<strong>I'm sorry, but I'll just ignore your comments in the future, Eric, until you get your facts straight. By saying something like (freely) RISC was designed to have high IPC, while being slower on MHz and thus still beating the CISC competition, you made such a HUGE arse of yourself, it almost hurts.
Maybe you should spend the next month reading Arstechnica (ars as in latin for art, not arse), and not here posting facts from Eric's Wonderland.</strong><hr></blockquote>
I never said that.
From my other post:
"The best-laid plans of mice and engineers?
Seriously though. despite their best efforts. the RISC CPU turned out to be a fantastic "Low Frequency" chip. I wasn't talking about the original dreams of the first RISC engineers. I was talking about current reality. but the best things sometimes come from the least expected places.
<strong>That's plain wrong. When the Alpha was introduced, its clock speed was way ahead of the x86 chips of that time.</strong><hr></blockquote>
I wasn't speaking of all eternity. I mean't for a fair while.
[quote]Originally posted by RazzFazz:
<strong>x86 CPUs only started to surpass them MHz-wise when the only CISCy thing left was the x86-to-uOP decoder in the first few pipe stages.</strong><hr></blockquote>
I already knew that. note my constant usage of the term "CRISC"(<a href="http://www.mackido.com/Hardware/x86RISC.html"; target="_blank">Complex Reduced Intruction Set Computer</a>).
[quote]Originally posted by RazzFazz:
<strong>When was the last time you saw a true CISC CPU from Intel or AMD?</strong><hr></blockquote>
Refer to "CRISC" comment.
[quote]Originally posted by RazzFazz:
<strong>It's true that an Alpha neatly outperforms an Athlon or Pentium running at much higher clock speeds, but that's not some sort of an inherent benefit of being RISC, but rather a tribute to the Alphas vastly superior, clean architecture. The same is true for most other classical examples of RISC chips - unlike the x86 processors, they had the luxury of starting from scratch with a clean, sane architecture, and they were initially targeted at the very high-end market, and could thus afford to put features into that architecture that were at that point unavailable to the x86 CPU makers for economical reasons.</strong><hr></blockquote>
<strong>Right, that's probably why both the Athlon and the P4 use a RISC back-end.</strong><hr></blockquote>
[quote]Originally posted by RazzFazz:
<strong>High frequency RISC is reality (albeit with a CISC-to-uOP decoder slapped on) - that's why all current CPUs (VLIW aside) use a RISC architecture (at least internally). There's no way a true CISC architecture could scale anywhere close to today's clock speeds.</strong><hr></blockquote>
good on ya Eric for posting your opinion and backing it up time and time again.
shame on the rest of you for getting all personal.
on the subject of RISC vs CISC, does anyone here have any personal first-hand knowledge of what they are talking about or are we all just i-got-my-info-from-here-so-your-info-must-be-wrong-people?
Well it's obvious that an indepented source is more unbiased than DEC hammering on the competitors technologies. And on the RISC vs CISC issue, Arstechnica is one good read, even if other stuff they post isn't always to my liking.
As for the source of information, masta:
You know that we get 80% of our total knowledge and information out of the media, while the other 20% are very basic things usually, like, well like the very basic things a human being has to be capable of doing. Thus I think it's legal to put one opinion against the other, even if both sides didn't invent the RISC design. OK?
You should read DEC's paper. the first link I posted was bad. the engineer that wrote it is probably one of the finest chip engineers. and. Alpha bias or not. this guy _definitely_ knows what he's talking about.
[quote]As for MIPS. the research money for high end MIPS has dried up since that slime Belluzzo(Whome promptly moved to COO at Microsoft?) decided that SGI would start migrating to Intel's sucky IA64. and when Compaq subsequently killed it's himalaya line for the same reason. if they(MIPS) still had those two income sources. the MIPS would still crush all CISC and CRISC comers.
Actually. I dug up the dirt on Intel suspiciously killing perfectly good RISC architectures myself.
[quote]Originally posted by johnsonfromwisconsin:
<strong> No wonder you're so messed up Eric; David K Every is a frickin' shill.
yeah he got most of the ISA portion right (by cooincidence), but his microarchitecture explanation falls short and is saturated with spin.</strong><hr></blockquote>
[quote]Originally posted by johnsonfromwisconsin:
<strong>posting mackido links won't help the credibility of your argument. DKE has, well, err, not exactly the best reputation...</strong><hr></blockquote>
Hmph. David K. Every is a nifty guy. the only thing most of his arguments seem saturated with is unbreakable strings of logic.
