New barefeats article proves G4s sometimes faster, sometimes slower than AMD/Pentiums

[javajedi]

http://members.ij.net/javajedi

You're more than welcome to download the Java version, or the Mac OS X native one. When I said C, I really should clarify. It's actually a Cocoa version so the source is a .m objective c file, however the math function itself is from the C library. It's really cool how in objective c you can use regular C


For integer testing:

int x1,x2,x3;
for (x1=1; x1<=20000; x1++) {
for(x2=1; x2<=20000; x2++) {
x3 = x1*x2;
}
}

if you don't want to mess with the cocoa thing, just use that snippet and use time.h to time it.

 

[javajedi]

Originally posted by Backtothemac


And I care why? It doesn't matter how fast you can surf on your PC. I can get around fast enough on my Mac. People who say Mac's are too slow are the same people that never take the time to watch a sunset or spend a day with their kid.

They are fast enough. They do what they are supposed to do the way they are supposed to do it.

The don't crash, don't get viruses, and don't look like something from the 1980s!

More fallacies...

#1 My PC doesn't crash
#2 It does not get infected with virii
#3 It doesn't look like something from the 1980s


You take a look for yourself
http://homepage.mac.com/kevindecker/PhotoAlbum3.html


Oh and one more thing Back2TheMac: I've noticed now you are signing quite a different tune, before the G4 was supreme... now.. it is slower and... uhh.. doesn't matter?? How convenient.

 

[alex_ant]

Originally posted by Backtothemac
And I care why? It doesn't matter how fast you can surf on your PC. I can get around fast enough on my Mac. People who say Mac's are too slow are the same people that never take the time to watch a sunset or spend a day with their kid.

Or perhaps the people who say Macs are too slow are the ones who would like more time to watch a sunset or spend a day with their kid?

 

[alex_ant]

Javajedi, what you've done with your benchmarking is very helpful and I believe provides much insight. I too was surprised to see that the PowerPC performed as poorly as it did. Sorry if I missed you addressing this, but did you use GCC 3.x on the PPC?

There are a few conclusions I could draw from this performance data:

1) AltiVec acceleration is crucial to attain performance competitive with x86.
2) In the best case, AltiVec-accelerated code will perform several times faster than optimized x86 code. However, the best case is very rare and limited to specialized tasks like BLAST, RC5, SETI, certain Photoshop filters, and so on.
3) In the worst case, AltiVec-optimized code will perform barely any better or perhaps even worse than non-optimized code.
4) The G4's integer and floating-point units are extremely weak.
4a) Even MHz-for-MHz, they appear to be slower than those of the Pentium 4.
4b) The 750FX's integer unit is stronger than the Pentium 4's clock-for-clock, but considering the Pentium 4 is clocked 4x higher at the moment, it does about 4x better overall.
5) The c't SPEC benchmarks from a while back (the only source of G4 SPEC results I'm aware of) weren't that far off.

I'm disappointed but not surprised to see that gopher has split from the thread. Oh well, I'm sure he'll reappear at a later date oblivious to everything that has just been presented in this thread.

Alex

 

[javajedi]

Originally posted by javajedi
What you are saying makes a lot of sense. Now that I think about, I too recall reading this somewhere.

Now that we know the real truth about the "better standard FPU", I thought it was time to shed some light on non vectorized G4 integer processing.

It still does 200,000,000 calculations, but this time I'm multiplying ints.

Motorola 7455 G4@800Mhz: 9 seconds (Native)
IBM 750FX G3@700Mhz: 7 seconds (Native)
Intel P4@2600Mhz 2 seconds (Java)

PowerPC 7455 integer processing is consierabley better than floating point (obviously less work doing ints), but still less per cycle than the Pentium 4.

Very intresting the G4 looses both floating point and integer to the IBM chip, at a 100MHz clock disadvantage.

I'm still waiting to see that "better standard FPU" in the G4. It seems the G4 is absolutely useless unless you are fortunate to have vectorized (AltiVec) code.

Alex, yeah, the native version was compiled under 3.1. It really is interesting to note that despite the 750FX's 100MHz clock disadvantage, it is able to outperform it by 22%. Since there is a 13% difference in clock speed, and if clocks were equal, the 750FX is technically 25% more efficient in scalar integer. I should also re-emphasize that I never bothered compiling the test natively for x86, I left it java, so it's not out of the question the P4 could do this in 1 second - and that is *NOT* using any vector libraries, just plain old integer math.

