You might want to take a look too. The report also says that the tests were not optimized for the G5 either.

The spelled "Macs" instead of "Mars"

Reports are saying Panther isn't 64-bit...talk about lame. I hope there is a G5 edition or something.

If they have a pre-release G5 there are several functions not yet enabled on the G5 chip that are enabled on the Pentium.

When Apple did their tests they dissabled the Pentium fuctions that were not ready on the G5 (but would be at time of release).

With all everything functional the G5 will scream past the Intel chip and later when software versions get bumped for 64 bit processing, It will be like getting a bonus speed boost!

Within 12 months there will be a 3Ghz version wich will widen the gap that much more!

Finally, God's Kingdom has come! Satan is being overthrown!

If they have a pre-release G5 there are several functions not yet enabled on the G5 chip that are enabled on the Pentium.

When Apple did their tests they dissabled the Pentium fuctions that were not ready on the G5 (but would be at time of release).

With all everything functional the G5 will scream past the Intel chip and later when software versions get bumped for 64 bit processing, It will be like getting a bonus speed boost!

Within 12 months there will be a 3Ghz version wich will widen the gap that much more!

Finally, God's Kingdom has come! Satan is being overthrown!

"Reports are saying Panther isn't 64-bit...talk about lame. I hope there is a G5 edition or something."

And what the heck does that mean to you? If you are thinking it is slower, then go do some research. Making Panther 64-bit will NOT make it faster. It will just let it address more RAM.

If you were able to get through the NASA technical document, you would have no problem with the G5 white paper. Basically, the single G5 has dual Altivec units whereas the G4 has only one. The NASA apps were compiled for the single Altivec G4. Thus, when the NASA apps are recompiled to take advantage of the dual G5's quad Altivec units, I would expect the benchmarks to show results at least 4X what they got here. Factor in Panther and other compiler optimizations and that 3.2ghz P4 will seem like a joke. By the way, anyone else notice the NASA guy was running OS X 10.2.7??

Did anyone else notice that all the graphs in the article noted that the dual macs had their second processor disabled. So, chip to chip the G4 compared well to the P4 and more importantly the single G5 doubled those marks!

Where did .7 come from? Is that part of the security update?

>anyone else notice the NASA guy was running OS X 10.2.7??

That's the OS version the G5s will initially ship with.

I think the 3 ghz G5 will be available much sooner than 12 months. Steve is only saying "within" 12 months so as not to soften sales of the initial 1.6ghz -2.0ghz units. I would be surprised if it is more than 6 months after release of the first G5's and we don't see the 3.0ghz machines. The 12month remark was also intended to give Intel a false timeline so that IBM/Apple can surprise the world again in just a few short months. If you don't believe this, look at the facts: the G5 fab process is superior to that used by Intel and Intel is getting 3.2ghz out of a lesser fab; the IBM guy says they are already building G6 protos; 2.0ghz is not where Apple wants to be; Apple/IBM now have the chance to be ahead in MHZ once and for all and they will not pass up this opportunity; in 12 months there will be 4.0ghz P4s.

If you read the vector results, the G5 at 2ghz had about twice the MFLOPS of the Xserve at 1ghz. So the paper concluded that the G5's vector performance is on par with the G4 per mhz. But as mentioned above, the G5 actually has dual vector units (twice what is in the G4) therefore an optimized benchmark for the G5 would smoke the results of the unoptimized G5!

NASA confirms Apple's own tests.. This is big..

Maybe thats why Apple stock just went up 4% ~ 1 dollar? :)

I am I reading things wrong or did the Pentium systems only have 2MB of Ram (P4 2ghz) and less than 1MB RAM (P4 2.66)while the G5 and G4 systems had between 1GB and 2GB Ram?

