Intel at 4 GHz??? Come on Motorola...

1234689

Comments

  • Reply 101 of 170
    Just a little perpective on the RISC versus CISC wars...



    In the late 80s and early 90s, Sparc and MIPS were tearing up the place and the industry concensus was that it was simply impossible for CISC to scale. Billions and billions of dollars were based on this assumption -- DEC developed Alpha as a workstation/PC-class chip, Microsoft rewrote OS/2 to become portable (this became NT/XP), and IBM and Motorola teamed up to build a chip that would eat Intel's lunch in the mainstream desktop computer market (the PowerPC). The charts that showed Intel leveling off and RISC shooting upward were certainly real.



    Apple was suffering for a similar "Mhz gap" at the time, and was looking for a new chip that would both scale and have large economies of scale. The PowerPC looked like the right horse to climb onto -- the design was good, and when PPC machines running NT and OS/2 took over the corporate markets, Apple could get the benifits. For a while it seemed like it was working -- Moto was producing 300Mhz consumer chips while Intel was stuck at 166Mhz.



    Then in 95-96, the Pentium Pro (P6) shipped, using a hybrid-RISC design without sacrificing any compatibility. For a brief period of time, it had the fastest SPEC score in the industry.



    This lead to a collective OH SHIT! momemt. Why were all these companies spending billions in chip development when they could just buy kick-ass CPUs from Intel for cheap and get the _real_ economies of scale? Intel says "You think that's good? Wait until you see our upcoming 64-bit CPU with super VLIW technology! It will destroy RISC." Even Sun signed on for a while.



    I really think that the management decision to kill Alpha and MIPS was made back in 1996, they just had to wait it out until Intel got their chip out. (Which, assuming the next version is finally the one that counts, will be about 4 years late.) HP of course had been in on IA64 from the beginning -- seeing the writing on the wall for proprietary RISC back in 93 or so. Just in the last couple years we've finally seen this plan enacted with Alpha and PA-RISC finally being shut down.



    Meanwhile, those Windows PPC machines sold so poorly that most people forgot they were ever made. IBM moved back to server chips, and Moto went embedded. Apple's plan had completely backfired -- they ended up being just about the only customer for desktop PowerPC. They're in a worse situation than in the twilight days of the 68K.



    The bottom-line is that writing is on the wall for PowerPC in Macs. A single customer CPU is just not sustainable. There might be a G5, but there probably won't be a G6, and the replacement will probably be either Intel or AMD. There just aren't any other options out there (except Sparc, but that's no great performer either). x86 everywhere is a bleak future, I agree.



    My guess is that the only reason that Apple hasn't already gotten off the sinking PPC ship is because of OS X. The customer base just can't handle more than one major transition every few years. Not to mention that the conversion will be easier if Classic is minimized as much as possible.
  • Reply 102 of 170
    g-newsg-news Posts: 1,107member
    First of all I have to say that I don't think PPC is a sinking ship. Maybe Moto is, but PPC isn't.



    And then the Pentium Pro was everything else than cheap when it was released, in fact it was one of the most expensive desktop chips in history.



    G-News
  • Reply 103 of 170
    outsideroutsider Posts: 6,008member
    Boy, IntlHarvester , thats the best sensationalist, pessimistic, fluff, anti-PPC rant yet. How did you come to that warped conclution from the past you described? IBM will stop using PPC then too? Have you told them?
  • Reply 104 of 170
    programmerprogrammer Posts: 3,457member
    The problem with SPEC is that they compile and run the same source code on all the machines. While this might seem fair, the sad fact of the matter is that slight changes to the source code can make a dramatic difference to each processor. Compilers just aren't that good -- even Intel's much vaunted compiler leaves a lot to be desired.



    Several times I have had a piece of C code that I have carefully optimized for the PC, tweaked to get the best sequence of instructions out of it. When compiled for the PowerPC (this was a few years ago so the 604 was the particular target) it generate okay but not great code. I rearranged things a little (i.e. seperate variables instead of arrays, rearranging loops, and a few other things), the PowerPC suddenly ran 30% faster... but the x86 ran 30% slower. And this was a piece of code that is probably quite like the SPECfp benchmarking code.



