PDA

View Full Version : Much headroom for Dual 2 Ghz G5??


i_wolf
Oct 22, 2003, 12:56 PM
Hi guys,
I know that the G5 is very early in its life span and it has agreat future ahead of it.
I was not swayed one way at all by the pcworld bench with the G5 in it. I think there were too many issues at stake besides ram.
1) They were testing on Jaguar... which to thebest of my knowledge only has a patch to allow compatibility no taking advantage or optimizations at all.
2) They were using software compiled for older PPC G4, again not compiled for G5 at all except photoshop (and that killed the Opteron and the Xeon by a considerable margin).
3) That one bench showing a single Athlon FX beating the dual was a jok.....Premier was being run though effectively emulation mode .... OS9. Hardly even steven. And also not dual aware.
4) Very few apps if any were dual aware.... this is deabately i mean dual aware in the strictest sense that they didn't leverage at all the power of a duallie be it pc or mac duallie.


Which brings me to the main point of my post.
1) Do you think that a bigger gap between the G5 and PC's will emerge when Panther is released. Or do you reckon that it will take at least another OS upgrade before we have full 64 bit optimizations???
2) Do you think that many apps have potential for speed boost or do you reckon that a recompile will only give a minor speed boost??
3) Is now a good time to be looking at a G5 or is it likely that apple will go dual 2.5's?? Reason i ask is im told that stocks are extremely low and there is a big wait and usually that is the sign tthat a speed bump is immenent. Also i was thinking that because its Christmas Market and all, Intel will be releasing P5 or Prescott and AMD will be speed bumping the Athlon FX one more time, so apple might have to speed bump their G5 again???
Any opinions???

Kind Regards,
i_wolf

jonapete2001
Oct 23, 2003, 04:50 PM
The single g5 being beat by a single Athlon fx was not a joke. Remember the g5 was dual 2ghz. The fx chip was 2.2 ghz. For apps that are not dual aware the fx will beat out the g5. The G5 is a great chip but I also think the athlon fx broke new ground on the pc side. I also think the panther release should improve the g5. But it will take another 1 or 2 os releases before they are fully 64 bit. This can also be said for the x86 64 amd chips. they should also improve when the 64 bit windows release comes later next year. Over all the i think both chips are even. From what i see and hear the difference is negligible. Both the G5 and athlon fx are both good chips.

i_wolf
Oct 23, 2003, 05:59 PM
but isn't it a case of fact not necessarily that G5 on Jaguar is a stop gap to keep people going until panther. I mean panther will be be somewhat optimized for G5 like windows would be for SSE2 on P4 and Athlon 64 (i.e. panther still isn't 64 bit true OS) however it still isn't really going to truly shine IMO until one or two dot releases of panther where apple add more G5 optimizations. Neither the G5 nor Amd 64 benefit from 64 bit os. This will be very interesting actually because currently the Athlon does gains about 15% improvement in linux by moving to 64 bit. Hope and i would imagine that the G5 would do far better.
I look at it like this. THe native support in G5 and Opteron is different.
Essentially AMD are using a tweaked athlon core with 64 bit extensions bolted on. Ignoring the memory controller integrated.... it still means that most of the internal pathways etc... are similar to the Athlon. So running 32 bit code there is not much room for optimization left.... the code they ran was already optimized 32bit code that makes extensive use of SSE2.... the Athlon also benefits form this. I would imagine that there is little room left for the AThlon 64 in 32 bit apps for optimizations and compiler reworks.
On the other hand.. the G5 is nothing like the G4 architecturally....it was built from the ground up. It is also much more massively parallel ... so that you have many more instructions in flight 215 or something like it. So if you have many apps open at once, like in a real world the G5 should scream past the Intel and Opteron. This is the case with 32 bit and 64 bit code. However 32 bit code can be optimzed for the G5 Altivec unit which is enhanced over the G4... not new instructions but tweaked to be more efficient.. (was reading the ADC notes on the G5). Now current 32 bit apps like they were using ... were not optimized at all for the 32 mode of the G5. On the other hand the opteron made instant advantage from using otimized SSE2 standard x86 code.
The point is that as existing 32 bit apps are recompiled they tend to yield a 30% improvment... this is not even adding 64 bit support. So i reckon the in future apps... the G5 will really rocket ahead of the Opteron and Intel...
Even think about the fact that GCC has only started to begin to be optimized.... wow. So what number it did to hold its own and beat the opteron in the one or two bench's that were "fair".. .as in the G5 was running in classic mode but was still using unoptimized code on an unoptimized os... well that says alot. TIt must be raw sheer brute force power that is keeping it up at the moment. The great thing about the G5 is that it has the raw brute horse power and elegant effiency to boot..... now all developers need to do is start optimizing.
Incidentally in Mac Magazine (irish print) they reran some of the tests on panther and the G5 had a healthy lead even over the dual polywell numbers and the single fx 2.2. Its early days yet for the G5.
Incidentally i don't want to come off as negating the Opteron. I think its a phenominal chip, i have a dual 1.8 in college in the netsoc. They are awesome processors. Easily the best of the x86 bunch... its just that , the review that we saw was very biased and had a lot of testing loop holes.
Interestingly when the P4 was first release granted a really really crap chip and was noticibly slower than athlon at the time... once code was recompiled and optimized it left the athlon in the dust in a lot of applications like divx encoding etc.....
This is the same with the G5 except its a great design AND has the future potential that the P4 had back then.

