The problem is marketing, plain and simple. Computer makers need a way to compare their products against others that is easy enough for Joe Artist, or Patricia Housewive to understand in about 1 minute. You have to remember that things today are not always sold after the buyer has done extensive research, they are sold when a person decided they need a computer and wander into the local store and browse the isles. The comparison’s made using the info found on the little cards with the price printed on it. They look at price and the list of one line feature descriptions. One of those lines currently says “XXXMHZ BrandProcessor”. THAT’S what AMD and Apple is looking to replace with something that make their products look more appealing, not some in-depth examination of the comparison between RISC and CISC. Most folks couldn’t give a rat’s fart about that stuff, they just want what they perceive as the most bang for their hard earned buck while taking as little time as possible in making that decision. Hence Intel’s MHZ ratings.

To that end, the most value I can see is a tranlation of some benchmark that can be used on any CPU to give a reasonable estimate of realized speed. I now think MHZ and gigaflops are missing the boat entirely. What is needed is an application that does some standard stuff, move files around, fill in color, create a 3d image, whatever, add up the time is take to run through all of that, and use that as a rating. Give it a catchy name like ‘User Speed’, or ‘App-Spec’. (I like using Setiathome as a comparison for this reason. Yes, I know it has problems too, but its closer to what people really do with their computers.)

It’s not perfect but it fills that space next to the price with a number that has a bit more meaning to the consumer.

At least, that’s what I think.

>But you do understand it, right?
————————————————————————————————————-

Could you be more condescending?  Honestly…

————————————————————————————————————-
Quote
>I certainly agree with you that a Mac is a much more thoroughly >considered “Personal Computer” than any box running Windows, but trying >to claima performance lead for the Mac platform is completely without >foundation. 
————————————————————————————————————-

You don’t need to claim a lead.  Why not just try to inform the public that MHz are not king and that a 500Mhz G4 can at the very least keep up with a P4 running at twice the MHz rating. I’m running a 400MHz G4 at home and I’m just as happy with its performance as with my 1 Gig P3 at work.  Sure its anecdotal, but I don’t think that otherwise uninformed members of the general public would consider the G4’s performance to be bad at all.

I think that somehow breaking up the myth of MHz completely determining the speed of systems would be a good thing.  Floating point operations per second (FlOPS) run into similar problems in that they measure only a subset of what goes into making a system “fast”. I do, however, think they are at least a bit better; at least they measure, to some degree, work actually accomplished and not how fast the clock on the processor ticks.

But then, what about integer operations per second? Hard disk speed?  Ability to multitask effectively via large amounts of fast RAM? Even efficiency and design of the OS or Window Manager or file manager(Windows vs. MacOS, Finder vs. Explorer)?  The list of other factors is large, making a good “system rating” a hard thing to put your finger on.

To sum up: not everything is black and white. Lets all try to remember that.

Matthew

OK, I’ll admit my last post was somewhat peevish, but the fact remains that SPEC exists NOW, and has done for at least a decade. If any of the above mentioned microprocessor or system designers were in the least bit interested in a fair and objective way for the consumer to compare the performance of their systems, they would embrace SPEc and work to make their benchmarks ever more relavent. That they don’t (and Apple/Motorola are the arch criminals here) shows exactly what they think of the idea. Apple witter on about GigaFLOPS and Supercomputers, Intel pluck mp3 encoding statistics out of the air or feed us some nonsense about speeding up the internet, and AMD are actually proposing to pretend their processors are faster than they are by simply omitting verifiable SI units. Unbeleivable. And my comment about GFLOPS wasn’t without reason - i have investigated Apple’s claims for MY systems GFLOPS ratings, and I couldn’t get close. And what does a GFLOP tell you about performance anyway? i’d venture that Mhz was a far better indication of processor performance than GFLOPS. A long time ago i had a little Colour Classic Mac, to which I added a Motorola 68882 FPU so that i could run Quark Xpress on it. This had the effect of boosting it’s FPU benchmarks (as reported by Norton system info) well beyond those of my friends Performa 630, which the ancients among us will recall was powered by the none-too-shabby MC 68LC040. The fact that my little CC could knock the performa dead in the Sine and Cosine benchmarks had absolutely no impact on the fact that his machine performed all other tasks about 10 times faster.

I am neither coder nor chip-architect, but it’s as well to try and realise what one DOESN’T know about these things - those that DO know are engineers working at the cutting edge of technology whether they be Intel, Moto, IBM or AMD employees. If we want transparnecy in performance evaluation, we’re going to need some kind of legislation to force disclosure of an officially sanctioned measure. they do it with cars, why not computers?

I have to agree that there needs to be some sort of universal standard for comparing different processor architectures. Spec has it’s flaws, and the results can be skewed. Since it is a software test, other variables such as which compiler is used, and disk/memory performance can have profound effects the outcome. The only advantage is that it is freely available and therefore subject to some sort of peer-review.

Also, as others have pointed out, a balanced system design can lead to more efficient overall operation; regardless of individual component performance, especially the processor.