    Also, in my experience the GCC compiler does a lousy job of taking advantage of machines with lots of registers (i.e. MIPS & PowerPC), giving the advantage to the x86 world. It is that way for a good reason -- most of the people using GCC have x86 machines!



    In practice if somebody is actually trying to make a critical piece of floating point code go as fast as possible on the target machine, then the G4 & PIII come out about even or even with the G4 a little faster. In heavily mixed code where there is a balance of integer & float, the G4 comes out a little better as well.



    <img src="graemlins/oyvey.gif" border="0" alt="[No]" /> Its not a happy state of affairs, but unless Apple & Moto can make some serious improvements to GCC, then that's what we're stuck with.
  • Reply 105 of 170
    [quote]And then the Pentium Pro was everything else than cheap when it was released, in fact it was one of the most expensive desktop chips in history.

    <hr></blockquote>



    IIRC, it was still quite a bit cheaper than the Alpha's and MIPS targetted to squash the x86.
  • Reply 106 of 170
    amorphamorph Posts: 7,112member
    [quote]Originally posted by Programmer:

    <strong>

    <img src="graemlins/oyvey.gif" border="0" alt="[No]" /> Its not a happy state of affairs, but unless Apple & Moto can make some serious improvements to GCC, then that's what we're stuck with.</strong><hr></blockquote>



    Fortunately, that appears to be what Apple is doing as we speak. gcc 3.x should be interesting.
  • Reply 107 of 170
    g-newsg-news Posts: 1,107member
    both alpha and the MIPS were never desktop CPUs.

    (desktop as in "desktop below 5000$")



    The Pentium Pro was just the first market research for the P2, a prototype, if you want.



    G-News
  • Reply 108 of 170
    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>SPARC? &lt;snicker&gt; Got uncontested proof?</strong><hr></blockquote>



    Not quite. but SPEC'll have to do.



    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>Because I call bullshit. SPARC sucks in FP compared to IA64 processors.</strong><hr></blockquote>



    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>One thing's for sure...barring outliers and cheating, IA64 systems are probably always going to crush anything Sun can turn out in FP with SPARC, comparable systems wise. It's simply a vastly superior architecture for FP.</strong><hr></blockquote>



    [quote]Originally posted by johnsonfromwisconsin:

    <strong>I have little faith in Sun's ability to put out world class individual MPU's.</strong><hr></blockquote>



    I was surprised too. but it turns out that the UltraSPARC III Cu really is an incredibly fast chip:



    800Mhz Itanium:

    SPECint2000: 342

    SPECfp2000: 701



    1050Mhz UltraSPARC III Cu:

    SPECint2000: 610

    SPECfp2000: 827



    Congratulations Scott! Bring out that UltraSPARC IV and show em' what your made of.



    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>At any rate, "McKinley" is expected to have competitive SpecINT 2K performance, and possibly best in SpecFP 2K, at least until EV7 systems show up. It's unknown which will roll out first.</strong><hr></blockquote>



    I doubt it. and even if McKinley _DOES_ ship early(Which I doubt). the POWER5, UltraSPARC IV and Alpha EV7(And maybe even the new PA-RISC and MIPS CPUs) will not be far behind. so as to pound it into submission.



    [quote]Originally posted by johnsonfromwisconsin:

    <strong>I don't know where your hearing the Power5 stuff from, but the Power4 isn't even out in force yet.</strong><hr></blockquote>



    HmmmÉ it looks like the POWER5 is set to come out in early 2003. by which time the POWER4 will have reached about 1.63Ghz. or so IBM says.





    Eric,
  • Reply 109 of 170
    [quote]HmmmÉ it looks like the POWER5 is set to come out in early 2003. by which time the POWER4 will have reached about 1.63Ghz. or so IBM says.

