Apple Told to Halt ‘World’s Fastest’ Claims for G5

Eugenia Loli 2004-03-26 Apple 119 Comments

The Council of Better Business Bureaus has recommended (based on a tip by Dell) that Apple Computer discontinue comparative performance claims regarding its Power Mac G5 desktop.

About The Author

Eugenia Loli

Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.

Follow me on Twitter @EugeniaLoli

119 Comments

2004-03-27 9:27 pm

Anonymous
People complain to the BBB whey they’re getting hurt. It stands to reason that Dell is somehow getting hurt by Apple’s claim. I wonder if high end customers are making Dell prove they are the better performer, and it costs them money to do it. There’s got to be some legitimate business reason for Dell to complain to the BBB, because they are one of the last companies on earth who are concerned about customer interests.
2004-03-27 9:32 pm

Anonymous
“People complain to the BBB whey they’re getting hurt. It stands to reason that Dell is somehow getting hurt by Apple’s claim. I wonder if high end customers are making”

No, it doesn’t stand to reason. When Apple did their SPEC comparisons, they used a Dell machine. They misrepresented the performance of Dell’s products.

“Dell prove they are the better performer, and it costs”

Most “high-end” customers don’t buy Apple and wouldn’t consider them.

“them money to do it. There’s got to be some legitimate business reason for Dell to complain to the BBB, because they are one of the last companies on earth who are concerned about customer interests.”

I’ve always gotten good service from Dell.
2004-03-27 10:09 pm

Anonymous
Head on with your bad self. Because it would still just be a Windows(old fashioned DOS, with all the warts) machine. All 88,000 virusses, the worms, the really cool stuff that Macs just don’t have.

And if you really could do that, then Dell would be doing it already. Dell is whining and it is mainly because Jobs and Michael Dell just don’t seem to like each other.

A quick search of PriceGrabber comes up with this Opteron

http://www.sys.com/products/powerhouse/amd.cfm?id=SYSPHOD22002&proc…

When you check it, it costs more than a G5 and most tests say it isn’t faster. When you throw in some of the extras that it takes to match a stock G5, like DVD burning, then it isn’t even a contest. So you might be misinformed on this matter. Heck, the frontside bus isn’t even as fast as a dual G5.
2004-03-27 10:14 pm

Anonymous
Absolutely wrong! Assuming that bandwidth isn’t a limiting factor, clock speed helps integer and floating-point performance identically, that is, linearly with increased clock speed. If you extend the pipeline to increase the clockspeed, however, then the performance increase of the extra clockspeed is offset by the performance decrease of the extra pipeline bubbles. This decrease is proportional to the average number of missed branches in the code. Integer code has an extremely high branch density, often 25% or more. Floating-point code often has a branch density close to 10% or less. Ergo, there are many more missed branches in integer code, thus the performance impact of the longer pipeline is higher

Really, it seems to be a bad design descision by intel then. The have always climed that majority of the consumer code is integer and therefore high clockspeeds is better. Historically, Highend CPUs like itanium, PA-RISC, Ultrasparc, Alpha have had much better FP performance than pentiums or Athlons at lower clock speeds,

Clock speed: The 970’s high clock speed relative to the G4e will help significantly in giving it the edge in integer performance. How the 970 does relative to the P4 and the Opteron in this regard will depend on heavily on how aggressively IBM ramps up the chip’s GHz rating.

2004-03-27 10:37 pm

Anonymous
“Anyway you slice it Apple’s the marketing literature isn’t misleading no matter how much you want it to be.”

(actually one of the funniest comments I’ve ever read here.)

Let’s see… Apple made a claim.

They were called on it by the BBB.

Apple promised not to do that again…

Sounds like they were misleading to me…

Comparisons to Dell,(or anyone else) or to processor architecture, OS performance, heat output, or even price, are not a part of this story…

Apple was misleading people, they were called on it, and they said they wouldn’t do it again…

What is so hard about that?

2004-03-27 10:44 pm

Anonymous
Maybe you did not read I have a G5 at work. You seem to think the heat is an issue, it is not. As they move the desktops to the 90nm (like the xserve) heat will drop down.

BTW ALL G4 Powermacs have had HUGE heatsinks. I have several at work. All the power usage on them have been by far less then AMD or INTEL, yet they all are 2 to 3 times the size of the heat sinks on other machines.

