[Eug]

http://www.apple.com/powermac/performance/

[Arkham_c]

Funny, I was just going to post that. The page is here:

http://www.apple.com/powermac/performance/

Apple's dual-1.8 is the real value leader now. If I didn't already have the dual-2GHz I'd be getting the dual-1.8 now.

[soul searching]

Originally posted by Arkham_c:
...Apple's dual-1.8 is the real value leader now. If I didn't already have the dual-2GHz I'd be getting the dual-1.8 now.

I almost feel bad for those who got single 1.8s (let's face it, they still have a G5 ).

I wonder if we'll see any revisions in January-March now. This update almost kills those rumors.

[Eug]

Interestingly, the SPEC graph is now gone. Hmmm.... I guess Apple didn't like all the controversy. Or else they've already milked it for all it's worth and feel they don't need it anymore.

Originally posted by soul searching:
I almost feel bad for those who got single 1.8s (let's face it, they still have a G5 ).

I wonder if we'll see any revisions in January-March now. This update almost kills those rumors.

No it doesn't. This is just an interim rejigging of the lineup. The CPU speeds haven't even changed. I fully expect to see an update in Jan/Feb.

[driven]

I would almost be tempted to buy right now, but since I'm working 80-85 hour work weeks until mid-late February I wouldn't have time to play with it anyway.

So .... I'll wait and see what the line up looks like then.

(I'll certainly be buying something at that time unless these long hours continue.

[gperks]

Would be curious (and hopefully not depressed) to see a Powerbook G4 on there.

[DrBoar]

Now the midrange offers the most bang for the bucks instead of the top of the line . Let me guess that Apple keep the SP-DP-DP and in 04 Q1 do a speed bump, not bad at all.

[cenutrio]

Originally posted by gperks:
Would be curious (and hopefully not depressed) to see a Powerbook G4 on there.

Trust me, you would be depressed. The powerbook lineup, although good looking, lots of features, quite decent prices, etc, lacks a competitive processor and system bus.

To check my point
http://barefeats.com/pvp.html

Still, nice machines but weak compared to PCs performance ( tried to find a comparative between centrinos and P4 notebooks with the 15" and 17" alPBs, they were massacred).

I need to replace my 1st generation TiPB, but I'm waiting (I know it may take 16 months).

[olePigeon]

I love it when they use AltiVec enhanced gene sequencing software. The G5 is more than 3x faster than the Xeon.

Originally posted by cenutrio:
Still, nice machines but weak compared to PCs performance ( tried to find a comparative between centrinos and P4 notebooks with the 15" and 17" alPBs, they were massacred).

If you use an AlienWare as your example I'm gonna kick your butt. That thing has like 1 hour battery life when playing a game (2 hours with Microsoft Word.) It's a desktop replacement or mobile PC, but it is definately NOT a laptop in the traditional sense. You're not gonna use it without plugging it into the wall.

[cenutrio]

Centrinos kick G4s too. I hate it but I'm not blind.

[i_wolf]

centrinos are great. but, they dont run OS X!
On another note, they are about clock for clock with the G4 currently n'est pas????
So for the G4 to be competetive with the best Pentium M apple would need a G4 1.7????

[cenutrio]

Originally posted by i_wolf:
centrinos are great. but, they dont run OS X!
On another note, they are about clock for clock with the G4 currently n'est pas????
So for the G4 to be competetive with the best Pentium M apple would need a G4 1.7????

I fully agree in your OS X comment. Still, it is difficult to me to upgrade my trusty TiPB 400 G4 (40 GB 5,400/384 MB) for a new one, when PBs performance is just poor.

PC notebooks vs. Powerbooks benchmarks

Of course, the size, weight, looks, and features factors make the PBs extremely appealing. But, I'll wait until the notebook speed catch up with the windows world (as the desktops did 3 months ago).

Panther has improved my PB capabilities a lot (overall speed increase around 10-15 % based on my benchmarks). Besides, I have G4s and G5s around (in my lab) in case I need raw power.

[thunderous_funker]

Unless you're playing games or doing full professional workstation stuff, I just don't see how the Centrino performance edge outweighs the portability advantage of the PB.

Of course, once those crafty engineers figure out how to shoehorn a G5 in there, Apple will have its own Desktop-replacement mobile workstation.

