G4/G3 Equivalants To PC Proccessors

[Cubeboy]

Originally posted by patrick0brien
-Cubeboy

I just rememebred something else. You're probably not going to like it as it's not supported by Spec tests but if you may recall the Sawtooth G4 400mhz when introduced in 1999 was cedited as the "First Mainstream Supercomputer" because according to government (NSA I think) tests the computer produced a real-world performance of >1GFLOPs sustained and spiked to >4GFLOPs a second, thus Apple had a great marketing angle, but couldn't export the machines until they petitioned the goverment to up the bar of export. So if we do some dirty math on that we'd get 400/~1024=~2.56FLOPs/Cycle and 400/~4096=~10.24FLOPs/Cycle

I know it's not elegant, I know it's not Spec, but these were the legal and certified findings of the time.

Ah so your refering to the GFLOPS used by Apple's Marketing department, unfortunately, I assure you, those scores have little value if any at all. To measure GFLOPS (billion Floating Point Operations Per Second), you divide the number of floating point operations in your program by the execution time in milliseconds. The problem with this however is that the number of floating point operation vary depending on the program being benched. Depending on the programs, the FLOPS rating can vary anywhere between a few hundred MFLOPS to several 1000 MFLOPS. Looking at their website Apple never provided the information of whether what program was used or any comparison to similarly benched Pentium 4 or Athlon systems, thus rendering their score essentially useless. Linpack, the sole entree into the Top 500 supercomputers list is one of the benchmarks that actually measures flops in a credible way, below I've listed the results for the G4 533, Athlon 600, and the first Pentium 4. I would put the current results but unfortunately, the Performance Database Server isn't updated very often so I've listed comparable processors to the 533 Mhz G4. Of course, back than, the G4 was far closer performance-wise to the Athlon and Pentium 4.

Official Linpack Results

Motorola G4 533
N=100: 231 Mflops
N=1000: 478 Mflops
Peak: 1066 Mflops

Athlon 600
N=100: 260 Mflops
N=1000: 557 Mflops
Peak: 1200 Mflops

Pentium 4@1.5 ghz
N=100: 326 Mflops
N=1000: 1311 Mflops
Peak: 3000 Mflops

 

[patrick0brien]

-Cubeboy

Not to sound pissy or argumentative, but you appeared to miss my point of "Sawtooth G4 400mhz when introduced in 1999 was credited as the "First Mainstream Supercomputer" because according to government (NSA I think) tests the computer produced a real-world performance of >1GFLOPs sustained and spiked to >4GFLOPs a second."

This is fact, and history.

I do listen to Spec2000 test and others, but I've learned to take them with a grain or salt as (even as you mentioned) the compilers differ, the tests differ. Will all of these differences on different machines. Not to mention what the different OS's those different tests, on those different machines are running and how they affect the outcome of measurement.

It's all too damn muddy, and ultimately not very useful.

No wonder there is no unified measure of performance, and ergo the need to read into these tests some sort of value.

Sadly, this pinnacle of computer performance measurement is quite a way off. Until then, arguments such as this one will continue to happen, with numbers being used to bludgeon each other with still no solution at the end.

This is why I find even the "vaunted" Photoshop tests, and render test more valuble. Because that is the stuff that users need - and users, are what computers are for.

Believe what you will. And I shall as well.

 

[Cubeboy]

Originally posted by patrick0brien
-Cubeboy

The project was kicked off to try to answer client's inquiries as to why the P4 performed less work than the P3 at comparable clock rates. The Motorola performance was thrown in as a bonus to that measurement.

Clock to Clock, the Pentium 3 is around 30% faster at general usage, the Pentium 4 is over 70% faster at SSE. Of course the Pentium 4 has also substantially increased it's IPC since then, at least 25% according to microprocessor report. As I've said before, Real world performance is not dictated by the CPU alone, mostly it's dictated by the design of the program being benched.