From his site:
"I've always been an opinionated SOB. I dealt with some of the News Groups (political and computers), and got tired of the Advocacy fights that weren't getting anywhere, and always saying the same things over and over again. So I clarified many of my points, and create this site to explain things once (and in more detail) to try stop the misinformation, "
For since I don't like saying the same things over again either. I like to link(Or excerpt. if the #$!& site doesn't support links) from MacKiDo and others like it. it saves me time. and besides. it would take to long to think up some of the brilliant points, analyses and arguments on sites like that.
[quote]Actually. I dug up the dirt on Intel suspiciously killing perfectly good RISC architectures myself.<hr></blockquote>
SGI were very glad to drop MIPS in Favor of IA64 as they could get someone else to develop MPU's for them. SGI is the one who killed (rather is killing off) MIPS for economic purposes.
[quote]Hmph. David K. Every is a nifty guy. the only thing most of his arguments seem saturated with is unbreakable strings of logic.<hr></blockquote>
That may seem to be the case, coming from the POV of someone who just wants to believe in any old myths that benifits their platform or puts the "Wintel" world in a bad light.
<strong>SGI were very glad to drop MIPS in Favor of IA64 as they could get someone else to develop MPU's for them. SGI is the one who killed (rather is killing off) MIPS for economic purposes.</strong><hr></blockquote>
I'm afraid that explanation falls flat on it's face for the same reason that the classic "Apple should port OS X to 80x86 now!" one does.
Let's say you're SGI. and you think:
<Q> What would make a prospective customer want an SGI branded(And more expensive. do to lower component order volume discount) IA64 box loaded with IRIX, UNICOS or Linux over an utterly identical. dirt cheap one form Dell, Gateway etc. running MS Windows 64 or something?
<A> The exact same as in the Apple one. absolutely nothing.
<Q> What would keep some enterprising hacker from making IRIX or UNICOS run on other brands of computer against our will?
<A> The exact same as in the Apple one. absolutely nothing.
Obviously. you would need some kind of an outside force to point you down this road. and what could be more obvious than the very dirtbag that put HP's PA-RISC up on the same sacrificial altar mere months before. and currently resides in a cushy executive position at the second party most likely to benefit from such moves.
[quote]Originally posted by johnsonfromwisconsin:
<strong>That may seem to be the case, coming from the POV of someone who just wants to believe in any old myths that benifits their platform or puts the "Wintel" world in a bad light.</strong><hr></blockquote>
I judge your opinion on this matter to be fixed at this point. and I have already conveyed mine. further intercourse upon it is thus unnecessary.
[quote]I'm afraid that explanation falls flat on it's face for the same reason that the classic "Apple should port OS X to 80x86 now!" one does.
Let's say you're SGI. and you think:
<Q> What would make a prospective customer want an SGI branded(And more expensive. do to lower component order volume discount) IA64 box loaded with IRIX, UNICOS or Linux over an utterly identical. dirt cheap one form Dell, Gateway etc. running MS Windows 64 or something?
<A> The exact same as in the Apple one. absolutely nothing.<hr></blockquote>
So your blaming Intel for being able to outcompete SGI?
[quote]
<Q> What would keep some enterprising hacker from making IRIX or UNICOS run on other brands of computer against our will?
<A> The exact same as in the Apple one. absolutely nothing.
<hr></blockquote>]
And you blame that on Intel?
I find your analysis lacking for this reason:
It doesn't address the economic motivation for the likes of SGI to adopt a different microarchitecture. You talked like SGI had a lot of resources to develop MIPS with from the beginning, and Belluzo just sold out in some sort of cloak and dagger scheme. If I remember, MIPS didn't even show up in the running as far as processors in servers goes, and was only about third in scientific workstations behind IBM and Compaq.
The MIPS was never a high volume or even a great performing processor compared to other 64-bit MPU's from Compaq, IBM, or HP. MIPS is very niche even within the 64-bit market which itself is a bit of a niche and it couldn't produce processors to compete with these or even the 32-bit x86. This is the reason the decision to phase out the MIPS in favor of IA64:
SGI didn't have the resources to compete with other 64bit or 32bit processors and IA64 promised the resources and engineering of the largest Microprocessor manufacturer in the world. IA64 not only *promised* better performance but it promised to bring commodity pricing to the 64bit market. No longer would SGI have to swim up a raging stream to develop it's own architecture, neither would it have to shell out a lot of money to buy competing 64-bit processors.
This is roughly the same reason why HP decided to invest in IA64 as well.
I've talked to him several times and had a few interesting discussions with him, but while his argumentations rarely fall short of logic, they always also show a great degree of stubbornness.