I just hope they keep something in the line up that caters to those of us who prefer small, light, cool and unplugged.

[i_wolf]

a g5 iBook!
imagine it... the cool plastic casing would melt as soon as you powered it up. It could double as a camping cooker.
I want to patent that idea now....
iBook camping cooker (c) i_wolf
usb powered kettle (c) i_wolf
usb powered toaster (c) i_wolf

The usb powered devices would be great for lazy ass programmers like me.... dont have to get off my todd for a cup of tea or a some toast. Genius really.


Jokin aside
you could have ferrari engine in 15 year old lada body. Engine capable of 200 mph but aeodynamics and weight cripple it to 120.... before it breaks down and you gotta start pushin.
Or
You could have a Lotus elise engine in a Lotus elise body. Engine capable of 150, in optimum conditions and good road it will easily hit 150! Normally we talking 120 mph

Now the lada above on an engineers rack will test faster in pure grunt. Real world... nada

The Lotus ... well its pretty much what you see is what you get... damn damn fast and pretty efficient.

Which is the better buy above... its the same with Apple laptop and PC laptop. The body above is the analogy of the OS.

Regardless of raw speed. which is the most productive tool and which can do more. I reckon the PB would win that race.... and its portably too.

[Eug]

Car analogies are always bad. A nice PowerBook design does not preclude a fast chip. Otherwise we'd all be running 5 Watt G3 chips from 3 years ago or whatever.

There little stopping Apple from putting a G5 in a laptop today from a purely technical point of view. They could throw in a 1.2 GHz G5 if they wanted, but that doesn't really help much in terms of playing the GHz game.

Shortly after when the new 90 nm G5 appears, we'll see it in the PowerBooks. A 1.6 GHz G5 90 nm would work just fine methinks.

[Superchicken]

Even a 1.3Ghz G5 would be acceptable so long as it goes toe to toe with a centrino 1.5Ghz I'd be happy.

[DaBeav]

Dual Opteron benchmarks are conspicuously absent...

Will the Athlon 64/FX-51 be MP capable?

The 1.8 and 2.0 G5s do have the benefit of an extra CPU...

[Spliffdaddy]

There are many laptops on the market that safely dissipate much more heat than that produced by a 2GHz G5. Don't let those massive heatsinks fool you - the G5 still operates at a lower wattage than a comparable P4 or Athlon.

The problem with using a G5 in a laptop would be short battery life. This is no different than the short battery life you'll find in pre-Centrino Pentium4 laptops, for example.

Historically, when desktop CPUs have been used in laptops, they have employed some method of throttling the processor's clockspeed (ideally based on CPU load) in order to increase battery life.

[blackwind]

Originally posted by DaBeav:
Dual Opteron benchmarks are conspicuously absent...

Will the Athlon 64/FX-51 be MP capable?

The 1.8 and 2.0 G5s do have the benefit of an extra CPU...

The Athlon 64 FX-51 is not MP capable. That is what the Opterons are for...

As for floating-point tests, in many cases a 2-GHz G5 is about equivalent to a 2-GHz Opteron. In some things, the G5 will be slower but in others it will be faster - most notably, and most recently, from the November 2003 LINPACK supercomputer results.

http://www.top500.org/list/2003/11/

3) 2200 x 2-GHz G5's = 10280 Gflops
4) 2500 x 3.06-GHz Xeons = 9819 Gflops
5) 1936 x 1.5-GHz Itanium 2 = 8633 Gflops
6) 2816 x 2-GHz Opterons = 8051 Gflops

2-GHz G5 ~ 4.672 Gflops
1.5-GHz Itanium 2 ~ 4.459 Gflops
3.06-GHz Xeon ~ 3.928 Gflops
2-GHz Opteron ~ 2.859 Gflops

In this case, the G5 rules the roost. Of course, there will be other cases where the G5 pawned in the same way as its competitors in LINPACK...