 

[Cubeboy]

Originally posted by patrick0brien
-Cubeboy
As for Spec2000 measurements, I'm highly suspect of them as they measure aspects don't directly impact the user, and that was the point of the project I'm referring to above. The Spec2000 test may be accurate to the module and the numbers stated, but we couldn't use them because that actually hurt our conclusions by distracting the clients obsevation that the P3 was in fact faster in real world performance than the P4. It was counterinformative. While true in their own right, the Spec2000 numbers were not indicative of overall performance.B]

No Spec was not meant to reflect real world performance I don't think any benchmark or program can do that considering the vast amount variation in programs today. Spec was designed to compare different architectures and estimate the potential of a particuler architecture. This is what sets spec apart from the rest of the benchmarks for you see spec cannot be hand optimized, it's source code consists of large data blocks, not little snippets. It's a composite of hand picked benchmarks meant to test every aspect of the cpu, the scores are than weighted transformed into a final score to show the processors performance at different kinds of code (integer code and floating point code). This is why SPEC is so respected by much the industry (IBM, Intel, AMD, Alpha all endorse it as the standard) The only weakness I see in spec is that you can't have a accurate measure of performance unless you have a good compiler which for some reason right now, Powermacs don't have one, however looking at the variation in scores, it can't justify that the G4 simply can't compete in the current market (which would explain why current Powermacs use dual G4s).

 

[Cubeboy]

Originally posted by patrick0brien
-Cubeboy

Not to sound pissy or argumentative, but you appeared to miss my point of "Sawtooth G4 400mhz when introduced in 1999 was credited as the "First Mainstream Supercomputer" because according to government (NSA I think) tests the computer produced a real-world performance of >1GFLOPs sustained and spiked to >4GFLOPs a second."

This is fact, and history.

I do listen to Spec2000 test and others, but I've learned to take them with a grain or salt as (even as you mentioned) the compilers differ, the tests differ. Will all of these differences on different machines. Not to mention what the different OS's those different tests, on those different machines are running and how they affect the outcome of measurement.

It's all too damn muddy, and ultimately not very useful.

No wonder there is no unified measure of performance, and ergo the need to read into these tests some sort of value.

Sadly, this pinnacle of computer performance measurement is quite a way off. Until then, arguments such as this one will continue to happen, with numbers being used to bludgeon each other with still no solution at the end.

This is why I find even the "vaunted" Photoshop tests, and render test more valuble. Because that is the stuff that users need - and users, are what computers are for.

Believe what you will. And I shall as well.

No Patrick, I afraid your the one who's missed the point, as I've stated in my previous post gigaflops has no value unless we know the exact specifications of the program used. Unless we have a set program that all processors will be benched on (which is exactly what Linpack is) or the exact specifications of the program, we cannot determine the value of the gigaflop. It's like comparing a MDD running FCP4 to a MDD running Aftereffects, depending on the program, the scores can be vastly different. With the right program, I can easily have a Pentium III or Athlon with sustained performance over 1 Gigaflop. As can be seen from the Linpack benchmark, under equal conditions, the Pentium 4 and Athlon clearly surpass the G4 in terms of Flops.

 

[Ensoniq]

No one will ever win the "which processor is really the fastest" argument. It's as unprovable as the existence of God or life after death. People have their beliefs, and that's that. Even though it's theoretically based on specs and figures that seemingly can't lie, it's all completely subjective.

That said, I sat down one afternoon and did some research to compile a list of SPEC scores for a variety of processors. You can either choose to believe my numbers or not, but the link to them is here:

Spec Scores

The table doesn't list every possible processor/speed combination. But it gives a good range of what's currently available. And of course, the numbers come under immediate criticism depending on whether or not you actually believe SPEC is a fair and valid cross-platform benchmark...maybe it's not.

But based solely on SPEC scores, here are some interesting points that my personal analysis of the numbers indicate:

1 - Clock per clock, the P4 averages 17% faster than a G4 in integer performance. Considering the P4 is 1 year old, and the G4 is 3 years old, that's a far smaller margin than you'd expect.

2 - Clock per clock, the P4 is 100% faster than a G4 at floating point operations. Definite winner there.

3 - Clock per clock, the PPC 970 will be at least 70% faster than the G4 in integer operations, and will more than TRIPLE the speed of floating point operations.

4 - Clock per clock, the Opteron could beat the PPC 970 by 15% at integer operations, but is identical in floating point operations.