Your arguemnt :look at the heat sinks” does not hold. And yes the dual machine has 9 fans. Very good, do you know how they work? they are NOT on all the time. and infact the dual Power supply noise problem was a defective parts which can be replaced.

as far as your pwer usage

the 970 still uses less power (1.8 @ 42 watts vs P4 2.8 @ 68 watts)

SO if you are going to scream heat look at INTEL and AMD

2004-03-27 11:20 pm

Anonymous
Really, it seems to be a bad design descision by intel then.

Why? The P4 has certainly served Intel very well, and the P4 has lead the Athlon in multimedia performance for most of its life cycle. Its a very fast CPU at a very decent cost.

The have always climed that majority of the consumer code is integer

The majority of consumer code is integer. However, nobody cares if a P4 runs Office faster than an Athlon does. The majority of the consumer code that is performance sensitive, however, (games, MP3 encoders, video players, etc) is floating-point.

Historically, Highend CPUs like itanium, PA-RISC, Ultrasparc, Alpha have had much better FP performance than pentiums or Athlons at lower clock speeds

Well, SPARC was never a very good performer, but you have a point about the other chips. However, you have to take account of the fact that high-end CPUs have tons of resources to throw at the problem. The Alpha 21264, released around 1998, had twice as many transistors (15M) compared to the PII’s (8M) of the time. It was an extremely beefy CPU, in some measures (eg: # of registers) beefier than the G5. The Itanium is the same way. The I2 has 221M transistors! So comparing such huge and expensive CPUs to consumer-level CPUs makes absolutely no sense. Also, it is ironic you mention the Alpha. Historically, one of the reasons the Alpha was so fast was that it was clocked nearly twice as fast as any of its competitors!

That’s it, All else is not equal in this case. RPM and clockspeed are more anologus than your example example of RPM = inflight instructions/ pipeline width.

Don’t equate inflight instructions with pipeline width. For example, the Opteron is a wide CPU, but can have 72 instructions in flight:

http://www.chip-architect.com/news/2003_09_21_Detailed_Architecture…

The P4 is a narrow CPU, and can have 126. If you use Apple’s logic, Intel could make a comparison between the Opteron and the P4 and say (to paraphrase Apple):

“meaning the Intel P4 can pump through more than 126 in-flight instructions at a time, a whopping 75% more than the 64-bit Opteron.”

Would you not agree that such a statement would be a misleading assessment of the relative performance of the P4 vs the Opteron?

PS> While, number of pipelines is a legitimate measure of CPU architecture, I’d argue its still not analagous to # of cylinders in a CPU. Output power reliably increases with number of cylinders, while that’s not the case for the number of pipelines in a CPU. You usually don’t have underutilized cylinders in a car engine, but you quite often have underutilized pipelines in a CPU due to the fact that a lot of code cannot be so easily parallelized. In the end, to be honest, you should cite horsepower numbers, which show how much power the engine was able to produce in a “benchmark” (a test hooked up to a dynajet chassis). Similarly, you should publish such benchmarks for CPUs too.

2004-03-27 11:54 pm

Anonymous
Don’t equate inflight instructions with pipeline width. For example, the Opteron is a wide CPU, but can have 72 instructions in flight:

http://www.chip-architect.com/news/2003_09_21_Detailed_Architecture…..

The P4 is a narrow CPU, and can have 126. If you use Apple’s logic, Intel could make a comparison between the Opteron and the P4 and say (to paraphrase Apple):

“meaning the Intel P4 can pump through more than 126 in-flight instructions at a time, a whopping 75% more than the 64-bit Opteron.”

To be more accurate the p4 126 inflight instructions are uops, each x86 instruction can be 1-4 uop. So the actual instructions in flight for a p4 when drawing a comparison to opteron could be 4 times less than 126 (31 x86 instructions).

from the URL you linked:

It can decode, execute and retire three x86-instructions per cycle. These instructions can be quite complex (CISC) operations involving multiple (>2) source operands. The Pentium 4 handles 3 so called uOps per cycle where multiple of these uOps may be needed to implement a single x86 instruction.

So the opteron an the g5 have a larger number of real instructions (100 ppc instructions(1 instruction = 2iops in the g5) in flight than a p4 and produce better perfromance per clock.