[i_wolf]

i would imagine that the reason that the Operton dual's were absent is because according to apple bench's shown... at those speeds the single p4 3.2 (is it the emergency edition??) is faster tahn the fx51 2.2Ghz Opteron/Athlon 64. According to apple the baseline used is a single P4 3.2GHz.
Now given the extreme lead that the dual G5 had in those leads, even if the Opteron 2.2 bench's doubled with a second processor it would still be behind the dual G5 (if we extrapolate on the information given to us by apple).At least in the figures from apple the only bench that a dual 2.2 opteron would likely equal or come near the G5 would be in the mpeg 2 encoding in the others according to apple it would likely come away short.
The keyword here is "according". I would like to see some independent bench's first! No doubt though as software starts to become optimized for G5 and when a "compatible" optimized GCC compiler is released that meets the standards set by XLC and XLF so far, it will tear away from the rest of the pack more. At least theoretically the G5 dwarfs the Opteron and Xeon in terms of GFlops. But unless a suitable compiler is used it will only be competetive in real world.But as of this point in time i would take those bench's with a grain of salt.
On a separate note, ... seamingly there is a new compiler for AMD which according to the inquirer is promising as much as 40% improvment in 64 bit mode over current 32 bit mode with a recompile. This is awesome. While like the IBM compiler for PPC970, it is in beta presently and presently only works in certain situations this is really great.
http://www.theinquirer.net/?article=12749
This 40% improvement so far is only in a single test with all other tests showing 15% improvement. Just thought some of you guys would find this interesting.

[blackwind]

There are indeed a fair bit of third-party (mostly scientific-based) benchmarks comparing the G5 against its nearest competitors. In those tests (I wish Ars Technica was up so that I could find the original links), a 2-GHz G5 was almost always faster than the 2-GHz Opteron, but that was generally because of xlf's astounding performance.

As for the Pentium 4 EE, I am not sure why everyone likes to refer to it as the "emergency edition". The Pentium 4 was (still is in some cases) the fastest processor for most of 2002 and 2003. Just because it is getting long in the tooth now does not mean that it is an inferior processor.

A 3.2-GHz Pentium 4C beat the 2-GHz G5 by a slight margin in all of those scientific tests (except for one, if I recall correctly), and the G5 beat the 3-GHz Pentium 4C and the 3.06-GHz Xeon in most of those tests.

In Cinebench 2003, which is very optimized for the Pentium 4, the gap is even bigger. Consider these raytracing scores (higher is better):

3.2-GHz Pentium 4C: 384
2.2-GHz Athlon 64 FX-51: 310
2-GHz G5: 288
2-GHz Opteron: 281
2-GHz Athlon 64 3200+: 281

The G5 does well against the AMD processors, but the Pentium 4 is superior in this case. Clearly, the Pentium 4 is not as dead as some people seem to believe.

[geekwagon]

Originally posted by blackwind:

As for the Pentium 4 EE, I am not sure why everyone likes to refer to it as the "emergency edition". The Pentium 4 was (still is in some cases) the fastest processor for most of 2002 and 2003. Just because it is getting long in the tooth now does not mean that it is an inferior processor.

It has more to do with timing than anything else. It was the typical Intel "Oh crap, one of our competitors is about to make an announcement, slap something together and announce it the same day to ensure that any press about their processor mentions us too!" thing.
They do this every single time, and have as long as I can remember..

EDIT: Hah, beat you by one second, Eug.

[Eug]

Originally posted by blackwind:
As for the Pentium 4 EE, I am not sure why everyone likes to refer to it as the "emergency edition". The Pentium 4 was (still is in some cases) the fastest processor for most of 2002 and 2003. Just because it is getting long in the tooth now does not mean that it is an inferior processor.

It was released in response to the Opterons, and was never a planned CPU. It's quite an expensive part, because of the good chunk of L3 cache included. It was supposed to be L3-cacheless P4 --> L3-cacheless Prescott. But because of Prescott delays they had to release the P4 Emergency Edition, to take some of the lustre off of the Opteron and Athlon 64 releases.

[SamuraiDL]

I have a single 1.8...asuming you all have stopped vicariously cringing/laughing/crying for me, How silly does an update sound, financialy minded or not, does apple let you trade in, and if not, then why?