What does this mean? Probably nothing at all in real world terms. It's a bunch of numbers that don't even remotely factor in the highly optimized OS X vs. the clunky Windows. It doesn't take into account vector processing on either platform. You could spend hours looking at the numbers (I did) and calculate any formula you like to make one platform look superior to the other. It's all in your point of view.

But if anything, it shows two big things:

1 - The G4 is not as "slow" as some claim. If not stuck at such low speeds from Motorola, it would fare much better against the Wintel competition.

2 - The PPC 970 will be as capable as any chip that currently exists, and those that will come to market over the next year. Given equal clock speeds, we'll have nothing to be overly concerned with on the Wintel side. And that is worth smiling about.

There will always be those who will say "my 3 GHz P4 beats your Dual 1.4 GHz G4". And if those few extra seconds saved really matter that much to them, nothing you can say will change their minds.

 

[Cubeboy]

Originally posted by Ensoniq
No one will ever win the "which processor is really the fastest" argument. It's as unprovable as the existence of God or life after death. People have their beliefs, and that's that. Even though it's theoretically based on specs and figures that seemingly can't lie, it's all completely subjective.

That said, I sat down one afternoon and did some research to compile a list of SPEC scores for a variety of processors. You can either choose to believe my numbers or not, but the link to them is here:

Spec Scores

The table doesn't list every possible processor/speed combination. But it gives a good range of what's currently available. And of course, the numbers come under immediate criticism depending on whether or not you actually believe SPEC is a fair and valid cross-platform benchmark...maybe it's not.

But based solely on SPEC scores, here are some interesting points that my personal analysis of the numbers indicate:

1 - Clock per clock, the P4 averages 17% faster than a G4 in integer performance. Considering the P4 is 1 year old, and the G4 is 3 years old, that's a far smaller margin than you'd expect.

2 - Clock per clock, the P4 is 100% faster than a G4 at floating point operations. Definite winner there.

3 - Clock per clock, the PPC 970 will be at least 70% faster than the G4 in integer operations, and will more than TRIPLE the speed of floating point operations.

4 - Clock per clock, the Opteron could beat the PPC 970 by 15% at integer operations, but is identical in floating point operations.

What does this mean? Probably nothing at all in real world terms. It's a bunch of numbers that don't even remotely factor in the highly optimized OS X vs. the clunky Windows. It doesn't take into account vector processing on either platform. You could spend hours looking at the numbers (I did) and calculate any formula you like to make one platform look superior to the other. It's all in your point of view.

But if anything, it shows two big things:

1 - The G4 is not as "slow" as some claim. If not stuck at such low speeds from Motorola, it would fare much better against the Wintel competition.

2 - The PPC 970 will be as capable as any chip that currently exists, and those that will come to market over the next year. Given equal clock speeds, we'll have nothing to be overly concerned with on the Wintel side. And that is worth smiling about.

There will always be those who will say "my 3 GHz P4 beats your Dual 1.4 GHz G4". And if those few extra seconds saved really matter that much to them, nothing you can say will change their minds.

A very well done chart, I have only a few quarrels. Spec is designed to test the strength of a particular architecture, having Multiple processors, SMT, or CMP, will not help the spec score (multithreading is viewed as optimization which isn't allowed). Your comparing the score from the Pentium 4 to the percieved score dual processor G4 (which would really have the same score as the single G4) while comparing the IBM's estimate for the PPC970 to the single processor G4, needless to say, this hardly accounts for a very good comparison.

As I've mentioned before, Spec wasn't designed to test real world performance, programs can be optimized, environments can be changed, a million factors can affect real world performance. Spec was made solely to measure the performance of a processor and compare it to processors of other architectures, to provide a universal measure of absolute performance of a CPU.

Right now, Apple needs to put two G4s into a single machine just to keep competitive with a single Pentum 4 or Athlon (which it's not doing a very good job at). That alone should give us a good insight as to the state of how well a single G4 stacks up. G4 architecture is simply too old to compete well with the new designs we see today.

I have no doubts that the PPC970s will offer competitive performance to Pentium 5 and Opteron, with Motorola (who hasn't come out with a new cpu in three years) off our backs I don't think we'll have much trouble with performance. Most of the professionals believe that the PPC970 will be very competitive with the next Pentium 4 (Pentium 5) in Floating Point and Vector, but lag behind in integer code. This goes along rather well with the specmarks you've posted.

