A question about chip speeds

imac davidimac david
Posted:
in Current Mac Hardware edited January 2014
Hi,



apologies if this has been covered before, but I did a search and didn't find the answer so hence this thread.



Can anyone explain to a non-techie (ie me) why it is that chip speed development of the G4 lags so far behind the PC's?



It seems that every few months the PC people announce another 300 or so Mhz, and yet the G4 speed increases are far more infrequesnt.



Is it because there are more PC chip manufacturers out there who are competing with each other? Is there something about the architecture used that enables speed increases? The new 13 micron manufacturing ability has also allowed speed increases apparently for the PC side, so why not to the Mac side?



Please don't reply along the lines of 'chip speed doesn't matter, it's real life use and 800Mhz is fine' because a) I know, and there have been other threads covering that; b) my new (and first) LCD iMac IS quicker for me that my ex PC with a higher clock speed and c) I just really want to know if anyone can explain why Mac chip speeds are advanced more slowly than the PC side.



Cheers!



David



PS and for any doubters out there, I LOVE my new iMac. For the first time I can approach my computer with photographs to scan, film to edit WITHOUT experiencing fear and dread that what should be a 30 minute project will take all evening due to system conflicts, TWAIN drivers messing up etc etc

Comments

  • Reply 1 of 15
    digitalandresdigitalandres Posts: 115member
    I think it's a FLOPs issue. Essentially, Mac Mhz's do not equal PC Mhz's. Like you pointed out yourself, a Mac with say 500Mhz runs MUCH faster than a PC running 500Mhz.



    That's about as much as I can contribute without having to make things up to fill the gaps in my knowlege. Im sure someone else can throw out an actual explanation.
  • Reply 2 of 15
    jet powersjet powers Posts: 288member
    I'll be general, as I'm no techie.



    1) R&D. Intel and AMD spend a lot of money making their chips faster, and it shows. IBM and Motorola have a more diversified business, and do proc development on a smaller scale.



    2) Pipeline depth. Someone else will come along and explain why longer pipelines allow for higher MHz ratings. Basically, you can just fill the pipelines with more stuff. The trade-off is that if you fill the pipeline with the wrong stuff, it needs to flush before it can continue. This is an evolution of the old CISC vs. RISC debate, which is a whole 'nother ball of wax.



    3) Focus. Intel and AMD chips are designed for exactly one thing: to sell computers. Motorola's chips ned to be sold in many markets, so their designs reflect serving those markets. I'm not sure why IBM keeps it's hand in the game with the 750FX, but my iBook is glad they did.



    Jet
  • Reply 3 of 15
    rodukroduk Posts: 706member
    I think point three above (focus) is an important one. Motorola have already announced that they are concentrating on the embedded processor market. I'm sure a 1GHz G4 is already more than powerful enough for its intended purpose.



    [ 08-08-2002: Message edited by: RodUK ]</p>
  • Reply 4 of 15
    zapchudzapchud Posts: 844member
    1) Harder competition between x86-compatible processors



    2) Motorola had a huge, long stall in MHz-development (problems or incompetence, I dunno )



    3) x86-processors uses an architecture that more easely adds to MHz. They use more steps (longer pipeline) to execute an instruction, in very short, the G4 needs 7 steps to execute an instruction, and a Pentium 4 needs 20. Lets say both processors executes 100 million instructions pr second (MIPS), the G4 will then have 700MHz, and the Pentium 4 2000MHz.

    This is a very short, quick and undetailed explanation.



    4) x86-processors uses more power (watts).



    5) The (current) G4 is not manufactured with the .13 micron process (but the G3's in the iBooks are)



    [ 08-08-2002: Message edited by: r-0X#Zapchud ]</p>
  • Reply 5 of 15
    ghost_user_nameghost_user_name Posts: 22,667member
    PLEASE take some time to read ArsTechnica's two articles comparing the P4 vs G4e processors. They will answer many if not all of your questions regarding the old MHz myth.