Pentium 4 (2GHz)MemTotal: 2065600 kB

Pentium 4 (2.66GHz)MemTotal: 905720 kB

G5 (dual 2Ghz PowerMac)hw.physmem = 1073741824

G4 (dual 1Ghz Xserve) hw.physmem = 1610612736

G4 (dual 1.25GHz PowerMac)hw.physmem = 2147483648

The G5s whitepaper states that a single G5 can handle 215 inflight instructions, over 13 TIMES AS MANY as the 16 instructions the G4 can handle. The NASA guy will be blown away when he runs his apps as native G5. I found it interesting that the nasa guy had to disable SSE/SSE2 compiler flags (Intel's equivolent to Altivec) because his apps ran faster without them! So what this means that an app optimized for the P4 actually runs slower than one that is not. And everyone knows that to tweak performance you have to disable Hyperthreading on the new Intel chips. Intel calls themselves a technology leader? Without high MHZ, they have squat.

The Pentium memory totals are listed in kB while the Apple memory totals are listed in bytes.

...is a browser that can actually render all websites as accurately & quickly as my pee cee...

Oh yeah & an email app as fast as Outlook Express in OS-9...?

Perhaps even for the same or lower price?
;-)

The G5 does not have 4 Velocity Engine or even 2. It has one Velocity engine. It's VMX implementation is identical to the Altivec engine that was found on the PPC 7400 and 7410. The 350 to 533 MHz G4's. This implementation had two executable units. A Vector and a Vector Permute. These two units are both part of the whole Vector unit. The PPC 7450 had 4 sub units. Vector Integer, Vector Float, Vector Permute, and one other I think a simple integer unit perhaps. This did not double the Vector units speed. but gave programers and compilers more flexibility in dispatching commands to the unit. So that G5 actually has a older Vector unit than the current G4's but it has a clock rate advantage and a bandwidth advantage. Which in the test done by this NASA bench mark seem to equal out. Video work should be more efficient on the G5 and things like RC5 will probably be more efficient on the current G4's.

But what impressed me was the complete stomping the Vector enhanced code gave the PC. That's real world. They are obviously going to use that code base on the G4's and G5's. and note that using the PC Vector unit (SSE2) slowed Jet3D performance. so the G5 in real world use stomped the P4 with a rate of 5177 MFLOPS vs. 255 MFLOPS.

I know which platform and code base I would pick for this application after reading this test. A cluster of G4 Xserves or if just for a single workstation a G5.

Macnn stated that the G5 beat the P4 3Ghz, but the only comparison was against a P4 2.66Ghz which was beat with a slight margin only.

Furthermore, they concluded that the G5 would lag behind the P4 3.2Ghz based on their extrapolation of the testing. And this is with Hyperthreading disabled and against a dual G5. Hmm. Another test to prove otherwise.

"I would be surprised if it is more than 6 months after release of the first G5's and we don't see the 3.0ghz machines."

I can't say one way or the other whether that will happen. I just would not encourage anyone to use that as an excuse not to purchase a machine because 'faster ones are coming'. Faster ones are always going to be coming out, but at some point you have to make a buying decision.

I was so happy to finally read that Mac lovers were happy with speeds for once. Let's not start 'give me a g6' conversation this soon.

We have a browser already that renders pages equally as fast if not faster than IE for the PC- Safari. I am sitting here with my 800 MHZ G4 PBook and my Micron T1000 at P4 2.2 GHZ and the only think that I can see is slower on the Mac is rendering Outlook Exchange Pages. Go figure.
If the Outlook Exchange pages were actually based on Web Standards, rather than proprietary MS Standards, it might load fast too. Outlook Express... yuck.

"Reports are saying Panther isn't 64-bit..."

Actually parts of the WWDC Beta of Panther are coded for 64 bit. Most is still 32 bit, but a lot of 64 bit clean code is in there.

OS X will be fully 64 bit before it reaches 10.4

"Furthermore, they concluded that the G5 would lag behind the P4 3.2Ghz based on their extrapolation of the testing. And this is with Hyperthreading disabled and against a dual G5"

Yes, they did estimate the G5 would be roughly 20% slower than the 3.2 GHz P4, but also mentioned that the G5 could easily gain 20% after optimization. So the 2GHz G5 is roughly equivalent to the 3.2GHz P4.