    <hr></blockquote>



    Interesting. Any link to that? I'm honestly interested. I have seen no mention of the Power5 from the likes of Paul Demone or on the first couple pages of a Google search. In any event, any chip IBM releases under the name Power5 is not likely to be any sort of generational leap over the Power4 (probably just a process shrink). McKinnley is clearly a generation ahead of Itanium. It will reach higher clockspeeds with a shorter pipeline on the same old .18u process



    here contains some details on estimated Spec Scores on McKinnley versus Power4



    <a href="http://www.realworldtech.com/page.cfm?section=columns&AID=RWT031401225852"; target="_blank">http://www.realworldtech.com/page.cfm?section=columns&AID=RWT031401225852</A>;



    [quote]I was surprised too. but it turns out that the UltraSPARC III Cu really is an incredibly fast chip:



    800Mhz Itanium:

    SPECint2000: 342

    SPECfp2000: 701



    1050Mhz UltraSPARC III Cu:

    SPECint2000: 610

    SPECfp2000: 827



    Congratulations Scott! Bring out that UltraSPARC IV and show em' what your made of.



    <hr></blockquote>



    Of course, there is a lot of contraversy in Sun's latest scores. It looks like they've made a compiler made for the exact data patterns presented by the SPEC benchmarks which aren't against Spec Rules. You may see other vendors release spec scores with such adjustments or a change in rules.



  • Reply 110 of 170
    spookyspooky Posts: 504member
    what about this:



    Power Mac 'is no supercomputer'



    An independent computer performance tester has concluded that Apple's Power Mac 'is no supercomputer', contrary to claims that Apple has consistently made.

    Using a series of tests - designed to be as platform independent as possible - the Standard Performance Evaluation Corporation (SPEC) concluded that compared to a 1GHz Pentium processor, the Motorola Power PC 1GHz processsor currently employed in Apple's top Power Mac, 'is far less suited for scientific applications', despite the fact that the Power PC FPU, 'with its 32 registers ought to have been superior to the x86 [Intel Pentium] FPU with its antiquated stack structure and eight registers only.'



    On the bright side for Apple, the test did show that 1GHz PowerPC gives a greater performance increase over the 867MHz version than mere clock speed would indicate, suggesting that Motorola is able to squeeze extra performance out of its processors beyond simply adding megahertz.



    According to SPEC, the eight per cent increase in clock speed for the 867 over the 800MHz processor delivered only a seven per cent increase in performance. However, the 15.5 per cent increase from 867MHz to 1GHz delivered an 18.6 per cent performance hike. Further tests pushed this up to around 27 per cent. This improvement was most notable when memory throughput was high due to the improved processor cache.



    SPEC's findings refute Apple's claim that the Power Mac is a supercomputer, though as both use different definitions, who is right and who is wrong is a moot point. Where SPEC relies on its own benchmarks, Apple counts the number of floating point operations performed every second. In its own words, 'What makes a supercomputer "super" is its ability to execute at least one billion floating-point operations per second, a staggering measure of speed known as a "gigaflop." The PowerPC G4, architected by Apple, Motorola and IBM, was the first microprocessor to deliver a sustained performance of over one gigaflop, thereby winning renown as the first supercomputer on a chip.'



    As Apple CEO Steve Jobs is keen to demonstrate as often as possible, the Power Mac visibly outperforms Pentium-based machines in what for Apple is the key processor intensive application - Adobe Photoshop. What SPEC's findings appear to show is in general the Power Mac may not bear up so well.



    A description and explanation of the tests used is available here.



    from MacUser UK 11/3/02
  • Reply 111 of 170
    [quote]Originally posted by Outsider:

    <strong>Boy, IntlHarvester , thats the best sensationalist, pessimistic, fluff, anti-PPC rant yet. How did you come to that warped conclution from the past you described?</strong><hr></blockquote>



    Alcohol helped!



    "IBM will stop using PPC then too? Have you told them?"



    They pretty much already have in desktop space, right? The Power4 server stuff is too expensive for Macs and they don't make very many.



    The point of all that is that chipmaking requires $billions in investments, and strategy shifts take 4 or 5 years to complete. I'm not at all anti-PPC -- just pointing out the straight facts that the desktop chip was not as popular as originally envisioned and is being abandoned by it's creators. Apple's eventually going to have to switch.
  • Reply 112 of 170
    Remember the c't article says that the specint PIII score was 236 using MS visual Studio. Specint scores using intel's special compiler get specint scores of 454 for a PIII 1Ghz. The real world uses MS visual studio. Intels compiler gets Specint scores of 636 for a PIV 2Ghz. If the same ratio holds true for MSVS you would expect a 2Ghz pentium to actually score 330 on Specint 2000. Compare this to the 309 score of a single processor G4 1Ghz. Then remember this score is using gcc which has a very imature PPC code branch. From what I have read PPC gcc contributers have had a tough time getting their contributions into the codebase.