So a consumer needs some way of comparing processors based on something other than raw clock cycles per second.

AMD has proposed a number based on the theoretical maximum number of instructions per clock times the clockspeed. This comes out to something similar to MIPS, but not quite. It is not limited by software or peripheral components, and it is a number based strictly on the theoretical maximum performance of the processor itself. In the case of an 867 G4, this number would be something like 6069 (based on 7 IPC, I don’t know if that’s right or not). The 2GHz P4 would have a number like 8000 (again, assuming 4 IPC though I don’t know if that is right either).

I don’t know the exact numbers, and I don’t want to take off on a tangent to find specifics. I just mean to illustrate that the G4s theoretical IPC is higher than the P4, and that the resulting number from these calculations shows a better representation of the PROPORTION of real world performance difference.

I think this rating system, theoretical and ultimately meaningless as it is, could be a win-win-win situation for everyone. Marketing gets to trumpet HUGE numbers, engineering is encouraged to produce more efficient processors, and consumers have a number that, at the very least, tells them the theoretical maximum performance available from that processor. At least in terms of how much info it can potentially shove around at a time. While it doesn’t mean anything for software and total system performance, that’s not what consumers are looking at anyway: they don’t understand, and can’t be bothered to be taught.

AND, just to make sure consumers slowly start to grasp the concept, a disclaimer should be added that says something like “This number represents the theoretical maximum performance of this processor. Actual performance will depend on the type and quality of software that is running, and the performance of other related subsystems that interact with the processor.”

Who knows… it could happen.

(PS: I just wanted to note that, using this rating, a 1.2 Ghz G4 would have a higher number than a 2Ghz P4 which sounds about right to me. Maybe there’s something in it after all?)

(PS: I just wanted to note that, using this rating, a 1.2 Ghz G4 would have a higher number than a 2Ghz P4 which sounds about right to me. Maybe there’s something in it after all?)

hmmmm…
you can BUY a 2Ghz P4 for about the price difference Apple will charge you for the difference between a 400 and 500Mhz G4. Can’t buy a 1.2ghz G4 though…

OK, last post (honest :-])

My favourite computer parts supplier advertises the 1.33 Ghz AMD Athlon thus:

“The AMD Athlon Processor provides the ultimate performance for Cutting Edge Applications. It has the abiltiy to support PC2100 DDR memory technology and it has 1.33GHz of power. It boasts a 384K total full-speed on-chip system cache including 128K L1 cache, and 256K on-chip full speed L2 cache. It also includes a superscalar floating point engine for x86 platforms.

The AMD Athlon Processor is among the world’s most powerful engines for PC computing, and represents the industry’s first 7th generation x86 microarchitecture. The AMD Athlon processor family is designed to power the next generation in x86 computing platforms. It meets the computation intensive requirements of cutting edge software applications running on high-end desktop systems, workstations and servers.”

WOW! That’s pretty confused! 1.33GHz of POWER, no less! But, generally, it’s fairly spot-on - the salient performance features are addressed and you’re left in no doubt over which market space it’s aimed at.

This chip costs Ј101 retail in the UK. The whole chip, that is - not some small percentage upgrade from a lesser one.

If there were a standardised system of performance devised, the immediately obvious thing to do would be to calculate benchmarks/pound (dollar, euro etc). How would the G4 look then? if this Athlon scored maybe 30BMs, and the G4 867 came in at 25BMs, we’d be pretty impressed at the greater number of BMs per cycle, yes? If we then found that the G4 cost Ј500 and the Athlon Ј100, what would be the obvious conclusion? That’s right, goodbye G4.

And THAT’S why Apple will NEVER submit to such a scheme.

Here’s a silly idea, but I kind of like it.

Rather than trying to hide megahertz or convince people that megahertz doesn’t matter, come up with a measure that contains the word “megahertz”, based upon the latest Pentium Chip from Intel.

For example, let’s say that Apple’s benchmark claims that the 867MHz PowerPC is 40% faster than the 1.7GHz Pentium 4.  Thus, the PowerPC is the equivalent of a 2.4GHz Pentium 4 (Well, 2.38GHz, but we’ll round up).

Thus, we are introducing the new Power Macintosh G4/2400 with an amazing speed of 2.4 VGHz*.  What does the ‘V’ stand for?  Why, “Virtual” of course!  Sitting at the bottom of the advertising page/screen is the footnote:  “* Virtual Gigahertz computed using ByteMark Benchmark against a 1.7GHz Pentium 4.  Actual processor speed is 867MHz.  For more information, visit http://www.apple.com/vghz.”

It’s no more misleading than the 17” Monitor (15.9” Viewable).  It gives people a speed rating they understand.  If they want the details, they can visit the web page or contact their friendly-neighborhood Apple salesperson.

Problem solved.  Pass the beer nuts.

...Continued from above.  Sorry, I couldn’t edit it.