While, number of pipelines is a legitimate measure of CPU architecture, I’d argue its still not analagous to # of cylinders in a CPU. Output power reliably increases with number of cylinders, while that’s not the case for the number of pipelines in a CPU. You usually don’t have underutilized cylinders in a car engine, but you quite often have underutilized pipelines in a CPU due to the fact that a lot of code cannot be so easily parallelized. In the end, to be honest, you should cite horsepower numbers, which show how much power the engine was able to produce in a “benchmark” (a test hooked up to a dynajet chassis). Similarly, you should publish such benchmarks for CPUs too.

I wasn’t the one who came up with the care analogy you did, I merely made the anology fit better. I guess you are taking the word analogy to literally, Cpu pipelines are always, in college classes compared to factoy assembly lines. The same pitfall you just discribed to the car engine analogy applies to the assembly line too but it is means of comparison to understand and concept only.

2004-03-27 11:57 pm

Anonymous
wasn’t the one who came up with the care analogy you did, I merely made the anology fit better. I guess you are taking the word analogy to literally, Cpu pipelines are always, in college classes compared to factoy assembly lines. The same pitfall you just discribed to the car engine analogy applies to the assembly line too but it is means of comparison to understand and concept only.

need a spell checker

I wasn’t the one who came up with the car analogy you did, I merely made the anology fit better. I guess you are taking the word analogy too literally, Cpu pipelines are always, in college classes compared to factory assembly lines. The same pitfall you just described to the car engine analogy applies to the assembly line too. But analogies are a means of comparison to understand a concept only.

2004-03-28 12:07 am

Anonymous
You calimed there was no megahertz myth

The problem with the concept of “megahertz myth” is because proponents of the “megahertz myth” push the idea that “megahertz does not matter.” That’s as dumb an idea as the one that says megahertz is the only thing that matters.

and that intel didn’t design the P4 with clockspeed asone of the main goal

I never denied the fact that Intel designed the P4 with clockspeed as one of the main goals. I said that emphasizing clock-speed was not a marketing gimmick, but design decision. Let’s put this in mathematical terms:

P = Overall performance

W = Average # of instructions that can be executed at once

H = Clock speed in Hz

M = Probability of mispredicting a branch on a given cycle

C = Cost of mispredicting a branch in clock cycles

D = Branch density of code, or % of instructions that are branches

P = W * H – (C * M * H * D)

You see, increasing H increases P, but if you increase H by increasing the length of the pipeline, you also increase C. So if you’ve got a processor with clockspeed = H’ and cost of mispredicted branch = C’, the total performance P’ will depend on whether the increase in the left-hand term is eliminated by an increase in the right-hand term. If you run the numbers using values appropriate to the P4 and the P6, you’ll see why Intel found that the increase from clockspeed more than offset the hit caused by mispredicted branches, especially for floating-point code, where D is small.

2004-03-28 12:17 am

Anonymous
To be more accurate the p4 126 inflight instructions are uops, each x86 instruction can be 1-4 uop. So the actual instructions in flight for a p4 when drawing a comparison to opteron could be 4 times less than 126 (31 x86 instructions).

Since the Opteron’s pipelines operates on RiscOperations (ROPS), which are similar to the P4’s u-ops, the 72 number is for number of ROPS in-flight at a time, not x86 instructions. So the hypothetical commercial comparing the number of in-flight instructions in the P4 to Opteron would be valid, but completely misleading, just like Apple’s ad.

So the opteron an the g5 have a larger number of real instructions (100 ppc instructions(1 instruction = 2iops in the g5) in flight than a p4 and produce better perfromance per clock.

There is no doubt that the G5 and Opteron get better performance per-clock, but the number of operations in-flight at a time has nothing to do with it. Its just a number characteristic of the CPU’s design. Taken out of context, its a meaningless number.

2004-03-28 2:54 am

Anonymous
Okay… so Apple used independent firm (VeriTest… industry standard for testing) to run cross-platform tests that used equivalent compilers avalable for both platforms (GCC compilers are actually more pro-x86 than PPC…) instead of non-existent PCC optimized compilers and specially rigged Intel compilers that run special test libraries when it detects a SPEC test being compiled…

Ran the best and fastest op system for x86 (Red Hat Linux. Windows cripples system performance…), gave the Dell crap boxes MORE memory than the G5 (mere 1.5 GB for the G5, but a full 2 GB for the Dell)…

Failed to cripple the G5 by installing G4 specific memory settings that cripple the G5…

Set the MALLOC flags in FAVOR of the Dell (despite false claims to the contrary. this originated wityh a certain Adfam Hincklry… considered to be a completre whacko in developer circles…), disabled a because it CRIPPLES x86 performance (Dell ships their machines with it off by default. Guess why?)… and this is called “unfair?!”