[i_wolf]

Blackwind: Those ars bench's with the G5, in most cases it does wallop the competition in scientific stuff. Correct. But they were bench's with GCC! the XLF stuff was truly incredibly faster by orders of multiples but wasn't officially tested by ars, it was run by users in the forums who were reporting their results. With regard to the bench's ars did conduct the G5 beat both the EE P4 and the 2GHz Opteron in photoshop overall results and in most scientific results. In the forums the science marks were practically two or three times faster than the competition in order of magnitude. But as impressive as those results are, i wouldn't imagine that those results are indicative of real world perf. more real world potential when a good compiler is used.
Cinebench is extremely extremely optimized for the P4. Those numbers if i can remember are now out of date since the manufacturers did a small optimization and recompile with GCC and got 30% performance improvement again on G5. As they themselves state on their website, its extremely early days and they are only starting out on the optimization path. Currently the existing Cinebench alpha/beta G5 version is faster than the Opteron 2Ghz but not quite as fast as P4. But again the manufacturers themselves on their website say they have a huge way to go in terms of optimization and that performance is definately gonna rise in that one.
I would also expect the Opteron situation to improve in Cinebench big time with optimizations also. Its on die memory controller should make a massive difference especially in cinebench, i think that existing results are not indicative of the potential the chip has in cinebench.
Remember the P4 has been out for years. When it was launched initially it was a bit of an ugly duckling. it took years of optimizations before it has become the swan we have today. Out the gates neither the G5 nor the Opteron are ugly ducklings, both are extremely fast using raw brute processing power, but especially in teh case of the G5, it aint even touching its potential yet.

[blackwind]

With regards to the scientific benchmarks (not done by Ars Technica, but they were the ones who had links... unfortunately, Ars Technica is down right now...), most - if not all - of them were done with xlf. (GCC cannot compile FORTRAN.)With Absoft's FORTRAN compilers, the G5 was lagging quite a bit. The only scientific benchmark where the G5 has a truly tremendous (multiple-fold) lead over its competition as a result of xlf is with NASA's Jet3D.

My work involves the sciences. The G5's good performance in scientific applications is good for me and is indeed indicative of real-world performance (at least for me).

The G5 is currently tops in Photoshop 7.0.1 (at least according to Ars Technica). The differences between high-end Photoshop machines, however, are still fairly small.

The G5 Cinebench score that I listed was from the G5-optimized version. (It is still only a beta, though.)

The Opteron would indeed benefit from optimizing Cinebench, as would the Athlon 64 FX. After AMD releases its compiler, I am certain that Maxon will provide an updated version of Cinebench, thus putting the G5 in last place again. The current G5-optimized version of Cinebench only allows the G5 to have a marginal lead over an equally-clocked Opteron.

The Pentium 4 is indeed a swan now, but it is not just because of software optimizations. Numerous hardware tweaks have helped enormously. If we take one of the original 1.5-GHz Pentium 4's now with optimized software, it still would be pretty bad compared to its contemporaries.

In multithreaded applications (such as Cinebench), the key advantage of the current Pentium 4 is its Hyperthreading. As a result, we have a processor with good per-clock performance with the highest clock speeds. This results in a monster processor with brute raw power. The G5, Opteron, and Athlon 64 FX are not alone in hardware advantages.

What we should really do is run Cinebench 2000, which does not have Pentium 4-specific optimizations, on the machines of interest and see how the respective processors compare. That way, none of them would be running specialized code, and it would be a more fair comparison of the processors themselves.

No matter the result of all this debating, it is clear that the G5 has established a fairly good reputation in its early days, unlike the ruckus of the G4. This in itself is the a great blessing.

[olePigeon]

Cinebench still doesn't use AltiVec. Until that happens these benchmarks are pointless.

[Eug]

Originally posted by olePigeon:
Cinebench still doesn't use AltiVec. Until that happens these benchmarks are pointless.

I don't buy these types of arguments, esp. when Cinema4D has already had *some* G5 optimizations.

Cinebench is a popular Mac program. If it runs slower on a Mac then Mac-heads will have to deal with it.

[driven]

Does the G5 still have an Altivec? (Pardon my ignorance)

[olePigeon]

Originally posted by Eug:
I don't buy these types of arguments, esp. when Cinema4D has already had *some* G5 optimizations.

Cinebench is a popular Mac program. If it runs slower on a Mac then Mac-heads will have to deal with it.

Oh I see how it works. So why were people up in arms about Apple's benchmarks and the compiler not being optimized for the Pentium 4?

Or is it that only Mac users can do benchmarks wrong and that it's OK for PC venues to use one-sided benchmarks?

