AMD's 45nm technology compared against Intel's
in [Chips]
|
From: Del Cecchi on 22 Dec 2008 23:45 Robert Myers wrote: > On Dec 22, 12:15 am, Del Cecchi` <dcecchinos...(a)att.net> wrote: > >>Robert Myers wrote: > > >>>Intel has done many things over the years (keeping the controller off >>>the die, killing Alpha, sticking with a front-side bus, NetBurst, >>>hyperthreading, not using SOI) that have occasioned critical comment. >>>There is always information missing from those discussions, which is >>>that Intel almost always has good business reasons for doing what it >>>does. Its judgment may be faulty, but the key is that *you do not >>>have access to those reasons*. >> >>"It seemed like a good idea at the time" in some cases clearly turned >>out not to be so good. Examples range from IBM's FS to their billion >>dollars worth of X-ray machine in East Fishkill. >> >>Likewise the business reasons Intel had may or may not have been "good" >>in the eye of a dispassionate observer. >> > > Dispassionate observers of SOI for x86 (or of Intel) don't work for > IBM. I don't work for either, have never worked for either, and I > don't own stock in either, either directly or indirectly. > > I'd *love* to know how Intel made some of its decisions. The only one > that's process-related is how Intel managed to be caught so flat- > footed at 90nm. Even there, the real question is why they didn't > abandon NetBurst sooner than they did. What kinds of lies were they > telling themselves? What did they know and when did they know it? > > Those are, to me, really interesting questions to ask, but they don't > lend themselves to the soccer stadium hooliganism that has so often > passed for discussion, and only rarely do we get to hear anyone who > actually knows anything speak to them. > >> >>>The one piece of information that is available (stock price) indicates >>>that, whatever missteps Intel may have made, it's business judgment >>>(as judged by markets) has proven to be superior to AMD's. The >>>situation with AMD has become so dire that it almost seems pointless >>>to talk about it, although there may be someone out there with >>>business judgment much better than I possess to see how a viable >>>enterprise can be created in the future. >> >>AMD has been a minor portion of the market and therefore at a cost >>disadvantage for years. Couple that with scratching to survive and it >>can lead to misteps. Monopoly level market share makes up for a lot of >>sins. >> > > You could just as well say that Intel has suffered for its size, and > it has, as I think IBM has, in the past, paid for its size. Intel is > famous for being able to replicate manufacturing on a large scale. On > the face of it, though, experience with Prescott suggests that Intel > is maybe not so good at managing huge design resources--but that's > only a guess. Big organizations can have big economies of scale. > They can also have bloated and dysfunctional org charts. > > >> >>>In any case, I suspect the decision against SOI was a matter of cost, >>>and I even vaguely remember some statements to that effect. I said >>>actually that I had cited stock prices as the one available indicator, >>>when margin (also publicly available) indicates that Intel manages to >>>have lower manufacturing costs. It all comes down to Intel being a >>>*business* and not a classroom project, a dorm room bull session, or a >>>soccer football team. >> >>Intel manages to have lower manufacturing costs than AMD primarily >>because it has much greater volume. Look up "learning curve". >> > > I'm not in the business, so I couldn't comment on how easy it is to > scale up process manufacturing. Not trivial, though, I'll bet. > > >> >>>Your post seems to confirm what I think is your ongoing delusion about >>>Intel: that it is simply a better marketing machine than AMD. That it >>>*is* a better marketing machine is probably correct. That that's all >>>there is to Intel is nothing short of corporate defamation, if that's >>>what you indeed intend to imply. Intel's real advantage, widely >>>acknowledged in the industry, is that it knows how to manufacture high- >>>end microprocessors at the lowest possible cost. >> >>>SOI is one of many dead and rotting horses on csiphc. Let it be. >>>What's interesting about it is historical: it played a key role in the >>>triumph of x86. Isn't that enough? >> >> >>Many processors are made with SOI, including all of IBM's, and the three >>game console processors. So I don't know why you think it is a dead and >>rotting horse. It has some advantages, after all. >> > > One can never be too careful when writing a post. I naively assumed > the implied context of the groups: ibm pc's and intel systems. From > those points of view, it's all old news, unless, of course, you're > scratching around for positive things to say about AMD's lagging > technology or trying to revive an old argument. > > Robert. Why would SOI have an advantage for some processors and not for others? I would attribute AMD's problems to other causes than their use of SOI for their processors, although honest men disagree over the magnitude and sign of SOI's advantage at a given process node. And scaling up manufacturing of these kind of processors isn't trivial, but it is doable. And the lesson of the learning curve is that the more you make the better you get at it and the less they cost. So the fact that Intel