Puh-LEEZE!

Okay, so Dell’s junk came in second best and at a higher price too!

let’s not mention the Wintel rags running rigged tests that pitted a G5 and a Dell running Adobe Premiere… conveniently failing to mention that…

1. Premiere has not been upgraded for the Mac in 3 years while the PC version is fully up to date

2. Premiere has NEVER been ported to M OS X and the rigged “tests” were running in “Classic” emulation on the Mac!

3. Adobe no longer publishes the Mac version!

Hmmm… anyone smell a rat here?

How much Dell money is floating about in the NAD back rooms?

2004-03-28 3:59 am

Anonymous
LOL- what utter bullshit.

Look at the empirical evidence regarding the heat and power consumed by the PPC970 and make up your own mind:

— Giant heat sinks

maybe overkill, but I would much rather have oversized heatsinks, as opposed to inadequate heatsinks. So I fail to see how this as a negative.

— 600W (and noisy) power supply for a system that only takes 2 low power drives

Noisy? You gotta be joking… your either very misinformed, or you have never really used a G5 before.

— 80W per CPU power system for 1.4Ghz CPU

Please; don’t shoot your mouth off without doing some basic research. To correct you: the G5 1.6Ghz chip dissipates 42 Watts.

— Lack of any ability to scale past 2Ghz over the past 9 months, even with 90nm chips

LOL. We all know Apple|IBM probably has the 970FX clocked well beyond 2Ghz, and in usual Apple fashion, they wont tell a living soul till its release. Thats quite obvious I would have thought…?

— G5 server runs only one (1) PPC970FX due to heat/power issues

Once again- Please check your facts. Im not even going to correct this statement…

— IBM running their blade servers at 1.6Ghz, slower than any Apple system

read above

— lack of Apple G5 laptop, G5 iMac, etc., all due to high power/heat.

Time will tell

sorry I just had to correct you on those points; they were outrageous

2004-03-28 4:06 am

Anonymous
I have to correct myself:

The G5 1.8Ghz chip dissipates 42 Watts.

http://www.xbitlabs.com/articles/cpu/display/powerpc-g5_5.html

2004-03-28 11:22 pm

Anonymous
The worst part for Dell is not that nobody cares or that the damage has already been made but that Apple’s claims are actually true.

2004-03-29 1:05 am

Anonymous
“The worst part for Dell is not that nobody cares or that the damage has already been made but that Apple’s claims are actually true.”

Apple’s claims weren’t true then – and aren’t true now.

2004-03-29 6:32 am

Anonymous
>Let’s see… Apple made a claim.

>They were called on it by the BBB.

>Apple promised not to do that again…

>Sounds like they were misleading to me…

This is not an accurate summary of what happened at all. Dell made a complaint to the BBB about the supporting evidence for claims Apple made in the original G5 campaign. The BBB did not say Apple’s claims were false, they merely said there was insufficient supporting evidence. A BIG DIFFERENCE.

As for “promised not to do it again”? Hardly. What Apple actually said was that the campaign had already finished and that they would keep the BBB’s concerns in mind when developing future campaigns. In other words, FUCK OFF!

Finally for the Dell defenders on the board: Are you aware that earlier this very month, Dell received a near identical reprimand from the BBB? Oh yes, for claims made in a server ad — that complaint was lodged by Sun Microsystems. Check out the NAD-BBB web site for particulars.

This story is really a tempest in a tea pot. If it weren’t such a slow news week and if stories involving Apple and allegations of wrong-doing weren’t so popular in certain circles, no one would have ever heard of it. After all, how many of you can say you were aware of the BBB’s “spanking” of Dell two weeks ago?

2004-03-29 6:47 am

Anonymous
>The “advertiser’s claim, ‘the world’s first 64-bit processor for personal >computers,’ could reasonably be interpreted to apply to workstations”.

Not only did they say that, but the BBB clowns had the further audicity to recommend that Apple make it more clear that this claim applies to personal computers and not workstations. Yep, fraid you actually read that right! Stating “personal computer” is not enough in their books. Apparently they want Apple to say “world’s first 64-bit processor for personal computers, except personal computers that are workstations“. Whatever other qualifications the BBB folks may have, it’s safe to say they’ll never find work as copy writers 🙂

2004-03-29 4:12 pm

Anonymous
BIG yawn. Surprise surpirse.

« Older Comments