He's going to tell you what he thinks is right, simply bypassing the fact that often more than one possibility is correct. He's a "different thinker" and always has to contradict a public opinion, even if that opinion is actually quite viable. As long as he objects to it, and even if he's alone to object to it, he's going to argue.
However I'm glad that we had the same opinion on the Xtrem Mac back then, when everyone was euphorically drooling in anticipation of something that was never going to happen, and I and he both knew it.
DKE is a moron. Anyone who quotes him or links to anything he wrote as evidence for their side instantly loses absolutely every shred of credibility or believability in any argument they put forth.
Do a search for him in ArsTechnica's Battlefront to find the reasons why. He's been amply discussed.
So, in all honesty, you actually do believe that a "CRISC" CPU, which basically is a RISC core with additional decoding stuff slapped on, can inherently scale higher than a much less complex RISC CPU?
Has it ever occured to you that the clock speed advantage of current Intel and AMD CPUs is more related to the fact that both of them invest insane amounts of money in their designs, fabs and processes, rather than some inherent "low-frequencyness" of all RISC CPUs?
The way you say it, you make it sound like one can tinker with the spec source code, or write special compilers that are only designed to produce good spec binaries. But the spec benchmarks have a lot of rules that must be adhered to. First, you cannot touch the source code, you cannot optimise it in any way shape or form. You CAN use compiler optimisations, but they're only allowed as far as they improve overall application performance in general, and are not just profiled for spec sources. The main rules are:
</strong><hr></blockquote>
This discussion about SPEC is ridiculous. By your own admission, you don't think a 700mhz P3 is faster than a 1GHZ G4. Likewise, it puzzles me why you put any credibility in SPEC results.
Let me tell you a few things about SPEC.
1) Did you know that SPEC has been proven to vary by more than 30% based on compiler alone?
2) Are you aware that the gcc compiler is not what's used for production Mac apps? Are you also aware that gcc (though free) is notoriously slow?
3) How about the relevance of some of the tests? When was the last time you needed to do weather forcasting on your PC? I'll agree that some of the individual tests hold some intellectual curiousity, but the aggregate score is not meaningful.
4) You are correct about SPEC not allowing for code changes or optimizations. Of course, this is SPECs biggest downfall. What commercial product runs on multiple platforms that has no optimizations for a given platform? What about SIMD? This is the G4's biggest strenth, yet there is absolutely no way to measure this via SPEC. Intel has done some work in "auto vectorizing" in it's compilers, but it's nowhere near as effective as manual SIMD optimizations as seen in commercial software. To my knowledge, no auto vectorizing exists in current Mac compilers. Therefore, it would seem that this gives Intel's results an artificially higher score.
That said, take a look at Apple's market and the programs that are key to Apple's success. That is, what programs are Mac users most likely to use? Photoshop, "i apps" (idvd,itunes, etc.) video compression, etc. All of these programs make extensive use of SIMD (Altivec or Velocity Engine), yet are not measured in any way by SPEC. Therefore, SPEC is irrelevant for the purposes of Mac to PC comparisons.
There are many valid uses for SPEC. SPEC is better served for measuring compiler performance than it is for cross platform performance, particularly in the PC market.
<strong>So your blaming Intel for being able to outcompete SGI?</strong><hr></blockquote>
No. I'm saying there's no logical reason whatsoever for SGI to migrate onto a mainstream chip alongside a major version of MS Windows.
[quote]Originally posted by johnsonfromwisconsin:
<strong>And you blame that on Intel?</strong><hr></blockquote>
No. I blame the fact of that SGI ignored it on Intel.
[quote]Originally posted by johnsonfromwisconsin:
<strong>I find your analysis lacking for this reason:
It doesn't address the economic motivation for the likes of SGI to adopt a different microarchitecture. You talked like SGI had a lot of resources to develop MIPS with from the beginning, and Belluzo just sold out in some sort of cloak and dagger scheme. If I remember, MIPS didn't even show up in the running as far as processors in servers goes, and was only about third in scientific workstations behind IBM and Compaq.</strong><hr></blockquote>
Yes. they used to have _quite_ a bit of resorces. SGI used to be first and foremost in the scientific/graphical computing world. and do you realize what an honored position third in scientific workstations behind IBM and Compaq is? that means that the MIPS was competitive with. and surpassed _only_ by the POWER and Alpha CPUs. this means that SGI manufactured the _third fastest CPU on earth_. whereas the IA64 can not be said to hold such a title.