[Eug]

Originally posted by olePigeon:
Oh I see how it works. So why were people up in arms about Apple's benchmarks and the compiler not being optimized for the Pentium 4?

Or is it that only Mac users can do benchmarks wrong and that it's OK for PC venues to use one-sided benchmarks?

Because the compiler was available. xlc/xlf for Apple was not.

I don't disagree with Apple's GCC test, but the Intel advocates had a point. People actually use icc/ifc to make programs, and anybody could buy it. In fact, it's free for educational use.

If Apple really wants to strut the G5's stuff, they really should do an xlc/xlf SPEC test and publish it at www.spec.org for the whole world to see. But they haven't yet for whatever reason.

[Turias]

Originally posted by driven:
Does the G5 still have an Altivec? (Pardon my ignorance)

Yeah, they do. I think they may have had to change the name, though. I heard something about the Altivec name being owned by Motorola, but I could be wrong.

[geekwagon]

Originally posted by Eug:

I don't disagree with Apple's GCC test, but the Intel advocates had a point. People actually use icc/ifc to make programs, and anybody could buy it. In fact, it's free for educational use.

Sure, but MOST Windows programs are going to be compiled by Visual Studio, and MOST Linux programs are going to be compiled by GCC. So I think that benchmarks using those compilers is much more interesting than icc (which seems to be pretty much limited to usage in the sciences, AFAICT..)

I guess I am coming at it from the angle of someone who uses off-the-shelf software, though, and not someone who needs to run physics models written in Fortran.

[olePigeon]

That's exactly my point. Why are they using Cinema when they could be using Lightwave 3D for benchmarking? Something that's designed efficiently for both Intel and PowerPC?

If people are going to nitpick about GCC vs. ICC/IFC, then they should nitpick about Cinema vs. Lightwave 3D or Maya (which I still haven't seen any benchmarks for.)

Originally posted by Eug:
Because the compiler was available. xlc/xlf for Apple was not.

I don't disagree with Apple's GCC test, but the Intel advocates had a point. People actually use icc/ifc to make programs, and anybody could buy it. In fact, it's free for educational use.

If Apple really wants to strut the G5's stuff, they really should do an xlc/xlf SPEC test and publish it at www.spec.org for the whole world to see. But they haven't yet for whatever reason.

[Eug]

Stating the obvious, but:

If you're doing SPEC, you pick the fastest available compiler for SPEC.

If you work in Photoshop, you test Photoshop.

If you work in Cinema4D, you run Cinebench.

It you do a lot of scientific floating point, you test that.

There's not much point in screaming about how great the dual G5 is at Photoshop if your point is to run Cinebench.

[blackwind]

In response to olePigeon:

Cinema 4D XL (and consequently Cinebench) are not optimized for Altivec because Maxon engineers believed Altivec to provide negligible improvements.

For texturing and shading, Altivec may be used, but for raytracing (which is double-precision or higher), Altivec is useless.

Since raytracing takes up most of the processing time, there is little or no point for Maxon engineers to spend extraordinary efforts to implement Altivec code. It creates a lot more work for a couple of percentage points worth of improvement.

As for Lightwave 3D 7.5, a few benchmarks on the raytracing exist.

From www.blanos.com/benchmark:

Raytrace Scene:
Dual 2.8-GHz Xeon: 50 s
Dual 2-GHz G5: 63 s
Dual 1.42-GHz G4: 90 s

A hypothetical Dual 2-GHz G4 would score 64 s, so it is obvious that the G5 needs optimization in Lightwave 3D as well. (Version 8 is supposed to have improvements for the G5.)

Other Lightwave 3D scores are available at the aforementioned website for anyone who would like to peruse through some more scores.

I remember some Athlon 64 FX Lightwave 3D scores are available elsewhere as well, so I will try to find them.

[Kam Lion]

http://www.fftw.org/speed/


Originally posted by blackwind:
"As for the Pentium 4 EE, I am not sure why everyone likes to refer to it as the "emergency edition". The Pentium 4 was (still is in some cases) the fastest processor for most of 2002 and 2003. Just because it is getting long in the tooth now does not mean that it is an inferior processor."

What you have said has no proof in a well designed benchmarking. You are dealing more with speculations in terms of comparing G5 to Pentium 4. The Power PC 970 is very powerfull chip.