I've found some documentation on the Altivec C programming interface, and this weekend I'm going to make a first attempt at vectorizing it. The integer test should be no problem, but my FPMathTest app that did square roots will be more difficult. With Altivec, there is not recognized double precision floating point, so this complicates doing square roots. If you want more accurate, precision square roots, you have to do Newton Raphson refinement. In other words more bull**** you have to go through. I believe in SSE2 you have double precision floating point ops, and if you were to vectorize it, you wouldn't have to compensate for this.


Another theory as to why the P4 is scoring so good is because if I'm not mistaking (and I'm not), the P4's ALU runs at double its clock. So in my case, 5.6GHz. I'm sure this relates to the issue.


I don't know how true this is, but I wouldn't be suprised if there is some truth to it, surely some food for thought:

http://www.osopinion.com/perl/story/17368.html


The G4 was just a hacked-up G3 with AltiVec and an FPU (floating point unit) borrowed from the outdated 604

If this is the case, then no wonder why we are getting these abysmal scores, and no wonder why a 400mhz Celeron can nearly equal it, and no wonder why the 750FX can outperform it (different company, different fpu)

 

[jefhatfield]

Originally posted by Backtothemac


And I care why? It doesn't matter how fast you can surf on your PC. I can get around fast enough on my Mac. People who say Mac's are too slow are the same people that never take the time to watch a sunset or spend a day with their kid.

They are fast enough. They do what they are supposed to do the way they are supposed to do it.

The don't crash, don't get viruses, and don't look like something from the 1980s!

but look at dis man

gots to be able to surf da net fastest to see all da nekkid pictures before da wife comes de home

 

[javajedi]

Originally posted by gopher

Maybe we have, but nobody has provided compelling evidence to the contrary.

Originally posted by gopher

And until someone does I stand by my claim.

Originally posted by Backtothemac

My reference is to the flaws of the X86 vs the PPC architecture

Originally posted by Backtothemac

Why is the PC faster? It is the OS, not the processor. Windblows uses .dll's Dynamic link libraries.

Originally posted by Backtothemac

the X 86 structure sucks

I gave you what you asked for, a fair and balanced benchmark, one even created by a Mac user. You guys have seen the code to the simple floating point and integer benchmarks, have you learned anything? You read my other posts, as well as those by others who ported the code themselves and tried it on a variety of platforms. I think it's good you both are passionate about the Mac, but hopefully this has done you both good, opened your eyes to the truth of G4 performance. Personally, I think the truth was there all along, you just didn't want to see it. Knowing what you now know you can no longer rationalize claims such as "x86 sucks", that it is the "OS that makes the other computer so fast "(a crappy one at that according to both of you).

You have both seen some very serious flaws in the G4 (to the extent a G3 can out perform it) and so you can no longer claim you have no irrefutable proof. I think you also both know that those who ran the test, including myself are not "PC biased".

You also know this has nothing to do with how much you hate "Winblowz", what you think of the CEO of chipzilla, how much nicer the Macintosh UI is, or how pretty you think youre computers are.

 

[jefhatfield]

i agree with you that pcs are faster and that some mac users will not see the facts today, but what major advantage does the faster pc give to me (the average user with e-mail, internet, office, and sometimes light graphics and digital photos)?

but it would be nice for the macs to close the speed gap someday, whether it's done with motorola processors or ibm processors

when i tell die hard mac users that one can do graphic design with pcs, they look at me like i am crazy

but today, in late-2002, it's entirely possible and if a computer user is used to using pcs for office stuff and wants to learn graphic design, it's ok for them to learn on their pc (assuming it's not too old)

only if they insist on a mac should they learn that platform to do most types of graphic design

at one time, the mac was the only real choice

it's kind of like music and rock and roll in the early days of recording that medium...if the bass player wanted to record an electric bass, it was almost always a "fender" inc. bass that was used

even years after bass players started using bass guitars made by other makers, they would still often get credited on the album as having played the fender bass or being the fender bass player

today, of course, a bass player is simply referred to as a bass player and their instrument of choice is as likely to be another make as it is to be a fender instruments electric bass guitar

 

[snoopy]

Originally posted by javajedi


. . . You have both seen some very serious flaws in the G4 (to the extent a G3 can out perform it) . . .