In the example, I mention the SPEC benchmark.  I use that as an example—it could be any benchmark.  Apple should obviously use an outside benchmark.  Heck, go back to ByteMarks—Apple got lots of good mileage out of ByteMarks.

And for those of you who complain that ByteMark is a lousy benchmark, so what?  Megahertz is a lousy benchmark.  So’s Gigaflops.  So’s Quake frame rates.  So’s SETI scores.  It’s all static anyway.

The important thing is the conversion back to “Virtual Megahertz.”

Hiding the MHz rating is not really the answer.  Quoting Gigaflops is a better measure although not perfect.  Consider the following analogy:

Which is faster a 2.0 litre car or a 1.4 litre car.  Well it depends on the power of the engine.  Say the 2.0 litre car produces 180 bhp and the 1.4 litre car produces 160 bhp.  Now things look a bit closer.  Consider further that the 2.0 litre car is a family car weighing 2.0 tonnes and that the 1.4 litre car is a small car weighing 1.0 tonne. In sustained motorway driving both perform as well as each other.  For stop/start driving the 1.4 litre car will accelerate quicker after a stop and on average be quicker.  Also note that the 2.0 litre car consumes 60 litres of fuel a day, whereas the 1.4 litre car consumes only 40 litres of fuel a day.

Now re-phrase this to compare two personal computers:

Which is faster a 2.0GHz machine or a 1.4GHz machine.  Well it depends on the power of the CPU.  Say the 2.0GHz machine performs at 4.5 Gigaflops and the 1.4GHz machine performs at 4.0 Gigaflops.  Now things look a bit closer. Consider further that the 2.0GHz machine has a long 20 stage execution pipeline and that the 1.4GHz machine has a shorter 10 stage execution pipeline.  In sustained usage both perform as well as each other.  For stop/start processing the 1.4GHz machine will accelerate quicker after a pipeline stall and on average be quicker. Also note that the 2.0GHz machines consumes 60W of power, whereas the 1.4GHz machine consumes only 40W of power.

Note that the two machines could be an Intel 2.0GHz Pentium 4 machine and an AMD 1.4GHz Athlon.  On this basis a Motorola 867MHz PowerPC 7450 performing at 5.0 Gigaflops, with a 7 stage execution pipeline and consuming 20W of power, would be a 867cc car producing 200 bhp and is a 700kg sports car consuming a minuscule 20 litres of fuel a day.

As for the speed of these machines, that depends on other factors.  For the cars, the coefficient of drag, gear box ratios, ability of driver and the type of road (straight, winding, hilly) all have a bearing.  For the computers, the bus speeds, amount of memory, efficiency of code and type of application (text processing, number crunching) affect the performance.  For instance text processing does not need the power, which is like driving down a steep hill in a car, whereas Photoshop filters require power, which is like driving up a steep hill.

To sum up,  consider a computer’s clock rating (MHz) like a car’s engine capacity (litres) and the computer’s power (Gigaflops) rating like a car’s engine power (bhp).

With reference to a previous mention of using the SPEC benchmarks, Motorola actually quote SPECint95 and SPECfp95 figures for the PowerPC 750 and 7400.  The figures are similar for G3 and G4 processors of the same clock rate as these benchmarks do not take Altivec performance into account.

This means that we cannot even compare different PowerPCs with each other.  If SPEC was used as the measure of performance when purchasing a computer, nobody would buy a G4 machine costing twice as much as a “similar performance” G3 machine.

Altivec (or Velocity Engine in Apple parlance) obviously makes a big difference not visible in the SPEC figures.  A similar problem arises using the Drystone benchmark figures.  Note that an IBM 750CXe (G3) clocked at 500MHz yields 1160 Drystone MIPS , whereas a Motorola 7400 (G4) clocked at 500MHz yields 917 Drystone MIPS.

Altivec (or Velocity Engine in Apple parlance) obviously makes a big difference not visible in the SPEC figures.  A similar problem arises using the Drystone benchmark figures.  Note that an IBM 750CXe (G3) clocked at 500MHz yields 1160 Drystone MIPS , whereas a Motorola 7400 (G4) clocked at 500MHz yields 917 Drystone MIPS.

and here we can see the REAL problem; that the G4 was adopted before it was ready. The G3 chip was having no trouble outperforming Intel’s Pentium III until Intel started to scale their clock speed in response to AMD’s Athlon. If Apple/IBM/Moto had stayed together and worked to scale the speed of the G3, Apple could have maintained their performance lead. that they didn’t was a testament to the consortium’s lack of vision, and landed Apple in the unenviable postion of moving to a slower architecture exclusively. Those SPEC don’t lie, they just prove the everyday experience of anyone who ever moved from a 450Mhz G3 to a 450Mhz G4.

It’s worth remembering that using a G4 DOES NOT make a “big difference” over a same speed G3 for 90% of applications.

i dont give a rats ass how they rate them… im stuck making a party cd for my friends kids birthday party favors and im processing nsynk and janet jackson this job cant GO FAST ENOUGH…. i want a g5 RIGHT NOW AAAAAHHHHHHHH>>>>>>