 

[g30ffr3y]

guys... this is a circular argument... we all choose to use macs becuase we love them... sure we'd all like faster machines... we'd all like to tout our fast processors to our PC molested friends, but its just not in our cards right now... you have to ask yourself if your mac does what you need it to do... if it does than its the great machine that you were willing to pay a premium for... if not... well the 970's are on their way... personally i dont give a damn if there is a P4 out there with twice the power of my quicksilver or my tibook... i love my mac... i look forward to loving more and more macs in the future... to hell with m$... i dont give a **** if they come out with a P10 at 420ghz...

just please put doomIII on the mac...

 

[Cubeboy]

Exactly, buy what you need, I've got the original Cube, 450 mhz G4, 128 MB ram, 20 gig hard drive, and a 16 MB ATI Rage 128 pro video card, it's nearly three years old and it still runs all my apps on OS 9.2 just fine. For the most part, speed has little value in todays software market, we have reached a point in time that the computer has to wait on the person instead of vice versa. Any extra speed is essentially useless in these conditions. When your going with Apple, you get all the speed you need with style, elegance, quality, reliability, and the best OS in the industry, thats something your not going to get with a Dell or Gateway or whatever other PC company their is.

 

[g30ffr3y]

im glad that benchmarks arent the bible of computing to all... osx is a pleasure to work in... i often regret even changing the wallpaper... though i do often... everything is just so beautifully designed... you dont get that with windoze, dell, gateway, etc...

 

[maraczc]

Just in case anyone cared on TheMacMind website one contributor places the 800 MHZ G3 as a bit faster than an 800 MHZ P3.

 

[praetorian_x]

Originally posted by Ensoniq

What does this mean? Probably nothing at all in real world terms. It's a bunch of numbers that don't even remotely factor in the highly optimized OS X vs. the clunky Windows. It doesn't take into account vector processing on either platform. You could spend hours looking at the numbers (I did) and calculate any formula you like to make one platform look superior to the other. It's all in your point of view.

...

There will always be those who will say "my 3 GHz P4 beats your Dual 1.4 GHz G4". And if those few extra seconds saved really matter that much to them, nothing you can say will change their minds.

Optimized OSX vs clunky windows? Have you tried to resize a window in OSX lately? I have yet to see *any* machine (including the dual 1.4's) that can resize an iPhoto window smoothly. The windows window manager is one giant hack for performance. Ugly as sin, but fast as hell.

OSX sacrafices speed for flexibility and appearance. NIBs/Services/Plugins/AlphaBlending/etc. are all cool as hell, but they sure aren't optimized for performance...

Regarding the "few extra seconds", I can see how, for professionals, it matters. That's why Pixar went with a cluster of dual xeons to render "Finding Nemo". Your average user will not likely notice much of a difference between x86 and ppc, given one crucial point:

The OS running on top of it is the same.

But this is not the case. OSX is much more resource intensive than Linux/Windows, and is much younger and hence less optimized. Window resizing is one of the primary stumbling blocks in OSX, and its something that Quartz does not, and cannot address. Another area where OSX falls down is program loading, especially in Cocoa based apps with NIBs. Launch Terminal.app on a top end mac, and then go run xterm on a Linux box from three years ago. The difference is startling.

Because of this, the *percieved* speed of the macintosh platform is much lower, although in raw number-crunching terms, ppc isn't quite as badly off as some make out.

Anyway, not trying to convert anyone, just pointing out my thoughts.

Cheers,
prat

 

[donigian]

Re: G4/G3 Equivalants To PC Proccessors

Originally posted by maraczc
Intel Centario

Sorry, it's Centrino and a certification not an actual processor. The Centrino certification requires the use of the Intel chipset, the Intel Pentium-M processor, and an Intel WiFi radio. You want a comparison to the Intel Pentium-M processor.

And, topicolo, the AthlonXP 3000+ runs a 333 MHz FSB, only the 3200+ runs a 400 MHz FSB.

Ahh, but those are little points of clarification.