I'm sure Apple has known of those flaws for quite some time, and likely that is what started them looking for another processor vendor. I'm sure Apple and IBM have been working together for more than two years. Hopefully we will see the fruits of that venture in six months or so. Stay tuned for feedback from the October 15th Microprocessor Forum.

 

[jefhatfield]

that's prolly why sj goes ballistic when any reporter mentions anything to that effect...it may be true or ...maybe... apple and steve jobs may be with motorola 100% percent and hate the "ibm talk" because of how it undermines the high end computing relationship they have now

on the low end, the G3 has had a pretty good run and now with the G3fx and 512k level 2 cache, things are good in that sector for some time to come...hopefully

 

[alex_ant]

Originally posted by jefhatfield
i agree with you that pcs are faster and that some mac users will not see the facts today, but what major advantage does the faster pc give to me (the average user with e-mail, internet, office, and sometimes light graphics and digital photos)?

IMO, not much. A couple things would be the ability to do all of those a bit faster, but that only makes a difference if you're being held back by your Mac at the moment.

2 points: 1) I think the computing industry has historically been all about the trickle-down effect, where the highest of high tech starts at the very top - the high-end workstations, the mainframes, etc. - and trickles down into low-end workstations/servers, then desktops, then consumer electronics. This could be seen as a technological entropy of sorts, and if you look at it as a hierarchy, the PC (hardware wise) is closer to the root (top level) of that hierarchy at the moment. What that means is that it's closer to being the latest & greatest than the Mac is, which puts it in a position whereby its relative speed advantages are self-perpetuating, in that being closer to the source of the newest, best technology, it has a chance to incorporate that technology before the Mac does, thus raising itself up on the hierarchy yet further. This explains why PCs have been eating into the specialty markets of SGI and Sun (and Apple) and show no signs of stopping. The Mac is a fantastic platform, but it has some formidable competition that is driven by the pure force of the capitalist marketplace, and when you look at it that way, you realize how amazing it is that it has held on all this time.

2) Software is always getting more featureful and less efficient. (With a few exceptions, like the way the performance of OS X has improved between the public beta and Jaguar.) The kind of Mac that's adequate now (say an 800MHz TiBook) will probably seem quite slow in three years, whereas if you buy a top-of-the-line PC notebook today, it could easily last 5 or more. With OS X, the days of Macs lasting 5+ years are gone, at least for the moment. We do things with our computers today that we didn't do with them 5 years ago - mainly due to the trickle-down effect. We do pro-quality video editing on consumer-class machines, our resolutions and color depths are higher, our digital cameras take higher-resolution photos, our audio & video is encoded with more processor-intensive compression codecs, and hell, our email client has a little tray that slides out! (Imagine animation like that on a ca. 1997 computer running a ca. 1997 OS!) A Mac will always be able to check e-mail, but so will a Performa or a 486. But I don't know how many people Performas and 486s appeal to. Probably not many... you tell me why.

Alex

 

[jefhatfield]

Originally posted by alex_ant

The kind of Mac that's adequate now (say an 800MHz TiBook) will probably seem quite slow in three years, whereas if you buy a top-of-the-line PC notebook today, it could easily last 5 or more. With OS X, the days of Macs lasting 5+ years are gone, at least for the moment. We do things with our computers today that we didn't do with them 5 years ago - mainly due to the trickle-down effect.

Alex

because the way the pc software gets so overbloated so fast, any pc laptop is rendered too slow in two years and any pc desktop (with the desktop's higher specs and expandability) is rendered too slow in three years

i can't see any pc lasting four years comfortably, unless it's an ultra sparc, sun, or silicon graphics unit

i am assuming this for someone who would sometimes need to use photoshop, autocad, or a fifty dollar high end game

.....

as for macs, i give them the same time frame even though they are behind the pc speed curve

i don't see mac software titles pushing the mac hardware off the planet like in the pc world, which is seen more as a throwaway consumer electronic

thank god that macs are not seen or built as throwaway consumer electronics

even the "now" lowly crt imac is a sturdy machine that will outlast, on the physical level, most pcs on the market