[quote]Originally posted by johnsonfromwisconsin:
<strong>The MIPS was never a high volume or even a great performing processor compared to other 64-bit MPU's from Compaq, IBM, or HP. MIPS is very niche even within the 64-bit market which itself is a bit of a niche and it couldn't produce processors to compete with these or even the 32-bit x86</strong><hr></blockquote>
On the contrary. what do you think has powered the N64, PS and PS2. as well as numerous printer, digital copier and network router embedded logic systems? the MIPS was(And is) one of the _most_ mainstream 64-bit CPUs on the market
[quote]Originally posted by johnsonfromwisconsin:
<strong>This is the reason the decision to phase out the MIPS in favor of IA64:
SGI didn't have the resources to compete with other 64bit or 32bit processors and IA64 promised the resources and engineering of the largest Microprocessor manufacturer in the world. IA64 not only *promised* better performance but it promised to bring commodity pricing to the 64bit market.</strong><hr></blockquote>
Intel promised a lot of things with the IA64. but they only delivered half way. for instance. the MIPS R14000 beats the Merced in integer performance(As well as nearly every other CPU on earth ). while the Merced wins in floating point performance. if MIPS knew they still had SGI and Compaq as future customers. they could easily have roasted Intel's chip.
And commodity pricing is right! with PA-RISC out of the picture from the start. and later MIPS. thanks to Belluzo. Alpha dead after being handed on a silver plate to Intel by their thrall Compaq and even some rumors of the POWER being surrendered by IBM for some unknown reason. I think being the ONLY major 64-bit CPU on the market other than Sun's would make practically anything you can slap together go for "Commodity pricing".
[quote]Originally posted by johnsonfromwisconsin:
<strong>No longer would SGI have to swim up a raging stream to develop it's own architecture, neither would it have to shell out a lot of money to buy competing 64-bit processors.</strong><hr></blockquote>
Neither would much stronger companies like Dell. once their distinguishing advantages are gone. SGI's customers will scatter like a flock of pheasants after a shot. and I think that all the people at SGI obviously could have figured this out.
[quote]Originally posted by johnsonfromwisconsin:
<strong>This is roughly the same reason why HP decided to invest in IA64 as well.</strong><hr></blockquote>
While HP didn't have as much to lose(They do other things than HP-UX). and they were already a big Intel buyer. HP just doesn't seem to have been in line to benefit from the IA64 much. as a sign of the IA64's _very_ late and exceedingly underwhelming arrival. the PA-RISC project has been extended another five years. and the PA-RISC is currently a very quick chip(at this moment a bit ahead of the MIPS). and I think that if all of HP's CPU design resources were still dedicated to the PA-RISC than it too would easily be faster than the Itanium.
SGI was once considered quite powerful and strong enough for a bright future.
back in the 90s talking about an SGI Workstation, or calling "O2" would get every computer freak excited like a schoolgirl on halloween. "OMG I want one of these".
In the movie "Starship troopers" (which was made heavily on SGI), SGI even appears in the story as one HUGE world concern with dimensions of Microsoft and more, almost like the US today. everything is owned by SGI there (of course that is a joking hidden ad).
I guess they sat on their laurels a bit too long and now lost all but the movie customers.
<strong>In the movie "Starship troopers" (which was made heavily on SGI), SGI even appears in the story as one HUGE world concern with dimensions of Microsoft and more, almost like the US today. everything is owned by SGI there (of course that is a joking hidden ad).</strong><hr></blockquote>
There was a similar ad for a future SGI workstation(That bore a strong resemblance to the O2) in the film "Lost in Space" too.
Comments
<strong>I'm sorry, but I'll just ignore your comments in the future, Eric, until you get your facts straight. By saying something like (freely) RISC was designed to have high IPC, while being slower on MHz and thus still beating the CISC competition, you made such a HUGE arse of yourself, it almost hurts.
Maybe you should spend the next month reading Arstechnica (ars as in latin for art, not arse), and not here posting facts from Eric's Wonderland.</strong><hr></blockquote>
I never said that.
From my other post:
"The best-laid plans of mice and engineers?
Seriously though. despite their best efforts. the RISC CPU turned out to be a fantastic "Low Frequency" chip. I wasn't talking about the original dreams of the first RISC engineers. I was talking about current reality. but the best things sometimes come from the least expected places.
And RISC <a href="http://www.mackido.com/Hardware/WhatIsRISC.html"; target="_blank">isn't what it used to be</a> anyways?"
Eric,
[ 03-06-2002: Message edited by: Eric D.V.H ]</p>
<strong>That's plain wrong. When the Alpha was introduced, its clock speed was way ahead of the x86 chips of that time.</strong><hr></blockquote>
I wasn't speaking of all eternity. I mean't for a fair while.