    <a href="http://arstechnica.com/cpu/01q2/p4andg4e/p4andg4e-1.html"; target="_blank">The Pentium 4 and the G4e: an Architectural Comparison</a>



    <a href="http://arstechnica.com/cpu/01q4/p4andg4e2/p4andg4e2-1.html"; target="_blank">The Pentium 4 and the G4e: an Architectural Comparison - Part II: The Execution Core</a>







    This image represents the pipelines of the P4 and G4, respectively. What does it mean??? Read the article above and you'll find out!!!



  • Reply 6 of 15
    imac davidimac david Posts: 286member
    Brad,



    thanks for the articles - interesting reading though a bit beyond a non-techie like me.



    BUT, my question isn't about the "Mhz myth". I've already said that my new iMac with a chip running a lower Mhz than the PC it replaced, is as fast if not faster. As a non-techie I can accept a general concept such as 'a different methodolgy is used, with fewer exectuable steps, means the same output with slower speed'. A bit like gears on a bike - people can ride at the same speed though have their legs moving at very different speeds depending on which gear they've chosen.



    But if we agree that a G4 running at 1Ghz is faster than one running at 800Mhz and is desirable for improving the iMacs performance, then my question is:



    At a given point in time chip makers have their current offering. Why is it that Intel, AMD are able in months to increase their product's speed, whilst Motorola can't?



    Competition may be a factor, but don't IBM and Motorola compete to provide chips to Apple?



    If the articles did provide a reason why the architecture of P4's allow rapid increases in speed, then I apologise for not being able to read it



    Cheers!



    David
  • Reply 7 of 15
    rodukroduk Posts: 706member
    [quote]Originally posted by iMac David:

    <strong>

    At a given point in time chip makers have their current offering. Why is it that Intel, AMD are able in months to increase their product's speed, whilst Motorola can't?

    </strong><hr></blockquote>



    Again, I think this partly comes down to focus. Perhaps its not just a question of can't, but why bother. For the embedded processor market (network routers, mobile phones etc), Motorolas main focus, their chips are probably already fast enough. Its probably more beneficial for Motorola to improve other aspects of their chips, for example, reduce the heat produced, lower the cost etc rather than increase the speed.



    [ 08-09-2002: Message edited by: RodUK ]</p>
  • Reply 8 of 15
    zapchudzapchud Posts: 844member
    When you have a long pipeline, you'll easier get more MHz, or socalled "speed" BECAUSE one step in the pipeline represents one Hz. To execute an entire operation, you need more Hz when you have a long pipeline to execute it in the same time as a processor with a short pipeline. Since it is "no problem" to create a processor that is as fast, you'll automatically get more MHz because the pipeline is longer (has more steps).
  • Reply 9 of 15
    imac davidimac david Posts: 286member
    [quote]Originally posted by r-0X#Zapchud:

    <strong>When you have a long pipeline, you'll easier get more MHz, or socalled "speed" BECAUSE one step in the pipeline represents one Hz. </strong><hr></blockquote>



    Are you sure? My understanding was that 1 Hz represents one clock cycle per second - ie a measurement of the state of a 0 or 1. The more often a computer can check that the higher the Hz number. So if a computer can check 10,000 times a second it is said to have a 10kHz clock speed. The fact that a longer pipeline is in one chip means that for equivalent number of cycles the chip with the longer pipeline can do less.



    So a chip that doesn't change in pipeline length can get quicker if problems such as heat dissapation, timing etc can be removed.



    I tend to go with RodUK and the fact that Motorola have little incentive to improve, unlike the Intels of this world.



    Regards,



    David
  • Reply 10 of 15
    rodukroduk Posts: 706member
    I think your question also highlights how Apple and Motorola have different visions for the future of the PC. Its been suggested that PCs will eventually die out as every device will have its own processor and "intelligence". Apple are trying to prevent this from happening by reinventing the PC as a digital hub (closely followed by Microsoft). Motorola are in the process of jumping ship. For every processor Motorola sell to Apple for inclusion in a computer, they can probably already sell a thousand others to companies for inclusion in other devices.