makes maybe 5 times as many processors as AMD does gives them a cost advantage over and above that bestowed by spreading fixed chip design and process development costs over a larger base. As for Intel suffering for its size, I can't think of examples besides getting slapped on the wrist a few times for being a little too aggressive in pursuit of market share. Note that I am not saying that they didn't make mistakes in planning and execution along the way. Itanium is probably the most notable. Netburst wasn't the technically best solution, but Intel sold a lot of them and made a lot of money. And both it and Itanium would have been more successful had it not been for AMD. But being the underdog with small market share requires near perfection in execution. I don't claim to be totally dispassionate, although I never had any professional or financial skin in the SOI issue. Would you amplify on AMD's lagging technology since to my admittedly casual observations there seems to be a rough parity. I am deleting the intel newsgroup from this post since I am not trying to start anything. del
From: Robert Myers on 23 Dec 2008 00:21 On Dec 22, 11:45 pm, Del Cecchi <delcecchinos...(a)yahoo.com> wrote: > > Why would SOI have an advantage for some processors and not for others? It is evident that I am not a process guy. Nor do I have a clue about the internals of Intel's operations. They didn't want to use SOI, apparently because of cost. Does anyone have anything else to add? They didn't get what they expected from strained silicon. I'm sure that if I wanted to read up on the subject, I could get a clue as to why, but that would be energy wasted for this group. Intel's design and process technology flopped at 90nm--a double whammy. That still doesn't mean that Intel should have gone with SOI. No one here has enough information to say that. > I would attribute AMD's problems to other causes than their use of SOI > for their processors, although honest men disagree over the magnitude > and sign of SOI's advantage at a given process node. > I don't know enough about process technology to comment one way or the other. If there is anyone in this group who knows anything about solid state physics, no one has shown it, so what we've had is an endless repetition of content-free polemics. I do understand, at least in theory, how straining silicon affects the mobility of electrons. I infer that Intel thought it could get enough out of strained silicon and did not want to use and SOI process for reasons of cost. As to the *current* technologies, if SOI is working for AMD at 45nm, then fine. I have no brief for any particular technology. Yousuf's argument is apparently that SOI is superior. My response is 1. Who cares. 2. AMD and Intel are always leapfrogging on one performance measure or another. That a partiuclar performance measure goes a particular way at a particular time doesn't mean much. I don't want to put the cognitive effort into proposing what I think would be a meaningful comparison, but I do not think that what Yousuf offered was meaningful. > And scaling up manufacturing of these kind of processors isn't trivial, > but it is doable. There, I think you unfairly slight Intel, but reasonable people may disagree. Sure, it's doable, but, apparently, Intel does it exceedingly well. And the lesson of the learning curve is that the more > you make the better you get at it and the less they cost. So the fact > that Intel makes maybe 5 times as many processors as AMD does gives them > a cost advantage over and above that bestowed by spreading fixed chip > design and process development costs over a larger base. > Only if you can successfully replicate the technology at a sufficiently large scale, something that, if my memory is correct, IBM has not always been able to do. > As for Intel suffering for its size, I can't think of examples besides > getting slapped on the wrist a few times for being a little too > aggressive in pursuit of market share. GM's stock value is estimated by analysts to be zero. GM made an awful lot more cars than anybody else for an awfully long time. > Note that I am not saying that > they didn't make mistakes in planning and execution along the way. > Itanium is probably the most notable. > > Netburst wasn't the technically best solution, but Intel sold a lot of > them and made a lot of money. And both it and Itanium would have been > more successful had it not been for AMD. But being the underdog with > small market share requires near perfection in execution. > As one of Intel's architects has said publicly, Intel doesn't understand software. P4 and Itanium are great for some kinds of software, just like streaming processors are great for some kinds of software. Short-pipeline x86 is great for the ratty base of installed crapware, but that still doesn't mean it's going to survive in the long run. You read it here first. > I don't claim to be totally dispassionate, although I never had any > professional or financial skin in the SOI issue. > > Would you amplify on AMD's lagging technology since to my admittedly > casual observations there seems to be a rough parity. > They're six months to a year behind in scale shrinks. > I am deleting the intel newsgroup from this post since I am not trying > to start anything. I'd love to say something here, but it *would* start something. Whatever gets started, it won't be with you. Robert.