    Not so different after all. Search the web for Spec scores using MSVS and I find no scores. Real x86 compilers not spec tweaked show that the P4 is only narrowly faster than a G4 running at half the speed.
  • Reply 113 of 170
    airslufairsluf Posts: 1,861member
  • Reply 114 of 170
    [quote]Where do people find the time to write these posts that could choke an elephant? And so many? Guy's agree to disagree, the horse ain't moving anymore. Let's all move into some fresh material here.<hr></blockquote>



    My advice to you, if you don't enjoy a thread, find another. I am enjoying my conversation with Eric immensly even if we don't see eye to eye.



  • Reply 115 of 170
    [quote]Originally posted by IntlHarvester:

    <strong>Just a little perpective on the RISC versus CISC wars...</strong><hr></blockquote>



    A little common perspective that is? allow me to tell you what they don't want you to know.



    [quote]Originally posted by IntlHarvester:

    <strong>In the late 80s and early 90s, Sparc and MIPS were tearing up the place and the industry concensus was that it was simply impossible for CISC to scale. Billions and billions of dollars were based on this assumption -- DEC developed Alpha as a workstation/PC-class chip, Microsoft rewrote OS/2 to become portable (this became NT/XP), and IBM and Motorola teamed up to build a chip that would eat Intel's lunch in the mainstream desktop computer market (the PowerPC). The charts that showed Intel leveling off and RISC shooting upward were certainly real.</strong><hr></blockquote>



    Yup.



    [quote]Originally posted by IntlHarvester:

    <strong>Apple was suffering for a similar "Mhz gap" at the time, and was looking for a new chip that would both scale and have large economies of scale. The PowerPC looked like the right horse to climb onto -- the design was good, and when PPC machines running NT and OS/2 took over the corporate markets, Apple could get the benifits. For a while it seemed like it was working -- Moto was producing 300Mhz consumer chips while Intel was stuck at 166Mhz.</strong><hr></blockquote>



    Greatly incorrect. Here's where the PowerPC came from:



    Once upon a time. Motorola was continuing the development of their 680x0 CISC architecture as the new Motorola 80000. an advanced RISC based design. Apple(As usual) was one of their prime motivations in doing this. now for some reason. Apple decided that Motorola needed someone to help them out with the 80k.



    Instead of someone more logical. like DEC(With whom Apple was already working on the then new StrongARM), Acorn(That's the "A" in "ARM" in case you didn't know), MIPS, AT&T(From whom Apple had already bought DSPs for the Quadra 630), Hitachi or someone like that. Apple decided to choose their Evil(With a capital E!) arch nemesis IBM. now I know you and I wouldn't do something so silly. but Apple did. I think it was because that jerk Scully(Correct me if he didn't work there then) was trying to buddy up with IBM. so that they would buy Apple and make him and his evil cohorts a fortune in dirty stock money.



    But in any case. Apple decided that Motorola would blend their new 80k chip with IBM's POWER. thusly making? the PowerPC. and the PPC version of Windows NT wasn't some free-floating idea of Microsoft and IBM's. it was something for Apple's new CHRP(Also known as PPCP) computer. the only non Apple machines that would run Win NT on a PPC were going to be Apple licenced clones. which would also run Solaris, AIX, OS/2 and of course the Macintosh System Software. CHRP was actually ready to ship. except? the only reason CHRP failed is the same as OpenDoc. Apple. for no reason whatsoever. sayed "No".



    [quote]Originally posted by IntlHarvester:

    <strong>Then in 95-96, the Pentium Pro (P6) shipped, using a hybrid-RISC design without sacrificing any compatibility. For a brief period of time, it had the fastest SPEC score in the industry.</strong><hr></blockquote>



    Gee. I wonder why the Pentium pro/Pentium 2 weren't total pieces of #$*& like _every other product_ of Intel's? could it be that they resorted to that age old tactic so often employed by Microsoft, Intel and IBM? I think so.