[quote]Originally posted by RazzFazz:
<strong>x86 CPUs only started to surpass them MHz-wise when the only CISCy thing left was the x86-to-uOP decoder in the first few pipe stages.</strong><hr></blockquote>
I already knew that. note my constant usage of the term "CRISC"(<a href="http://www.mackido.com/Hardware/x86RISC.html"; target="_blank">Complex Reduced Intruction Set Computer</a>).
[quote]Originally posted by RazzFazz:
<strong>When was the last time you saw a true CISC CPU from Intel or AMD?</strong><hr></blockquote>
Refer to "CRISC" comment.
[quote]Originally posted by RazzFazz:
<strong>It's true that an Alpha neatly outperforms an Athlon or Pentium running at much higher clock speeds, but that's not some sort of an inherent benefit of being RISC, but rather a tribute to the Alphas vastly superior, clean architecture. The same is true for most other classical examples of RISC chips - unlike the x86 processors, they had the luxury of starting from scratch with a clean, sane architecture, and they were initially targeted at the very high-end market, and could thus afford to put features into that architecture that were at that point unavailable to the x86 CPU makers for economical reasons.</strong><hr></blockquote>
Isn't that what <a href="http://www.mackido.com/Hardware/WhatIsRISC.html"; target="_blank">I said</a>?
[quote]Originally posted by RazzFazz:
<strong>Oh, and besides, why would RISC have an advantage over VLIW?</strong><hr></blockquote>
Read <a href="http://citeseer.nj.nec.com/rd/74623007,266363,1,0.25,Download/http%3A%2F%2Fciteseer.nj.nec.com/cache/papers/cs/12850/http%3AzSzzSzwww.digital.comzSzhpczSzrefzSzref_a lpha_ia64.pdf/alpha-and-ia.pdf" target="_blank">this DEC paper on why VLIW(And by extension. IA64) suck</a>.
edit:
Whew! The DEC paper was originally at <a href="http://www.digital.com/hpc/ref/ref_alpha_ia64.pdf."; target="_blank">http://www.digital.com/hpc/ref/ref_alpha_ia64.pdf.</a>; and was then moved to <a href="http://www.compaq.com/hpc/ref/ref_alpha_ia64.pdf."; target="_blank">http://www.compaq.com/hpc/ref/ref_alpha_ia64.pdf.</a>; soon after Compaq sold the DEC Alpha division to Intel. the IA64 smashing article mysteriously disappeared. fortunately. after a LONG search. I managed to find a backed up copy at a whitepaper hoarding place called CiteSeer.
[quote]Originally posted by RazzFazz:
<strong>Right, that's probably why both the Athlon and the P4 use a RISC back-end.</strong><hr></blockquote>
[quote]Originally posted by RazzFazz:
<strong>High frequency RISC is reality (albeit with a CISC-to-uOP decoder slapped on) - that's why all current CPUs (VLIW aside) use a RISC architecture (at least internally). There's no way a true CISC architecture could scale anywhere close to today's clock speeds.</strong><hr></blockquote>
Refer to "CRISC" comment.
\t\t\t\t\t\t Eric,
[ 03-06-2002: Message edited by: Eric D.V.H ]
[ 03-06-2002: Message edited by: Eric D.V.H ]
[ 03-06-2002: Message edited by: Eric D.V.H ]</p>
good on ya Eric for posting your opinion and backing it up time and time again.
shame on the rest of you for getting all personal.
on the subject of RISC vs CISC, does anyone here have any personal first-hand knowledge of what they are talking about or are we all just i-got-my-info-from-here-so-your-info-must-be-wrong-people?
I have no knowledge, hence no opinion.
Peace,
G
As for the source of information, masta:
You know that we get 80% of our total knowledge and information out of the media, while the other 20% are very basic things usually, like, well like the very basic things a human being has to be capable of doing. Thus I think it's legal to put one opinion against the other, even if both sides didn't invent the RISC design. OK?
G-News
[ 03-06-2002: Message edited by: G-News ]</p>
Eric,
<hr></blockquote>
And Crimeny!
No wonder you're so messed up Eric; David K Every is a frickin' shill.
yeah he got most of the ISA portion right (by cooincidence), but his microarchitecture explanation falls short and is saturated with spin.
posting mackido links won't help the credibility of your argument. DKE has, well, err, not exactly the best reputation...
[ 03-06-2002: Message edited by: amyklai ]</p>
<strong>
And Crimeny!</strong><hr></blockquote>
Actually. I dug up the dirt on Intel suspiciously killing perfectly good RISC architectures myself.