From: Del Cecchi` on 23 Dec 2008 20:45 Robert Myers wrote: > On Dec 22, 11:45 pm, Del Cecchi <delcecchinos...(a)yahoo.com> wrote: > > >>Why would SOI have an advantage for some processors and not for others? > > > It is evident that I am not a process guy. Nor do I have a clue about > the internals of Intel's operations. They didn't want to use SOI, > apparently because of cost. Does anyone have anything else to add? Cost was it, and they didn't think they needed what advantage it might give. > > They didn't get what they expected from strained silicon. I'm sure > that if I wanted to read up on the subject, I could get a clue as to > why, but that would be energy wasted for this group. Intel's design > and process technology flopped at 90nm--a double whammy. That still > doesn't mean that Intel should have gone with SOI. No one here has > enough information to say that. Sorry, 90 nm is pretty much ancient history. But I don't recall a process flop. > > >> I would attribute AMD's problems to other causes than their use of SOI >>for their processors, although honest men disagree over the magnitude >>and sign of SOI's advantage at a given process node. >> > > I don't know enough about process technology to comment one way or the > other. If there is anyone in this group who knows anything about > solid state physics, no one has shown it, so what we've had is an > endless repetition of content-free polemics. I do understand, at > least in theory, how straining silicon affects the mobility of > electrons. I infer that Intel thought it could get enough out of > strained silicon and did not want to use and SOI process for reasons > of cost. Don't forget hole mobility. Very important in CMOS. And if I recall correctly, one of the strained silicon advantages was it let you increase both hole and electron mobility. As for polemics and knowledge of solid state physics etc, I resent that statement. I have some knowledge and have yet to engage in polemics. > > As to the *current* technologies, if SOI is working for AMD at 45nm, > then fine. I have no brief for any particular technology. Yousuf's > argument is apparently that SOI is superior. My response is > > 1. Who cares. > > 2. AMD and Intel are always leapfrogging on one performance measure or > another. That a partiuclar performance measure goes a particular way > at a particular time doesn't mean much. I don't want to put the > cognitive effort into proposing what I think would be a meaningful > comparison, but I do not think that what Yousuf offered was > meaningful. > > >>And scaling up manufacturing of these kind of processors isn't trivial, >>but it is doable. > > > There, I think you unfairly slight Intel, but reasonable people may > disagree. Sure, it's doable, but, apparently, Intel does it > exceedingly well. Yes they do. I was responding to your comment that I interpreted as casting doubt on the "learning curve" since scaling up production is hard. > > And the lesson of the learning curve is that the more > >>you make the better you get at it and the less they cost. So the fact >>that Intel makes maybe 5 times as many processors as AMD does gives them >>a cost advantage over and above that bestowed by spreading fixed chip >>design and process development costs over a larger base. >> > > Only if you can successfully replicate the technology at a > sufficiently large scale, something that, if my memory is correct, IBM > has not always been able to do. Go read up on the semiconductor learning curve. TI came up with the concept and it goes way back, before Moore's law. Everyone has problems in their fabs now and then, going from development to production. But why the dig at IBM? Trying to provoke a polemic? > > >>As for Intel suffering for its size, I can't think of examples besides >>getting slapped on the wrist a few times for being a little too >>aggressive in pursuit of market share. > > > GM's stock value is estimated by analysts to be zero. GM made an > awful lot more cars than anybody else for an awfully long time. I thought you were talking about Intel, not GM. What does GM have to do with it? > > >>Note that I am not saying that >>they didn't make mistakes in planning and execution along the way. >>Itanium is probably the most notable. >> >>Netburst wasn't the technically best solution, but Intel sold a lot of >>them and made a lot of money. And both it and Itanium would have been >>more successful had it not been for AMD. But being the underdog with >>small market share requires near perfection in execution. >> > > As one of Intel's architects has said publicly, Intel doesn't > understand software. P4 and Itanium are great for some kinds of > software, just like streaming processors are great for some kinds of > software. Short-pipeline x86 is great for the ratty base of installed > crapware, but that still doesn't mean it's