Listen carefully, please...Hyperthreading is 95% marketing c***. Bench tests from several PC sites have shown it to usually be maybe 2% faster overall, but even slower much of the time. That often-cited "up to 30%" faster hype is only for certain types of applications, and even then the increase is artificial and limited to 20%. For example, the tests Veritest ran for Apple showed *slower* results when Hyperthreading was enabled. That's not all that uncommon.

They mention more than once that their tests only use 1 processor, so this data is for chip-to-chip comparison. Having a 2nd G5 chip in the box will add considerable speed to many apps, meaning that the only fair comparisons at this point are with Xeon and Opteron systems. The single P4 (3.2GHz) is soundly trounced by a dual G5 machine.

"Macnn stated that the G5 beat the P4 3Ghz, but the only comparison was against a P4 2.66Ghz which was beat with a slight margin only."

No....MacNN did not say this. As for the actual test, read the whole thing. Closely. It's quite impressive for the G5.

"Furthermore, they concluded that the G5 would lag behind the P4 3.2Ghz based on their extrapolation of the testing. And this is with Hyperthreading disabled and against a dual G5. Hmm. Another test to prove otherwise."

No... They concluded that based un unoptimized code for the G5, vs. the 3.2 Ghz P4....and this based on a SINGLE PROCESSOR. So....if optimization can make up the 20%, which they said is quite reasonable, then ONE G5 will roughly compare to the P4 3.2 on the test the P4 is best at.

So at WORST an app which ignores the second processor will be as fast on the G5 as on the P4 3.2. Not too bad really.

Gimme a G6! ;) You know by the time G5 ships, Intel would pmost likely have a new processor that's faster than the 3.2 right?

How do you like me now???...

Same as before: we like the hardware but hate the ignorant zealots.

Actually, there is some talk over at MacRumors.com (which was reposted from MacBidoulle) that development of the PPC980 (G6) is being done concurrently with the POWER5. This would make sense as the PPC980 is supposed to basically be one core from the POWER5. Improvements include SMT (HyperThreading), eLiza, and possibly FastPath (speeding up commonly used operations like TCP/IP by moving them to hardware). POWER5 has already been booting code so the G6 may not be as far off as some of us think. (Definitely not as long as the span between G4 and G5 generations.)

They will probably call the 980 a G5

the G4 included many CPU's
7400, 7410, 7440, 7450, 7455, and a 7457 coming soon.

Wait a minute, I thought Maclots poo-pooed the idea of HyperThreading when Intel announced it? Would they also dismiss it as hype if IBM does it?

Don't get me wrong - I am all for anything, even if it's introduced by a competitor, that improves performance of the OS X experience and I believe Hyperthreading would definitely help. I just don't like zealots.

" And what the heck does that mean to you? If you are thinking it is slower, then go do some research. Making Panther 64-bit will NOT make it faster. It will just let it address more RAM."


Gee...wouldn't that make it faster?

I know it won't make the hardware faster, but it sure as heck would make the software faster....otherwise what is the point? Maybe you need to do a little more research.

So I'm a "Maclot" or a zealot if I relate findings from PC sites such as Anandtech and TomsHardware, as well as Slashdot?

I have nothing against Intel, but I have everything against hyped-up marketing-speak, whether it's from Apple or Intel. HyperThreading has been hyped by nearly all the PC mags and columnists who have an interest in pleasing Intel, but independant tests show it to be bogus again and again. I don't know how many times I've seen tests that show better results with HyperThreading OFF.

As for IBM implementing some sort of hyper-threading, if it's as bogus as Intel's then I'll be the first to call them on it.

"Gee...wouldn't that make it faster?

I know it won't make the hardware faster, but it sure as heck would make the software faster....otherwise what is the point? Maybe you need to do a little more research."

No, I wouldn't really make your software any faster. I would allow you to run applications that need such large RAM requirements. Like large scale databases.

But the reports of Panther's 64 bit capability indicate that not all of the OS code will be compiled 64 bit clean. But apps that are compiled that way will take advantage. I see no problems with this. Maybe you should do some research.