[quote]Originally posted by johnsonfromwisconsin:
<strong> No wonder you're so messed up Eric; David K Every is a frickin' shill.
yeah he got most of the ISA portion right (by cooincidence), but his microarchitecture explanation falls short and is saturated with spin.</strong><hr></blockquote>
[quote]Originally posted by johnsonfromwisconsin:
<strong>posting mackido links won't help the credibility of your argument. DKE has, well, err, not exactly the best reputation...</strong><hr></blockquote>
Hmph. David K. Every is a nifty guy. the only thing most of his arguments seem saturated with is unbreakable strings of logic.
From his site:
"I've always been an opinionated SOB. I dealt with some of the News Groups (political and computers), and got tired of the Advocacy fights that weren't getting anywhere, and always saying the same things over and over again. So I clarified many of my points, and create this site to explain things once (and in more detail) to try stop the misinformation, "
For since I don't like saying the same things over again either. I like to link(Or excerpt. if the #$!& site doesn't support links) from MacKiDo and others like it. it saves me time. and besides. it would take to long to think up some of the brilliant points, analyses and arguments on sites like that.
Eric,
SGI were very glad to drop MIPS in Favor of IA64 as they could get someone else to develop MPU's for them. SGI is the one who killed (rather is killing off) MIPS for economic purposes.
[quote]Hmph. David K. Every is a nifty guy. the only thing most of his arguments seem saturated with is unbreakable strings of logic.<hr></blockquote>
That may seem to be the case, coming from the POV of someone who just wants to believe in any old myths that benifits their platform or puts the "Wintel" world in a bad light.
<strong>SGI were very glad to drop MIPS in Favor of IA64 as they could get someone else to develop MPU's for them. SGI is the one who killed (rather is killing off) MIPS for economic purposes.</strong><hr></blockquote>
I'm afraid that explanation falls flat on it's face for the same reason that the classic "Apple should port OS X to 80x86 now!" one does.
Let's say you're SGI. and you think:
<Q> What would make a prospective customer want an SGI branded(And more expensive. do to lower component order volume discount) IA64 box loaded with IRIX, UNICOS or Linux over an utterly identical. dirt cheap one form Dell, Gateway etc. running MS Windows 64 or something?
<A> The exact same as in the Apple one. absolutely nothing.
<Q> What would keep some enterprising hacker from making IRIX or UNICOS run on other brands of computer against our will?
<A> The exact same as in the Apple one. absolutely nothing.
Obviously. you would need some kind of an outside force to point you down this road. and what could be more obvious than the very dirtbag that put HP's PA-RISC up on the same sacrificial altar mere months before. and currently resides in a cushy executive position at the second party most likely to benefit from such moves.
[quote]Originally posted by johnsonfromwisconsin:
<strong>That may seem to be the case, coming from the POV of someone who just wants to believe in any old myths that benifits their platform or puts the "Wintel" world in a bad light.</strong><hr></blockquote>
I judge your opinion on this matter to be fixed at this point. and I have already conveyed mine. further intercourse upon it is thus unnecessary.
\t\t\t\t\t\t\t\t\t\t\t Eric,
Let's say you're SGI. and you think:
<Q> What would make a prospective customer want an SGI branded(And more expensive. do to lower component order volume discount) IA64 box loaded with IRIX, UNICOS or Linux over an utterly identical. dirt cheap one form Dell, Gateway etc. running MS Windows 64 or something?
<A> The exact same as in the Apple one. absolutely nothing.<hr></blockquote>
So your blaming Intel for being able to outcompete SGI?
[quote]
<Q> What would keep some enterprising hacker from making IRIX or UNICOS run on other brands of computer against our will?
<A> The exact same as in the Apple one. absolutely nothing.
<hr></blockquote>]
And you blame that on Intel?
I find your analysis lacking for this reason:
It doesn't address the economic motivation for the likes of SGI to adopt a different microarchitecture. You talked like SGI had a lot of resources to develop MIPS with from the beginning, and Belluzo just sold out in some sort of cloak and dagger scheme. If I remember, MIPS didn't even show up in the running as far as processors in servers goes, and was only about third in scientific workstations behind IBM and Compaq.
The MIPS was never a high volume or even a great performing processor compared to other 64-bit MPU's from Compaq, IBM, or HP. MIPS is very niche even within the 64-bit market which itself is a bit of a niche and it couldn't produce processors to compete with these or even the 32-bit x86. This is the reason the decision to phase out the MIPS in favor of IA64:
SGI didn't have the resources to compete with other 64bit or 32bit processors and IA64 promised the resources and engineering of the largest Microprocessor manufacturer in the world. IA64 not only *promised* better performance but it promised to bring commodity pricing to the 64bit market. No longer would SGI have to swim up a raging stream to develop it's own architecture, neither would it have to shell out a lot of money to buy competing 64-bit processors.
