x1200, maximum settings: 28.9 / 100.4 / 314.6, score = 77.91
Cheers, Oliver"