Please, read the link.

[blackwind]

I found a few more scores. All tests are single-threaded to test only one processor at a time. Note that the Pentium 4 can actually run two threads simultaneously as a result of Hyperthreading, but these scores do not take advantage of it due to the single-thread restriction.

G5 score: www.blanos.com/benchmark
Other scores: http://www.xbitlabs.com/articles/cpu...4-3200_13.html

*Dual 2.8-GHz Xeon: 50 s
*Dual 2.0-GHz G5: 63 s
*3.0-GHz Pentium 4C: 85 s
*Dual 1.42-GHz G5: 90 s

2.2-GHz Athlon 64 FX-51: 88 s
3.2-GHz Pentium 4 EE: 90 s
3.2-GHz Pentium 4C: 93 s
2.0-GHz Athlon 64 3200+: 95 s
2.0-GHz Opteron 146: 97 s
2.0-GHz PowerPC G5: 114 s
2.2-GHz Athlon XP 3200+: 133 s

*Indicates that multiple-threads are used.

Let us hope that Lightwave 3D 8 really does have some impressive G5 optimizations...

[stingerman]

Originally posted by DaBeav:
Dual Opteron benchmarks are conspicuously absent...

Will the Athlon 64/FX-51 be MP capable?

The 1.8 and 2.0 G5s do have the benefit of an extra CPU...

Considering that an Opteron costs $900 a piece and that the A64 FX-51 is a Opteron, the benchmarks are pretty well matched.

[stingerman]

Originally posted by SamuraiDL:
I have a single 1.8...asuming you all have stopped vicariously cringing/laughing/crying for me, How silly does an update sound, financialy minded or not, does apple let you trade in, and if not, then why?

The Single 1.8 has been re-priced to 2099. So sell yours on ebay and get the dualie. Of course in a couple of months, they'll upgrade again

[stingerman]

Originally posted by cenutrio:
I fully agree in your OS X comment. Still, it is difficult to me to upgrade my trusty TiPB 400 G4 (40 GB 5,400/384 MB) for a new one, when PBs performance is just poor.

PC notebooks vs. Powerbooks benchmarks

Of course, the size, weight, looks, and features factors make the PBs extremely appealing. But, I'll wait until the notebook speed catch up with the windows world (as the desktops did 3 months ago).

Panther has improved my PB capabilities a lot (overall speed increase around 10-15 % based on my benchmarks). Besides, I have G4s and G5s around (in my lab) in case I need raw power.

Their comparing different types of Notebooks, if you are going to lug an Alienware, around why don't you just go ahead and attach an LCD and car battery to the G5 and lug that around too. However, a 1.25 G4 is a pretty fast Notebook, especially with those 9600 GPU's in them. The benchmarks you are looking at do not really tell the performance of the system. Try this test and you'll see what I mean, burn a DVD video at the same time as you try browsing the web and playing a song on iTunes, then do a similar multi-tasking job on the Wintel Notebook!

The Mac with OS X will not even skip a beat, with the Wintel Notebook, it will just die until the burn is done. The benchmarks do not show you the real performance of these systems. The Powerbooks with OS X are well balanced systems, that handle multiple tasks and perform very well because of this.

[Spliffdaddy]

Burning a CD or DVD doesn't negatively affect Windows. Multitasking works very well - just as in OSX.

Remember, kids, FUD is not an adequate substitute for facts.

[driven]

Originally posted by Spliffdaddy:
Burning a CD or DVD doesn't negatively affect Windows. Multitasking works very well - just as in OSX.

Remember, kids, FUD is not an adequate substitute for facts.

Large file copies over a network sure does slow it down though. (At least on our P4 2.8s ....