    Since about 1993. DEC had signed Intel into a chip making pact. in which DEC would use some of Intel's chip fab-ing plants to make DEC Alpha's. however. it appears that Intel couldn't restrain themselves. and started to take advantege of the fact of that intricate designes for one of the most sophisticated CPUs in the world were flowing right under their noses. but. after looking at the new Pentium Pro. DEC engineers thought there was something fishy going on. and eventually managed to catch Intel with their hand _deep_ in the cookie jar. In May 12 of 1997. DEC <a href="http://www.courttv.com/legaldocs/cyberlaw/digital.html"; target="_blank">sued</a> Intel in federal court. saying that Intel was violating a number of Digital patents with its Pentium Pro and Pentium II product lines. in other words. Intel swiped their sophisticated CPU design from DEC. makers of the world's fastrest CPU. this little tale has even wider ramifications farther down too.



    [quote]Originally posted by IntlHarvester:

    <strong>This lead to a collective OH SHIT! momemt. Why were all these companies spending billions in chip development when they could just buy kick-ass CPUs from Intel for cheap and get the _real_ economies of scale? Intel says "You think that's good? Wait until you see our upcoming 64-bit CPU with super VLIW technology! It will destroy RISC." Even Sun signed on for a while.</strong><hr></blockquote>



    "Super VLIW technology"!? ha ha ha ha *Ack* *Snort* <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> . VLIW is probably one of the most laughed at computing designs there is. the design. made famous(Or infamous) by "EPIC" in the late 80s. was widely known as one of the biggest flops in CPU design ever. I don't think an engineer would have been able to mention VLIW in a positive light back then and leave the room with their pride intact.



    [quote]Originally posted by IntlHarvester:

    <strong>I really think that the management decision to kill Alpha and MIPS was made back in 1996, they just had to wait it out until Intel got their chip out. (Which, assuming the next version is finally the one that counts, will be about 4 years late.) HP of course had been in on IA64 from the beginning -- seeing the writing on the wall for proprietary RISC back in 93 or so. Just in the last couple years we've finally seen this plan enacted with Alpha and PA-RISC finally being shut down.</strong><hr></blockquote>



    Doubt it. I think I already explained that colluding traitor Belluzzo's involvment in tilting the odds against HP and MIPS/SGI. as for DEC. the financial collapse. purchase by Compaq. and eventual conquest by Intel of DEC is a long and winding tale. picking up from when DEC sued Intel:



    Soon after hearing of DEC's suit. Intel filed a countersuit alleging that DEC had stolen some of _Intel's_ information while under contract. just as the two giants were winding up for a terrific legal battle. out of the blue. DEC and Intel made an absolutely bizzare out of court settlement. in it. DEC was to surrender nearly all of it's Alpha development centers, every single one of it's chip fab-ing facilities and all of it's involvement in the design, production, distribution and marketing of the StrongARM to Intel. the only thing Intel had to give in return was the sum of $700M. DEC soon faltered. and was purchased by Compaq. which led to the Ultimate demise of the Alpha.



    [quote]Originally posted by IntlHarvester:

    <strong>Meanwhile, those Windows PPC machines sold so poorly that most people forgot they were ever made. IBM moved back to server chips, and Moto went embedded. Apple's plan had completely backfired -- they ended up being just about the only customer for desktop PowerPC. They're in a worse situation than in the twilight days of the 68K.</strong><hr></blockquote>



    What "Windows PPC machines"? the _only_ PPCs that were supposed to run Windows NT were Apple licenced CHRP systems. they sold so poorly because they _didn't exist_.



    [quote]Originally posted by IntlHarvester:

    <strong>The bottom-line is that writing is on the wall for PowerPC in Macs. A single customer CPU is just not sustainable. There might be a G5, but there probably won't be a G6, and the replacement will probably be either Intel or AMD. There just aren't any other options out there (except Sparc, but that's no great performer either). x86 everywhere is a bleak future, I agree.</strong><hr></blockquote>



    [quote]Originally posted by IntlHarvester:

    <strong>My guess is that the only reason that Apple hasn't already gotten off the sinking PPC ship is because of OS X. The customer base just can't handle more than one major transition every few years. Not to mention that the conversion will be easier if Classic is minimized as much as possible.</strong><hr></blockquote>



    No. it's because. like I already said about SGI. they'd get eaten alive. the G4 is a slightly better than equivelant chip to the latest +2Ghz 80x06 chips. and the G5 will be the _only_ consumer priced 64-bit PC chip on earth. when 64-bit addressing in OS X. as well as 48-64-bit synthisizers and digitizers in the Mac's A/V hardware get rolled in. Apple will have a significant lead over it's competitors.