This is roughly the same reason why HP decided to invest in IA64 as well.
[ 03-06-2002: Message edited by: johnsonfromwisconsin ]</p>
I've talked to him several times and had a few interesting discussions with him, but while his argumentations rarely fall short of logic, they always also show a great degree of stubbornness.
He's going to tell you what he thinks is right, simply bypassing the fact that often more than one possibility is correct. He's a "different thinker" and always has to contradict a public opinion, even if that opinion is actually quite viable. As long as he objects to it, and even if he's alone to object to it, he's going to argue.
However I'm glad that we had the same opinion on the Xtrem Mac back then, when everyone was euphorically drooling in anticipation of something that was never going to happen, and I and he both knew it.
That transscript can be read here:
<a href="http://www.g-news.ch/articles/xtrem.html"; target="_blank">www.g-news.ch/articles/xtrem.html</a>
G-News
Do a search for him in ArsTechnica's Battlefront to find the reasons why. He's been amply discussed.
[ 03-07-2002: Message edited by: TheAlmightyBabaramm ]</p>
<strong>
Refer to "CRISC" comment.
</strong><hr></blockquote>
So, in all honesty, you actually do believe that a "CRISC" CPU, which basically is a RISC core with additional decoding stuff slapped on, can inherently scale higher than a much less complex RISC CPU?
Has it ever occured to you that the clock speed advantage of current Intel and AMD CPUs is more related to the fact that both of them invest insane amounts of money in their designs, fabs and processes, rather than some inherent "low-frequencyness" of all RISC CPUs?
Bye,
RazzFazz
<strong>And I think those huge 10-20 stage pipelines help out in the MHz area too.</strong><hr></blockquote>
Yes Intel said that the architecture of the Pentium 4 will be able to reach 5 ghz in the future, the 20 stage pipeline help certainly.
<strong>
The way you say it, you make it sound like one can tinker with the spec source code, or write special compilers that are only designed to produce good spec binaries. But the spec benchmarks have a lot of rules that must be adhered to. First, you cannot touch the source code, you cannot optimise it in any way shape or form. You CAN use compiler optimisations, but they're only allowed as far as they improve overall application performance in general, and are not just profiled for spec sources. The main rules are:
</strong><hr></blockquote>
This discussion about SPEC is ridiculous. By your own admission, you don't think a 700mhz P3 is faster than a 1GHZ G4. Likewise, it puzzles me why you put any credibility in SPEC results.
Let me tell you a few things about SPEC.
1) Did you know that SPEC has been proven to vary by more than 30% based on compiler alone?
2) Are you aware that the gcc compiler is not what's used for production Mac apps? Are you also aware that gcc (though free) is notoriously slow?
3) How about the relevance of some of the tests? When was the last time you needed to do weather forcasting on your PC? I'll agree that some of the individual tests hold some intellectual curiousity, but the aggregate score is not meaningful.
4) You are correct about SPEC not allowing for code changes or optimizations. Of course, this is SPECs biggest downfall. What commercial product runs on multiple platforms that has no optimizations for a given platform? What about SIMD? This is the G4's biggest strenth, yet there is absolutely no way to measure this via SPEC. Intel has done some work in "auto vectorizing" in it's compilers, but it's nowhere near as effective as manual SIMD optimizations as seen in commercial software. To my knowledge, no auto vectorizing exists in current Mac compilers. Therefore, it would seem that this gives Intel's results an artificially higher score.
That said, take a look at Apple's market and the programs that are key to Apple's success. That is, what programs are Mac users most likely to use? Photoshop, "i apps" (idvd,itunes, etc.) video compression, etc. All of these programs make extensive use of SIMD (Altivec or Velocity Engine), yet are not measured in any way by SPEC. Therefore, SPEC is irrelevant for the purposes of Mac to PC comparisons.
There are many valid uses for SPEC. SPEC is better served for measuring compiler performance than it is for cross platform performance, particularly in the PC market.
Steve
<strong>So your blaming Intel for being able to outcompete SGI?</strong><hr></blockquote>
No. I'm saying there's no logical reason whatsoever for SGI to migrate onto a mainstream chip alongside a major version of MS Windows.
[quote]Originally posted by johnsonfromwisconsin:
<strong>And you blame that on Intel?</strong><hr></blockquote>
No. I blame the fact of that SGI ignored it on Intel.