.....

when i got my ibook, even though the single usb port left me stranded peripheral wise two years later, it was built to last and last

when i got my pc laptop, made by compaq, the thing was definitely sold as a throwaway unit

the rubber feet fell off which i had to glue back on

one screen hinge kept on popping off so i have to avoid touching it on that left side

when i close the pc laptop unit, i have to do it slowly since that particular model had thin plastic latches that broke off easily and the ribbon cable connecting the lcd had a tendency to get unplugged inside the unit

and the battery was useless after a year and wouldn't hold a charge anymore

i never shelled out the $199 bucks to get a new battery and now i just use the short length ac adapter

.....

in contrast, my ibook's only deterioration has been the battery's ability to hold a 4 1/2 hour charge...the thing never got 6 hours in real world everyday use like advertised...using just word processing with the lcd dimmed way down, a reviewer got five hours on a new rev a. ibook battery

now the laptop's battery, after 34 months of daily use, holds a 2 3/4 hour charge...actually, not bad compared to the pc laptop whose battery died after just a year

.....

when i looked at a computer accessories catalog, they recommended that i replace my pc model's battery after one year of part time use

but they also recommended that i replace my rev. a ibook's battery after just one year, also...how wrong they were...ha

if i still have my 300 mhz ibook two years from now, even if i wouldn't likely be using it much, i will give it a five year birthday party on macrumors...ibook's in late-2004 will be at 1.9 ghz by then if apple still has an ibook on the consumer end...this is based on average speed climb in industry

right now, the earliest rev. a ibooks are now 3 1/4 years old, originally had os 8.5, and i bet most are still working

 

[nixd2001]

Originally posted by javajedi
I gave you what you asked for, a fair and balanced benchmark, one even created by a Mac user. You guys have seen the code to the simple floating point and integer benchmarks

It would be interesting to see the code generated for the loops - it won't change the answers but it might give some of us a bit more understanding on the perfomance differences.

 

[benixau]

Originally posted by jefhatfield

thank god that macs are not seen or built as throwaway consumer electronics

Too right. I had an LC630 up until about 2 months ago. First problem occured on week into use, apple monitor stoped working. 10 yrs later, power supply went. Only problem after the monitor problem and before the power supply porblem was the family needed to use it. That only meant that i couldnt sometimes.

A pc we had was upgraded every 6 months to make sure it would run. 2 months ago, four years into its life, the celeron 333 couldn't handle windows XP (it came with 98) after a serious OS problem. We chucked it, bought an AMD XP 1800+ and got two brand new power macs (MDD model). Cant afford an iBook or Powerbook yet.

Macs last longer, as long as you dont want the latest and greatest app to run, like Office v.X on a performa. If you can live with Office 2001 or 98 then, why get something untested.

A mac cost 2x as mush upfront. They last 4 - 5 times as long. Who saves money???

 

[MacCoaster]

Originally posted by nixd2001


It would be interesting to see the code generated for the loops - it won't change the answers but it might give some of us a bit more understanding on the perfomance differences.

javajedi's Java and Cocoa/Objective-C code has been available here for a couple of days. My C# port is available for examination if you e-mail me.

 

[nixd2001]

Originally posted by MacCoaster

javajedi's Java and Cocoa/Objective-C code has been available here for a couple of days. My C# port is available for examination if you e-mail me.

I was thinking of the x86 and PPC assembler produced for the core loops. I could bung the C through GCC and get some assembler on my windy tunnels, true, but I'm not geared up to do the Windows side of things.

 

[MacCoaster]

Originally posted by nixd2001
I was thinking of the x86 and PPC assembler produced for the core loops. I could bung the C through GCC and get some assembler on my windy tunnels, true, but I'm not geared up to do the Windows side of things.

You could add the argument --funroll-loops to gcc to `unroll' the loops and make it faster by predicting it more accurately at compile-time.

 

[benixau]

for crying out load, who cares if a pc can do its sums better than a mac. My brother does maths better than me but i kick him in english.

In other words if i am more productive on my mac then it doesnt matter that it might be a little 'slower' it is a faster machine because i can work faster. End of story. New Thread.

