the mac world tests were conducted in conjunction with pcworld. they are obviously flawed. ignore em.
for a start adobe premiere doesn't run native on OS X,so the G5 was running in emulation mode in classic and then running an old NON multithreaded app... ?!?!??! What a dumb test... meanwhile the pc 's were using the latest and greatest optimized versions. Kinda dumb....
Then look at the mp3 encoding.... the two apps they used use totally different encoders ... hahahaha... i mean on my pc i can encode with LAME at about 5 times longer than musicmatch on the same pc. So that test means jack. they should have compiled the lame encoder from source for all machines usign the optimum compiler for each platform and then tested. Personally i have done this and the G5 clobbers the Opteron and Xeon by a significant margin.
With regard to even the OS that was being tested.... .they were running on Jaguar which merely has a plugin that allows jaguar to adress the 8 gig of memory.... there is little or no other optimizations for G5. They should have used panther for that. Panther still isn't fully 64 bit. But at least it would be in the same position as Opteron on Win XP Pro which still can't take advantage of 64 bit mode in Opteron.
While people will say that the Opteron isn't optimized because its also new, that rubbish... The opteron is pretty much based on the Athlon XP core with on die memory controller, with SSE2 instructions and x86-64 instructions bolted on. Still the primary architecture is the same as the Athlon XP.... and software is very optimized for that architecture already, even the SSE2 instructions don't need any optimizing for because a lot of software takes advantage of SSE2 nowadays already. Hence Opteron is pretty much as optimized for on the 32 bit front as it will get.
On the other hand.. the G5 shares very little with the G4 bar the altivec unit and even then the altivec units are very different. People need to recompile their altivec code on the G5 for it to see its fully performance benefit. The G5 is a much more radical evolution from the G4 than the Athlon 64 is from the Athlon XP. It needs the optimization and new compilers for adequate performance evaluations to be made. While extremely competitive with competition and probably faster than competition on legacy code, it doesn't really begin to shine until code is compiled with compatible compilers ,and software starts being optimized at the programming level. For example a lot of the potential of the g5 is simply not exploited at all in any of todays apps, and certainly not in the apps macworld showed. dual fpu, dual integer etc... and its ability to have lots and lots ofintructions in flight at once... 215 compared to the 15 on the G4.
Macworld didn't even addresss this. They didn't mention the difference improvement that apps were seeing from simple recompile... for example cinebench improved 30% from a recompile with Gcc 3.3 and some optimizations. They are expecting much bigger perf gains again soon.
They also failed to mention that the GCC compiler that Apple used in their claims and bench's benefited the P4 much more than the G5. GCC has had a lot of P4 optimizations made in it, the G5 being so new and so ahead of its time has very little GCC optimizations. At least IBM has shook this up a bit by releasing a G5 compiler recently. again this is very early and performance can and will improve.
Check this for a start.
http://arstechnica.infopop.net/OpenT...963285&p=1
Neither P4 nor Opteron can claim that they have anywhere near this level of performance to be expected.
Notice how an EARLY IBM compiler generated code that was a couple hundred percent faster than GCC code for G5. Incidentally the performance improvment over GCC was also seen by the G4 as well.
"
Type of Code G4 G5
Scalar +70% +210%
Vector +40% +70%
If this holds that places his 2.0GHz G5 (single)=254 at 787
Note also that the P4 2.66 is 255."
original score of 254 went to 787.... bloody hell impressive.
Incidentally this is from the JET 3d test. Basically what they did was compile the same code for each architecture and compared. G5 clobbered all the oposition. But they were surprised when the compiled again and check the performance improvment by compiling with IBM compiler instead of GCC.What wasn't shown here is that the G5 sees a 40 to 50% improvmentin general in integer based code. IF you read down through that thread that is the general concencus.
All macworld did was make evaluations on old, out of date code on an unoptimized system with different applications on different systems. That is not a benchmark. Its a fraud.
Look at the photo shop results.... these are extremely questionable. arstechnica (a much more reliable source) did a test against all the platforms macworld used and decided that unequivically the G5 clobbered the oposition. Did macworld pick one or two filters or did they do the full 30 or so filters that are used most regularly by professionals. Thats what arstechnica did, and it made for a much more thorough test.Incidentally, the G5 plugin for photoshop only accelerates a few filters, there is huge room for perf improvement in them. Did macworld mention that the length it took to load the different filters and run them... and the overall time it takes to run a test like that... if they take the over all time the mac wins again. Did they mention the questionable nature of the dual SMP advantage in photoshop... how there is pretty much 50 /60 % over all cpu usage on G5 running phtoshop and there is nearly 80 - 90 % cpu usage on Pc.... yet the G5 still clobbers the competition on 30 filters tested by arstechnica.
As far as gaming bench's go....well that was a joke.... one cannot make or take a definitive decision about any platform based on that rubbish... different video cards with different memory configs.. different drivers...
Why didn't they test with the latest patch for the Mac in unreal or Quake 3... why didn't they use a 9800pro like they did with the pc???
The "word" test was also a joke.... why did they include it anyway??? to be honest, i would be very surprised if Microsoft had released a version of Word that ran faster on a competitors OS than on their own. I mean this test to me woudl have had the same meaning as quicktime perf on windows versus quicktime perf on Mac. Obviously the mac would be much faster on quicktime but it wouldn't really be afair test would it? Apple would be very unlikely to release quickitime on windows if it performed better than quicktime on mac??
So what can be said... i already wrote into macworld and pcworld. I am not a mac fanboy, but i hate scewed benchmarks used for publicity. Apple were guilty of that for years. I still think that their bench' on their site are extremely generous to the intel platform and that I wouldn't be surprised to see them re release results based on the IBM compiler which would see a MASSIVE boost given to SPEC int and SPEC fpu results. I think if anything they were extremely humble with their results. Even the software used wasn't properly optimized for G5 never mind the compiler.
I agree that people should write in to macworld. Better yet, stop buying their magazine. If i were apple i would be very pissed off. People will go and buy their magasine because they are thinking of buying a new mac and they will be reading rubbish and go away thinking... all apple hype... which isn't fair or accurate at all.
Top