[quote]Originally posted by johnsonfromwisconsin:
<strong>I find your analysis lacking for this reason:
It doesn't address the economic motivation for the likes of SGI to adopt a different microarchitecture. You talked like SGI had a lot of resources to develop MIPS with from the beginning, and Belluzo just sold out in some sort of cloak and dagger scheme. If I remember, MIPS didn't even show up in the running as far as processors in servers goes, and was only about third in scientific workstations behind IBM and Compaq.</strong><hr></blockquote>
Yes. they used to have _quite_ a bit of resorces. SGI used to be first and foremost in the scientific/graphical computing world. and do you realize what an honored position third in scientific workstations behind IBM and Compaq is? that means that the MIPS was competitive with. and surpassed _only_ by the POWER and Alpha CPUs. this means that SGI manufactured the _third fastest CPU on earth_. whereas the IA64 can not be said to hold such a title.
[quote]Originally posted by johnsonfromwisconsin:
<strong>The MIPS was never a high volume or even a great performing processor compared to other 64-bit MPU's from Compaq, IBM, or HP. MIPS is very niche even within the 64-bit market which itself is a bit of a niche and it couldn't produce processors to compete with these or even the 32-bit x86</strong><hr></blockquote>
On the contrary. what do you think has powered the N64, PS and PS2. as well as numerous printer, digital copier and network router embedded logic systems? the MIPS was(And is) one of the _most_ mainstream 64-bit CPUs on the market
[quote]Originally posted by johnsonfromwisconsin:
<strong>This is the reason the decision to phase out the MIPS in favor of IA64:
SGI didn't have the resources to compete with other 64bit or 32bit processors and IA64 promised the resources and engineering of the largest Microprocessor manufacturer in the world. IA64 not only *promised* better performance but it promised to bring commodity pricing to the 64bit market.</strong><hr></blockquote>
Intel promised a lot of things with the IA64. but they only delivered half way. for instance. the MIPS R14000 beats the Merced in integer performance(As well as nearly every other CPU on earth
And commodity pricing is right! with PA-RISC out of the picture from the start. and later MIPS. thanks to Belluzo. Alpha dead after being handed on a silver plate to Intel by their thrall Compaq and even some rumors of the POWER being surrendered by IBM for some unknown reason. I think being the ONLY major 64-bit CPU on the market other than Sun's would make practically anything you can slap together go for "Commodity pricing".
[quote]Originally posted by johnsonfromwisconsin:
<strong>No longer would SGI have to swim up a raging stream to develop it's own architecture, neither would it have to shell out a lot of money to buy competing 64-bit processors.</strong><hr></blockquote>
Neither would much stronger companies like Dell. once their distinguishing advantages are gone. SGI's customers will scatter like a flock of pheasants after a shot. and I think that all the people at SGI obviously could have figured this out.
[quote]Originally posted by johnsonfromwisconsin:
<strong>This is roughly the same reason why HP decided to invest in IA64 as well.</strong><hr></blockquote>
While HP didn't have as much to lose(They do other things than HP-UX). and they were already a big Intel buyer. HP just doesn't seem to have been in line to benefit from the IA64 much. as a sign of the IA64's _very_ late and exceedingly underwhelming arrival. the PA-RISC project has been extended another five years. and the PA-RISC is currently a very quick chip(at this moment a bit ahead of the MIPS). and I think that if all of HP's CPU design resources were still dedicated to the PA-RISC than it too would easily be faster than the Itanium.
Eric,
[ 03-08-2002: Message edited by: Eric D.V.H ]
[ 03-08-2002: Message edited by: Eric D.V.H ]
[ 03-08-2002: Message edited by: Eric D.V.H ]</p>
back in the 90s talking about an SGI Workstation, or calling "O2" would get every computer freak excited like a schoolgirl on halloween. "OMG I want one of these".
In the movie "Starship troopers" (which was made heavily on SGI), SGI even appears in the story as one HUGE world concern with dimensions of Microsoft and more, almost like the US today. everything is owned by SGI there (of course that is a joking hidden ad).
I guess they sat on their laurels a bit too long and now lost all but the movie customers.
G-News
<strong>In the movie "Starship troopers" (which was made heavily on SGI), SGI even appears in the story as one HUGE world concern with dimensions of Microsoft and more, almost like the US today. everything is owned by SGI there (of course that is a joking hidden ad).</strong><hr></blockquote>
There was a similar ad for a future SGI workstation(That bore a strong resemblance to the O2) in the film "Lost in Space" too.
Eric,