The speed advantages of the G5 have to do with it's increased execution unit's, increased FSB bandwidth, and increased number of in flight and dispatched instructions. being able to do 64 bit word length integer calculations are great and so it the 42bit addressable memory space. but saying "64 bit make faster" is just incorrect.

I VERY MUCH doubt that the 980 (if it truly is based on POWER5) would be called a G5. POWER5 is a very different architecture (AFAIK) than POWER4, unlike Mot's 7400 series processors which mostly varied by differences in pipeline and vector unit implementation. IBM has been saying that POWER5 could have up to four times the performance of POWER4. I highly doubt that it would be appropriate then to categorize this processor with the POWER4/PPC970.

As far as SMT (HyperThreading), those in the know will tell you that it's benefit depends on application (as is the case with most things). However, CPU manufacturers must put at least some credit in the technology, since Intel, IBM, and Sun are all implementing some sort of SMT. You wanna drag this school of thought out to the max, look at what Sun did with Time-Space Computing on their MAJC processor. Neat stuff, but it's still in its infancy.

64 bit does not make software run faster. Forgetting memory and bus speed for a second and with everything else being even, clock speed is the #1 determination factor for making software run faster.

Case in point:
#1 - 1.8 ghz G5 running Jaguar and Panther (32 bit vs 64 bit)- there should be little speed difference there.

#2 - 1.8 ghz G5 running Jaguar and 2.0 ghz running Jaguar. 2.0 will run faster.

#3 - 1.8 ghz G5 running Panther and 2.0 ghz running Panther. 2.0 will run faster.

You know, I'm getting sick of some of you trying to debate hardware you haven't actually used. It's all theory to most of you at this point, and with any theory it's debatable. You're so wanting to prove your hardware knowledge is better than the other persons that it just becomes old.

Simple questions should help answer this:

Do any of you have a PowerMac G5 in hand? No
Are any of you running a G5 with Panther loaded on it? No
Do any of you run a system using a 64-bit architecture? Probably not.
Have any of you gone to school and are acredited in any sort of hardware or software developement? Probably not.

More importantly, have you ever worked for Apple???? No.

What bothers me the most is that you're on the outside looking in, trying to be knowledgable about a topics you know little about. It's like a baseball fan saying, "I could've hit that pitch" and really meaning it. Um, no you couldn't have. Your baseball career probably died in Little League for good reason - and the same could be said toward your understanding of hardware.

Don't get me wrong, I like debate and all. But the whole mine is bigger than yours is kind of old.

Any discussion of G5 benchmark is premature, especially discussion of Apple's benchmark which compares a to-be-avail in 3 month system to a 3 month old then system.

By the way, the above comment is not from a Winlot. Challenging Apple's benchmark results is no different from challenging Intel's benchmarks which always show them to be faster than AMD's, or AMD's which always show them to be faster than Intel's, Nvidia's, ATI's, etc.

This discussion is only meaningful after the G5 ships.

I don't know why are you people arguing?

These bencmarks are supposed to be measuring compute
performance and, as far as serious scientific
computations are concerned, the P4 can't even
match the G4.

G5 should be compared to the Itanium 2. There is
absolutely no point in this type of bickering.

In this game, the G5 beats the P4 hands DOWN. They
are not even in the same league. P4 does not even
have a real vector unit.

The fact that the G5 has a better MFLOPS/Mhz ratio
than the G4 is a very good thing. It suggests that
IBM has made a large investment and that the AltiVec
unit is not just an add on to the Power4 derived core
like some have previously suggested.

Bob (Math Ph.D. - What are your credentials?)

Why compare the G5 to the Itanium, you can't buy a desktop computer equiped with itaniun, it's for servers. G5 is a desktop processor. Compare the Power4 to the Itanium.

Itanium 2 is used in workstations.

G5 will be used in servers. PERIOD.

Mac OS X 10.3, aka Panther, will not be a 64-bit operating system, despite running on a 64-bit processor, the PowerPC 970 aka the G5.