 

[jefhatfield]

Originally posted by benixau



A pc we had was upgraded every 6 months to make sure it would run. 2 months ago, four years into its life, the celeron 333 couldn't handle windows XP (it came with 98) after a serious OS problem. We chucked it

even swapping out parts and stuff, for any pc box to last four years is admirable...he he

but even if your celeron 333 worked after four years, it wouldn't be able to run much these days on the pc software shelf

however, a 333 mhz imac could prolly still run stuff off the shelf as it still exceeds min requirements of a lot of mac software

i have seen pc games which have min requirements of pentium II or athlon already...and we know what that means...recommended requirements mean it should be a pentium III 650 or so

to me, that is built in obsolescence...it's not that way in the mac world...at least, not as bad

if i see mac software that says min requirements, g4 and os x, then i will protest

 

[snoopy]

Originally posted by benixau


for crying out load, who cares if a pc can do its sums better than a mac. . . . . if i am more productive on my mac then it doesnt matter that it might be a little 'slower' . . .

True for many of us. For applications that use a lot of math functions, it makes a big difference. So, for others it does matter. They may be in the minority, but a very important group of users. In less than a year the picture will change, and that small group will be very pleased with the Mac. For now, there is nothing anyone can do about it.

 

[MacCoaster]

Originally posted by benixau
for crying out load, who cares if a pc can do its sums better than a mac. My brother does maths better than me but i kick him in english.

In other words if i am more productive on my mac then it doesnt matter that it might be a little 'slower' it is a faster machine because i can work faster. End of story. New Thread.

Believe me, a lot of people do. Thanks to my UNIX knowledge, I am so much more productive in Linux/BSD on a PC than a Mac. For beginners to computers, sure Macs could be much more productive.

We were just discussing the G4--it was never intended to be an explict vs war between Mac and PCs. It's not a software thread. It's a frickin' hardware thread where we are discussing the inferiority of the G4.

Research scientists should think twice before using a Mac for research--since the G4 blows so much. That's where it matters. It's faster for them to use PCs than Macs. Gee, by 100 seconds. Think about it... a lot of scientific formulas are a lot more complex than our simplistic benchmark programs--100 minutes is sure much longer than 5 minutes.

 

[jefhatfield]

Originally posted by snoopy


True for many of us. For applications that use a lot of math functions, it makes a big difference. So, for others it does matter. They may be in the minority, but a very important group of users. In less than a year the picture will change, and that small group will be very pleased with the Mac. For now, there is nothing anyone can do about it.

those math functions are extremely complex and hard to do fast if we stay way behind the curve of the pc world

i was in this computer repair class where we had to do the math, some of the math that a processor did, so we could appreciate that little thing

in the old days of computing, way back when in the 1970s, many computing funtions had to be done by phd mathematicians and there were very few silicon "math co-processors"

early computer science college programs were thus a lot like math programs...it's so funny, actually sad, to see how many older, math literate techies were completely unable to relate when gui came along...it was like the great slaughter in silicon valley...we take the mouse and gui for granted but not only did it take away jobs, it also was a curve ball many inflexible older techies could not adjust to

change is never easy in the IT field and that is why it is rare to see anybody go from mathematician with vacuum tubes to green screen coder to gui to "whatever" the future holds

i also had a friend who had memorized hundreds of key combinations like ctrl-a and such and he only just learned to use the mouse two years ago...he took literally five years to learn how to use it with its two buttons...he could never remember, "was that right click, left click, double click, and where do i keep my fingers?"

i could go on with old man stories from the trenches of san jose, but i will stop NOW

if you started with a mouse, it only takes a few weeks to learn how to interact with windows and modern computers

one family friend, a computer professor at stanford, never got used to gui and he still uses his trusty 286...he says he can't think when there is more than one color on the screen and he never got used to the mouse

kind of the way i feel like when i use "hex-pee" or i try to play a game console thingy like x-box with all those buttons...as a ten year old yanks the keypad/console from me at the computer store and memorizes the keys and buttons within minutes as it relates to that game being played

 

[jefhatfield]

Originally posted by MacCoaster

Believe me, a lot of people do. Thanks to my UNIX knowledge, I am so much more productive in Linux/BSD on a PC than a Mac. For beginners to computers, sure Macs could be much more productive.