agdickinson
Oct 24, 2003, 12:21 PM
HI,

Do you have further details of the article?

Andrew
:)

couch potato
Oct 25, 2003, 11:29 AM
HAHAHA andrew:) :D

it seems like i_wolf practically did his own article.

yeah, andrew, i got a link somehwhere....let me get back to you on that one ill find it.:(

actually, i got panther on our g5 last night. it feels like there is an improvement, or like the g5 is more complete:rolleyes: .

as for speed, if there is an increase, its a kind of hard to tell since this thing screams in 10.2

i think ill have to use it a little more to really tell.

thanks i wolf. after posting that pcworld benchmark, this thread is very interesting:)

i_wolf
Oct 25, 2003, 04:41 PM
its just that there were so many holes. I am not a MAC fanboy at all. Indeed i own an Opteron. So i can compare like versus like.
I actually went out and bought the macworld magazine today. Its filled with even more garbage tests....
For example....
they ran an MP3 encode test... comparing the encode speed of itunes versus music match.
Now there are a few HUGE problems with this... besides the drive speeds which make a huge difference what they should have done was read in as WAV files and tested the convert to mp3 using a standard mp3 encoder that is available on all platforms.
The mp3 encoder that comes with music match is terrible. Sure i can encode mp3's extremely fast but they sound terrible. As far as im aware .. i could be wrong because i dont use music match anymore they use the Xing encoder, which is extremely fast but creates terrible quality encodes and which does no analysis before it encodes. The itunes encoder is remarkably better than the music match however it still isn't great. it does a better job. HOWEVER.... i ran an encode test on Opteron dual using iTunes for windows and compared against a dual G5. the G5 was double the speed. Is this a fair test. No. without a doubt the itunes version for windows is not as fast hasn't been optimized. However the itunes for mac looks like it has absolutely no G5 optimizations... and the G5 is simply running legacy G4 code in 32 bit mode. At the very least apple should improve this by recompiling itunes for G5 this in of itself should give it 30% improvement. If they want to double or tripple the performance they should make calls to the specific hardware on the G5.
Further more when i encode mp3 on my pc. I don't use itunes or musicmatch. I use the most popular and BEST encoder around . LAME. hydrogen audio and headfi and other audiophile places all agree this produces the best mp3's. Its also open source so one can compile your own version of lame for your specific processor. Even when using LAME Pentium 4 optimized version of LAME (it is compiled with SSE2 extensions and there is SSE2 calls in the code) Lame still encodes at probably 1/3 of the speed of the Xing encoder used in Music Match. But the quality difference is huge. THere is much more audio processing done on what should and shouldn't be thrown away. This results in higher quality output.
Now for a fair MP3 encoding test. They should have gotten the unoptimized code / basic LAME source code.
They then should have used a general compiler for all three processors (P4, Opteron, PPC970)... such as GCC 3.3. they should have made 3 separate builds of each using the compiler optimizing setting for each architecture.
They then should have compared each processor using its own optimized version of each build of LAME.
Incidentally the AMD and INTEL would probably benefit more from the compiler than apple since its the amount of optimization that GCC does for your code for the G5 isn't great right now. It is such a new processor that the performance will improve. Unfortunately the Opteron and P4 will not make such big strides forward. Their software is as optimized as it will get at the moment what with extensive use of SSE2. However the athlon should get a 15% boost when it moves to 64 bit.
Incidentally i plan to compile LAME for all three architectures and run a test and see how i get on. I expect the G5 to handily beat the P4 and Opteron.
Incidentally the rest of the tests conducted by macworld and pcworld are just a scewed....

