View Single Post
Old 02-20-04, 03:34 PM   #32
Nitz Walsh
Registered User
Join Date: Jul 2002
Posts: 622

As someone who's job is primarily administering to the needs of 100+ Mac users, most with G4's and G5, and having a Dual 1.4 G4 and now 1.8 G5 in my office, I can state without reservation that it is very likely the PC version of Doom3 will be the fastest.

In every cross-platform gaming comparison I've benchmarked myself, the Mac usually gets its ass handed to it in games. Yes, a $3000 Dual 2GHZ G5 is looking better these days, but every game benchmark I've seen still has a 3/3.2 P4 with the equivalent graphics card handing the Mac its ass, while costing just over a third as much with equivalent hardware. From a gaming perspective, even the 1.6G5 has a truly awful price/performance ratio compared to its PC brethren.

Those xlr8 UT 2004 benchmarks are extremely suspect. For one, flyby is completely fill-rate limited, so we're testing graphics cards here, not the overall system. Botmatch is a far better indicator of "true" UT 2003 performance, any in my benchmarks some time ago when I had a Dual 1.4, in some maps at 640*480 (in order to eliminate the graphics card, the Mac had a 9000Pro, I had a 4200), it was running at half the framerate of my home Athlon XP1800 system (other maps were closer, none were faster).

Edit: Accurate UT2003 PC flyby benchmarks here. One has to wonder how the heck TechTV got their numbers, or rather from what bodily orifice.

Note the above benchmarks are done on a 3.0ghz P4 - not 3.4, or an A64 or FX which provides even better performance (although it shouldn't affect flyby that much).

Things have improved since the demo, and there is a patch that provides additional performance for the G5 which I haven't tested as I don't have the full Mac retail version, but I have seen other benchmarks and they certainly don't correlate with xl8's findings. Bear in mind one of the apps that was helped most with Catalyst drive revisions was UT2003- for a few releases there, it was improving in the double digits (percentage wise) with each release, so the driver version is of paramount importance. It's the first benchmark I've seen in gaming where the Mac is actually faster (but as I said, flyby is largely useless to judge UT2003 performance) - and certainly in the G4 era, where in gaming benchmarks it was absolutely humiliated (a 3ghz P4 benchmarked at 2.5-3X the average framerate of a Dual 1.25 G4 in Jedi Knight:Outcast, for example).

I recently tried out the UT2004 Mac demo. On the Assault map at 512*384 (Assault 2004 - at least according to the demo - is massively CPU limited, plus this 1.8G5 has only a 5200 so I wanted to eliminate the graphics card bottleneck), it was definitely slower than the same map on my Xp2400/GF4 4200 machine at home, and that was running at 1024*768.

The G5 is a new chip, and compilers and skills needed to eke the most out of it are constantly evolving, so things can change. However, bear in mind a few things:

1) These games are designed with X86 code and Direct3D in mind from the outset (Doom is obviously OpenGL).
2) The drivers on the PC are far more optimized for game situations simply due to market economics.
3) Most game code does not scale that well with dual processors, heck most code in general doesn't. You can't simply state "Look at that! 2 64 bit 2 ghz processors! That will surely defeat a single 3.4ghz P4!"...uh, no.

All of those caveats aside that help explain poor Mac performance in gaming, the fact of the matter is, even with professional apps that can take good advantage of dual CPU's and Altivec, a 2GHZ G5 is roughly "on par" with a 3.2 ghz P4 system. Some benchmarks it will be faster, others slower. The G5 will continue to improve as compilers mature no doubt, but of course Prescott will be helped by maturing compilers and SSE3 support as well.

Apple equips its high-end PowerMacs with dual CPU's for a reason: it has to. Talk about which is more "efficient" is irrelevant, it's all about bang for the buck – who cares about Mhz (if I did, I wouldn’t own an Athlon). If Apple's highest end G5 at launch was a single 2.0ghz, it would look like a joke if benchmarked against 3.2ghz P4 systems. Of course, considering such a system these days retails for less than a single 1.6G5, that harsh criticism may still apply if you're not hooked on OSX (which I can understand, it has its faults but I'd pick it over XP if given the choice anytime).

Macworld's G5 vs. Athlon/P4 benchmarks

This isn't the definitive Mac vs. PC benchmark study, some of the apps are indeed poorly optimized for the Mac (Premiere has been dropped for the Mac, FinalCut Pro performance is significantly better), but since it's from MacWorld it's at least some guard against accusations of bias (which there shouldn't be - if Apple goes, I'm probably out of a job - or at least one I enjoy!). As mentioned, it also gives some indication on how much the crippled memory subsystem of the G4 machines hurt it – the single 1.8 G5 is faster in Quake3 than the Dual 1.42 model, for example.

Yes, the Dual G5 is the "Worlds faster computer" - according to Apple. According to independent third party benchmarks - no. For certain tasks it can be, and for certain tasks a PC will be - such as games.

As for the Xbox version – please. The Doom3 textures alone take up more than 64 megs in their highest setting from what I’ve read. Sure, it will probably run OK on the Xbox (640*480*30fps), but with significantly reduced texture quality, aliasing, shorter levels, joypad instead of mouse/keyboard (ack!) and who knows what else turned down. To even suggest it could run at 1280*720p is absolutely inane, there’s a reason less than a handful of games run beyond 640*480 on the Xbox – horsepower. Programmer ability has zero to do with it, you’re advocating that a game based on a graphics engine which pushes fill-rate like there’s no tomorrow should run at a res far less graphically demanding games can’t muster? Er….ok.

Last edited by Nitz Walsh; 02-20-04 at 03:53 PM.
Nitz Walsh is offline   Reply With Quote