    Eric,
  • Reply 116 of 170
    [quote]Originally posted by johnsonfromwisconsin:

    <strong>Interesting. Any link to that? I'm honestly interested. I have seen no mention of the Power5 from the likes of Paul Demone or on the first couple pages of a Google search.</strong><hr></blockquote>



    Okay. here's some scraps I managed to pull together:



    [quote]From <a href="http://www.theregister.co.uk/content/53/24321.html"; target="_blank">a release</a> on The Register:

    <strong>These releases seem to be timed with future Power4-II servers due in late 2002 and Power5 servers, due sometime in the second half of 2003.</strong><hr></blockquote>



    [quote]Originally from <a href="http://www-3.ibm.com/chips/news/1999/microprocessor99.pdf"; target="_blank">an issue</a> of the Microprocessor Report:

    <strong>IBM is planning for just that. Its roadmap shows frequency increases of 25% every year,with performance growing at three times that rate before jumping dramatically with the mid-decade introduction of a new Power5 design.</strong><hr></blockquote>



    [quote]Originally posted by johnsonfromwisconsin:

    <strong>Of course, there is a lot of contraversy in Sun's latest scores. It looks like they've made a compiler made for the exact data patterns presented by the SPEC benchmarks which aren't against Spec Rules. You may see other vendors release spec scores with such adjustments or a change in rules.</strong><hr></blockquote>



    Hey. I can't afford to be picky when I don't have access to a $10,000 Blade server . let's just say there's a fairly good chance that Sun has made a _really_ fast chip.



    Eric,
  • Reply 117 of 170
    [quote]Originally posted by G-News:

    <strong>both alpha and the MIPS were never desktop CPUs.

    (desktop as in "desktop below 5000$")</strong><hr></blockquote>



    Wrong. prior to that whole "Intel gets caught bootlegging DEC technology and muscles them into coughing up practically all of their material possessions for a week's worth of Intel's profits" thing. DEC was actually on the brink of releasing an "AlphaPC"(Also known as the 21164PC).



    It worked by using cutting edge software emulation(Which DEC called binary code translation) combined with the Alpha's incredible native speed to run 80x86 code(Read: Microsoft Windows/DOS executables) faster than a _real_ 80x86 chip. as well as running native Alpha code at it's usual blinding speed. this would have put a _major_ kink in Intel's napoleonic plans of conquest in the PC industry. as in addition to being a lot faster than any other CPU on the market. the Alpha(Like nearly every other RISC chip on earth) would actually be cheaper than an 80x86 if produced in equivalent volumes.



    Eric,
  • Reply 118 of 170
    [quote]Originally posted by johnsonfromwisconsin:

    <strong>



    My advice to you, if you don't enjoy a thread, find another. I am enjoying my conversation with Eric immensly even if we don't see eye to eye.



    </strong><hr></blockquote>



    Me too. although I plan to correct that last part?



    Eric,
  • Reply 119 of 170
    rickagrickag Posts: 1,626member
    Eric D.V.H

    Thank you for the history lesson, very interesting stuff.



    [quote]Eric D.V.H

    "and the G5 will be the _only_ consumer priced 64-bit PC chip on earth. when 64-bit addressing in OS X. as well as 48-64-bit synthisizers and digitizers in the Mac's A/V hardware get rolled in. Apple will have a significant lead over it's competitors."<hr></blockquote>



    My question is when?



    It now appears that any G5 for Apple may be a long way off, as Motorola and Apple keep saying the G4 has plenty of room to grow. And the G5 at least initially will in all likelyhood be 32 bit.



    By the time a 64 bit G5 hits the market where will Intel's and AMD's 64 bit products be? Already in selling machines, either enterprise servers and/or desktops, ?
  • Reply 120 of 170
    g-newsg-news Posts: 1,107member
    they already are in enterprise servers (itanium).



    The G5 is a big chance for apple to make a lot of past mistakes good. Somehow I feel they're going to **** up again.



    G-news
Sign In or Register to comment.