[i_wolf]

blackwind true they could release an AMD compiler, however the AMD processor does not have the potential for future optimization that the G5 has. Most of the bench' in ars forums "only" showed 50 to 60% improvement over the next best competition equivalent (amd opter 2Ghz) in some cases. But these were primarily integer bench's where people would traditionally have expected the opteron to crush the G5. Jet 3d was probably the only bench that was more a bench of pure FPU power. AGAIN XLC and XLF are not altivec auto vectoring compilers. And in this bench the G5 double and tripled a Xeon's score.
however the XLF and XLC compilers are extremely young. And probably the first "compatible" compiler for the G5 in the sense that it can generate code that will execute in parallel across the multiple units of the G5. Currnet G5 as ars themselves speculated originally does not do this and the advanced design (wide AND deep) of the G5 which allows for extreme parallelism for huge performance gains, does need a compiler to generate optimal machine code. Ars were proven right, when the original bench's were released for XLC and XLF. It seams that the reason that the G5 was so competitive and neck and neck so far has been because of raw brute horse power (probably like the opteron). However particularly in the case of the G5 it is a much much newer and arguably more advanced design than the G5 that allows for much much higher performance if both chips run under optimal conditions. I think the GFlops of the 970 is nearly double that of the Opteron. However for the G5 to actually operate near its theoretical it needs a modern compiler like XLC and XLF.
The optimizations for cinebench so far are extremely young and basic. GCC 3.3 has huge huge headroom for performance enhancement for the G5 in future releases. Indeed Apple and IBM are both apparantly working on this heavily. At the developers conference IBM said they are commited to bringing GCC performance to similar levels as what XLC and XLF will be expected to produce.
Note that presently i am not comparing 32 bit perf since most users would use operon with win xp and G5 with panther which are both 32 bit. Currently the opteron is heavily based on the Athlon XP. GCC and other compilers already produce good code for the Athlon XP and hence very good for the Opteron. It can also make extensive use of SSE2 in apps which have p4 SSE2 optimizations. The G5 has neither benefit. Its a much much newer and arguably more advanced design, wide and deep and extreme parallelism, whose current performance (which operating well well well below its theoretical performance) is so extremely competetive becase of raw horse power. Also the G5 cannot leverage its altivec unit the same way as the Opteron can leverage its SSE and SSE2 SIMD units because of autovectorised code that compilers presently produce.
http://www.theinquirer.net/?article=12749
This is interesting as it provides 40% perf increase and is a highly tweaked 64bit compiler for AMD. However the XLC and XLF are already producing much much higher gains for the G5 and again this is on a much younger architecture than that found in the opteron.
Dont get me wrong, i love AMD chips. I am a regular at 2cpu.com and am a big AMD fanboy. I own a dual 2.8Xeon on iwill DP533 and i also own a dual Opteron 246 2GHz workstation. While the Opteron is undeniably faster than the Xeon. There is no way that my Opteron is faster than the G5. I spent hours recompiling code with ICC (intel compiler) for the Opteron and using XLC for the G5. In most heavy number crunching apps that we use which doesn't have any altivec useage presently but which we have SSE2 modules for, the G5 crushes my opteron by a considerable margin. I didn't believe it myself until isaw it. Having said that, its unlikely that the SSE2 made much diff for my xeon or opteron since its only a small optimization. Comparing scores between G5 opteron and xeon, with Gcc on the same code, the G5 was marginally faster maybe 15% than the Opteron which was significanly faster than the Xeon.
For all the advanced tech in the P4, it is pretty much limited to multimedia apps as any type of heavy number crunching with lots of branches , whether hyperthreading or not, it takes a mass penalty certainly much more than the Opteron or G5.
As far as Hyperthreading yes it certainly benefits from more parallslism than anything else. This is why i would imagine that a chip like the G5 should benefit from optimizations more than the Opteron. The G5 is designed for extreme parallelism as already explained with the different functional int , fpu etc.. units. Simply put the current build of cinebench (which i understand (correct if im wrong) is compiled with GCC3.3) simply could not be running the different units in parallel as is the case with XLC. It also does not use Altivec. I would have to disagree that altivec would make a big difference also if it was incorporated. But thats open for debate.
Im not saying that what I say here is gospel. im sure there will be some apps that would benefit better from an opteron under optimal conditions better than a G5 under optimal conditions. but i would imagine that when support matures in compiler and software optimizations for both platforms that they can run at their respective potentials, the G5 should crush the Opteron in most apps and majority of conditionsto an equiv clocked Opteron.
In the mean time im in the lucky situation that i get to run a dual Opteron 2GHz and 2GHz G5 in work side by side!

[riverfreak]

Hey, that is cool to see them using HMMer for benchmarking! Right on! I was getting sick of the boring photoshop filter tests.