We were just discussing the G4--it was never intended to be an explict vs war between Mac and PCs. It's not a software thread. It's a frickin' hardware thread where we are discussing the inferiority of the G4.

Research scientists should think twice before using a Mac for research--since the G4 blows so much. That's where it matters. It's faster for them to use PCs than Macs. Gee, by 100 seconds. Think about it... a lot of scientific formulas are a lot more complex than our simplistic benchmark programs--100 minutes is sure much longer than 5 minutes.

too many of those programs are only on pcs

one research scientist my wife works with started coding in dos on the mac compiler and if he succeeded in getting into the server, which would not happen anyway, he would have caused major damage

this phd had no idea that the g4 and the mac os was not dos...he was sure everything was dos like his windows 98 box he and all the other research scientists use

the sas program they have only works on 95 and 98

 

[benixau]

well well well

If you want to get exceptional mathematical performance then why are you getting a micro computer???? I cannot out-type my computer and i cannot do mathematical functions fater than it, or even excel with all of its overhead.

BTW, my g4 is soooo slow at doing maths functions that i finished an assignment a whole 5mins ahead of a mate. In excel. these were some serious slowdown stuff, 10 cross-referenced, dependently linked, nested functions sheets. Now my mac only has 2 867s with 256ddr, his p4 2.53 with 512 couldnt beat me, WITH WIN95.


Now any more real world tests you would like????

 

[ddtlm]

Wow I missed a lot by spending all of Friday away from this board. I am way behind in posts here, and I'm sure I'll miss a lot of things worth comment. But anyway, the code fragment:

int x1,x2,x3;
for (x1=1; x1<=20000; x1++) {
for(x2=1; x2<=20000; x2++) {
x3 = x1*x2;
}
}

Is a very poor benchmark. Compilers may be able to really dig into that and make the resulting executable perform the calculate radically different. In fact, I can tell you the answer outright: x1=20000, x2=20000, x3 = 400000000. It took me 2 seconds or so. Does this mean that I am a better computer than a G4 and a P4? No, it means I realized that the loop can be reduced to simple data assignments. I have a better compiler, thats it.

Anyway, lets pretend that for whatever reason compilers did not simplify that loop AT ALL. Note that this would be a stupid stupid compiler. At each stage, x1 is something, we ++x2, and we set x3 = x1 * x2. Now notice that we cannot set x3 until the result of X2++ is known. On a pipelined processor that cannot execute instructions out of order, this means that I have a big "bubble" in the pipeline as I wait for the new x2 before I can multiply. However, after the x3 is started into the pipe, the next instruction is just another x2++ which does not depend on x3, so I can do it immediately. On a 7-stage in-order chip like a G4, this means that I fill two stages of the pipe and then have to wait for the results on the other end before I can continue. You see that this is very inefficient (28% or so). However, the G3 is a 4-stage design and so 2/4 of the stages can stay busy, resulting in a 50% efficientcy (so a 700mhz G3 is "the same as" a 350mhz G3 at 100% and a 800mhz G4 is "the same as" a 210mhz G4 at 100%). These are of course simplified cases, the actual result may very a bit for some obscure reason.

Actually the above stuff is inaccurate. The G3 sports 2 integer units AFAIK, so it can do x3 = x1*x2 at the same time as it is doing x2++ (for the next loop of course, not this one). This means that both pipes start one bit of work, then wait for it to get out the other end, then do one bit of work again. So this is 25% efficientcy. A hypothetical single-pipe G3 would do x3 = x1*x3 and then do x2++, however it could not do x3 = x1 * x2 again until the x2++ was out the other end, which takes 4 cycles and started one after the previos x3 = x1*x2, which should mean 3 "bubble" stages and an efficientcy of 20%.

Actually, it may be worse than that. Remember that this is in a loop. The loop means a compare instruction (are we done yet?) followed by a jump depending on the results of the compare. We therefore have 4 instructions in PPC I think per loop, and we can't compare x2 to 20000 until x2++ has gone through all the pipe stages. (Oh no!) And we can't jump until we know r]the result of the compare (oh no!). Seeing the pattern? Wanna guess what the efficientcy is for a really stupid compiled version of this "benchmark"? A: really freaking low.

I'll see about adding more thoughts later.