    [ 08-09-2002: Message edited by: RodUK ]</p>
  • Reply 11 of 15
    amorphamorph Posts: 7,112member
    [quote]Originally posted by iMac David:

    <strong>



    I tend to go with RodUK and the fact that Motorola have little incentive to improve, unlike the Intels of this world.

    </strong><hr></blockquote>



    Mot had and continue to have competition in the space they are targeting (from ARM and MIPS and IBM and for that matter Intel), so lack of competition isn't the answer. They've had a lot of problems, but the chief reason their chip is so different is one of focus: You can boost clock speed with deep pipelines, which (for various reasons) results in big, complex chips, and you can boost it by consuming a lot of power. Mot doesn't want to do either. It's not in their interest to.



    So Mot decided to stay with relatively elegant designs that don't use much power. It's better suited to their market, and frankly it's better suited to Apple in many ways. You can have a G4 stuffed into your iMac (or into a TiBook, or an XServe) because Mot focused their designers (who are, historically, some of the best in the industry) on keeping it small and cool relative to its computational power.



    Intel, on the other hand, focused on MHz at all costs, and so they have big, hot, complex, expensive processors that run very fast, but limit the design of the machines they can be run in.



    It's not a matter of Mot being worse. Intel has tripped over their own feet as well (I think it was the Pentium II where they pulled a G4? Or heck, just look at the Itanium's history). Mot has had to deal with near-bankruptcy and drastic cost-cutting measures, but the main reason their chips are so different is because of conscious, deliberate and intelligent design decisions.
  • Reply 12 of 15
    macaddictmacaddict Posts: 1,055member
    [quote]I think it was the Pentium II where they pulled a G4? Or heck, just look at the Itanium's history<hr></blockquote>



    Don't forget the Coppermine 1.13GHz



    BTW, I don't think Intel's chips are that expensive. Just a huge markup. Look at AMD...cheap cheap cheap!
  • Reply 13 of 15
    pyr3pyr3 Posts: 946member
    [quote]Originally posted by Brad:

    <strong>PLEASE take some time to read ArsTechnica's two articles comparing the P4 vs G4e processors. They will answer many if not all of your questions regarding the old MHz myth.



    <a href="http://arstechnica.com/cpu/01q2/p4andg4e/p4andg4e-1.html"; target="_blank">The Pentium 4 and the G4e: an Architectural Comparison</a>



    <a href="http://arstechnica.com/cpu/01q4/p4andg4e2/p4andg4e2-1.html"; target="_blank">The Pentium 4 and the G4e: an Architectural Comparison - Part II: The Execution Core</a>







    This image represents the pipelines of the P4 and G4, respectively. What does it mean??? Read the article above and you'll find out!!!



    </strong><hr></blockquote>



    The only thing that they forget to mention in these 'MHZ myth' papers is branch prediction. Nowadays the branch prediction algorithms are like 99.99% accurate at predicting whether or not a branch will branch or not.
  • Reply 14 of 15
    amorphamorph Posts: 7,112member
    [quote]Originally posted by pyr3:

    <strong>The only thing that they forget to mention in these 'MHZ myth' papers is branch prediction. Nowadays the branch prediction algorithms are like 99.99% accurate at predicting whether or not a branch will branch or not.</strong><hr></blockquote>



    They don't mention it because it's a necessity, not an accomplishment.



    Caches have to hit over 99.9% in order to be effective, and so to BPUs (for long pipelines.



    99.99% success is failure once every 10,000 times. That sounds rare, but remember that a P4 is churning through billions of ops per second. A higher rate of failure would be lethal.
  • Reply 15 of 15
    eskimoeskimo Posts: 474member
    [quote]Originally posted by pyr3:

    <strong>



    The only thing that they forget to mention in these 'MHZ myth' papers is branch prediction. Nowadays the branch prediction algorithms are like 99.99% accurate at predicting whether or not a branch will branch or not.</strong><hr></blockquote>



    Actually the practical limit of BPU accuracy is about 99.5% for the level of complexity that is economical to put in a desktop processor.
Sign In or Register to comment.