Instead, the next major release of the Mac operating system will be a hybrid, much like version 10.2.7, codenamed 'Smeagol', which Apple has running on its pre-production Power Mac G5 machines and with which it will almost certainly ship production units.

Smeagol is a 32-bit operating system, though certain libraries and other elements have been recoded to allow applications - and the OS itself - to make use of the 64-bit addressing and datapaths, sources close to Apple said. For example, that's how the Power Mac G5 is able to support at least 8GB of memory, double the 32-bit limit of 4GB. Panther will adopt the same approach.

The PowerPC architecture was always defined as a 64-bit environment - 32-bit operation was defined as a sub-set of that environment. According to IBM, the PowerPC ISA contains an "optional" 64-bit bridge to "facilitate the migration of operating systems from 32-bit processor designs to 64-bit processors" and to "reduce the amount of software change" necessitated by that migration. "This bridge allows 64-bit implementations to retain certain aspects of the 32-bit architecture that otherwise are not supported, and in some cases not permitted, by the 64-bit architecture," says IBM PowerPC documentation seen by The Register. It's this bridge that enables the 970 to offer native 32-bit support and in turn saves Apple and its developers from having to rebuild their apps.

When it runs 32-bit code, the 970 reconfigures the table it uses to map its 64-bit effective address space onto the 42-bit address space it can actually access to cope with 32-bit addresses. Two new instructions, which are part of the bridge spec., allow the first 16 entries of the table to point to any segment of memory within the effective address space.

The upshot that a 32-bit app can gain access to a much larger address space than it by rights is able to, though it has to be optimised for the 970 first.

Indeed, that probably explains why Adobe is so readily able to provide G5 support in Photoshop with a plug-in which presumably replaces the app's main memory management code.

This approach is clearly the one Apple favours, and the inclusion of the "temporary" bridge technology in the 970 may well have been made at Apple's behest. This way Apple and its developers can take advantage of the 970's 64-bit addressing and datapaths without having to maintain a new codebase for the G5 alongside code intended to be run on older processors. "All applications written for 32-bit implementations will run without modification on 64-bit processors running in 32-bit mode," says IBM's documentation.

Apple's approach with Smeagol and Panther allows developers to continue working on 32-bit apps, but

access to 64-bit features when their code detects it's running on a G5. As preciously reported, coding for the G5 requires programmers adopt the GCC 3.3 compiler. GCC 3.3 supports flags which allow developers to generate 970-only code, and to access 64-bit datapaths.

As we say, IBM describes the bridge as temporary: "These resources are not to be considered a permanent part of the PowerPC architecture," it sternly warns.

How temporary 'temporary' turns out to be will depend on Apple's G5 adoption rate, and the necessity of providing a smooth migration path for developers. And, indeed, the extent to which the bridge delivers the benefits of 64-bit computing without the need to re-compile everything. Apple may well feel that for now it's gaining enough extra performance by stepping up to the 970 in 32-bit mode and that it doesn't feel that there's much more to gained by going to 64-bit mode. Particularly if the chip can do one mode or the other, and neither simultaneously.

It's a classic industry conundrum: balancing the need to pull users forward to gain the benefit of new or more advanced technology with the damage that can be done by forcing an unwilling migration.

The bottom line is that it may be some time before Mac OS X goes truly 64-bit, and if it happens anywhere in the product line, it's most likely to be within the Xserve server line. It is not yet known whether the Server version of Panther will be offered in a 64-bit form. ®

G5 cannot be compared to any processor that humans create! G5 should be compared to an alien techonolgy processor. There is absolutely no point in this type of bickering.

Also Bob, PhD too.

:)

Who are these wankers that are complaining that their PeeSee opens email faster?? These are not power users. Crimony, you can get email on your cell phone for dog's sake. The new G5 wafers are superb. Period.

Bob, GeD.

I've been the Mac zealot,posting the NASA G5 thing, all over the net, for more than a month. I'm glad to see someone actually read it,and turned it into a real discussion.

Big Bill: Trucker/Various