However

i_wolf
Oct 25, 2003, 04:41 PM
its just that there were so many holes. I am not a MAC fanboy at all. Indeed i own an Opteron. So i can compare like versus like.
I actually went out and bought the macworld magazine today. Its filled with even more garbage tests....
For example....
they ran an MP3 encode test... comparing the encode speed of itunes versus music match.
Now there are a few HUGE problems with this... besides the drive speeds which make a huge difference what they should have done was read in as WAV files and tested the convert to mp3 using a standard mp3 encoder that is available on all platforms.
The mp3 encoder that comes with music match is terrible. Sure i can encode mp3's extremely fast but they sound terrible. As far as im aware .. i could be wrong because i dont use music match anymore they use the Xing encoder, which is extremely fast but creates terrible quality encodes and which does no analysis before it encodes. The itunes encoder is remarkably better than the music match however it still isn't great. it does a better job. HOWEVER.... i ran an encode test on Opteron dual using iTunes for windows and compared against a dual G5. the G5 was double the speed. Is this a fair test. No. without a doubt the itunes version for windows is not as fast hasn't been optimized. However the itunes for mac looks like it has absolutely no G5 optimizations... and the G5 is simply running legacy G4 code in 32 bit mode. At the very least apple should improve this by recompiling itunes for G5 this in of itself should give it 30% improvement. If they want to double or tripple the performance they should make calls to the specific hardware on the G5.
Further more when i encode mp3 on my pc. I don't use itunes or musicmatch. I use the most popular and BEST encoder around . LAME. hydrogen audio and headfi and other audiophile places all agree this produces the best mp3's. Its also open source so one can compile your own version of lame for your specific processor. Even when using LAME Pentium 4 optimized version of LAME (it is compiled with SSE2 extensions and there is SSE2 calls in the code) Lame still encodes at probably 1/3 of the speed of the Xing encoder used in Music Match. But the quality difference is huge. THere is much more audio processing done on what should and shouldn't be thrown away. This results in higher quality output.
Now for a fair MP3 encoding test. They should have gotten the unoptimized code / basic LAME source code.
They then should have used a general compiler for all three processors (P4, Opteron, PPC970)... such as GCC 3.3. they should have made 3 separate builds of each using the compiler optimizing setting for each architecture.
They then should have compared each processor using its own optimized version of each build of LAME.
Incidentally the AMD and INTEL would probably benefit more from the compiler than apple since its the amount of optimization that GCC does for your code for the G5 isn't great right now. It is such a new processor that the performance will improve. Unfortunately the Opteron and P4 will not make such big strides forward. Their software is as optimized as it will get at the moment what with extensive use of SSE2. However the athlon should get a 15% boost when it moves to 64 bit.
Incidentally i plan to compile LAME for all three architectures and run a test and see how i get on. I expect the G5 to handily beat the P4 and Opteron.
Incidentally the rest of the tests conducted by macworld and pcworld are just a scewed....