going to survive in the > long run. You read it here first. > > >>I don't claim to be totally dispassionate, although I never had any >>professional or financial skin in the SOI issue. >> >>Would you amplify on AMD's lagging technology since to my admittedly >>casual observations there seems to be a rough parity. >> > > They're six months to a year behind in scale shrinks. That is what lack of money and market share will do. The real question is are they cost and performance (including power) competitive? At some points moving to a smaller node increases the chip cost. It is only after you get down the learning curve a ways that crossover occurs. > > >>I am deleting the intel newsgroup from this post since I am not trying >>to start anything. > > > I'd love to say something here, but it *would* start something. > Whatever gets started, it won't be with you. > > Robert. I am sure there are folks who are pulling for AMD for reasons that are personal or financial rather than technical, because AMD is an underdog or they have stock in AMD or because they dislike Intel for whatever reason. I try not to get into those types of discussions. del
From: Robert Myers on 23 Dec 2008 22:05 On Dec 23, 8:45 pm, Del Cecchi` <dcecchinos...(a)att.net> wrote: > Robert Myers wrote: > > > They didn't get what they expected from strained silicon. I'm sure > > that if I wanted to read up on the subject, I could get a clue as to > > why, but that would be energy wasted for this group. Intel's design > > and process technology flopped at 90nm--a double whammy. That still > > doesn't mean that Intel should have gone with SOI. No one here has > > enough information to say that. > > Sorry, 90 nm is pretty much ancient history. But I don't recall a > process flop. > All recorded in this august group for posterity. Andy Grove showed up at a Transmeta rollout, apparently looking for clues to make things better at 90nm. This discussion has been going on here for years. And 90nm was AMD's opening (helped along in a big way by IBM), ancient history or not. > > As for polemics and knowledge of solid state physics etc, I resent that > statement. I have some knowledge and have yet to engage in polemics. > I spoke of knowledge as shown in csiphc. > > Everyone has problems in their fabs now and then, going from development > to production. But why the dig at IBM? Trying to provoke a polemic? > Polemicize away if the spirit moves you. You've carved this down to csiphc, where there are no Intel employees to write to me to say, "Thank God you're here." Instead, I'm surrounded by current and former IBM'ers whose envy of Intel is exceeded only by their loathing of Microsoft. > > > > GM's stock value is estimated by analysts to be zero. GM made an > > awful lot more cars than anybody else for an awfully long time. > > I thought you were talking about Intel, not GM. What does GM have to do > with it? > I'm pretty picky about what I read up on (read the best books first, as Ezra Pound said, you may not get a chance to read any others), and I'm not likely to spend my time reading about more microprocessor industry self-aggrandizement. To quote you: > > And the lesson of the learning curve is that the more > > >>you make the better you get at it and the less they cost. So the fact > >>that Intel makes maybe 5 times as many processors as AMD does gives them > >>a cost advantage over and above that bestowed by spreading fixed chip > >>design and process development costs over a larger base. > That applies differently to Intel or to IBM than it does to GM? The handling characteristics of a product with so many moving parts as a car can easily be duplicated from one copy to another? It's not even true of the manufacture of pianos. Ask any Steinway salesman. Both the human ear and the grip of rubber on the road are subtle and complicated. Just ask about the tires at last summer's Indy 500. Don't get me wrong. Microprocessors are nothing short of amazing. In the end, though, manufacturing is manufacturing and every industry has its bag of tricks and mysteries. Being bigger doesn't necessarily mean that your bag is better or even necessarily bigger. > > > > They're six months to a year behind in scale shrinks. > > That is what lack of money and market share will do. As my track coach used to say, saying "Yeah but they had all the horses" doesn't change the outcome of the meet. > The real question > is are they cost and performance (including power) competitive? That's determined by design wins. AMD ain't doing so hot, and arguing here isn't going to change that. > At some > points moving to a smaller node increases the chip cost. It is only > after you get down the learning curve a ways that crossover occurs. > We *are* headed toward monopoly. IBM will hang onto it's Power business somehow or other, someone else, maybe Asian, will fill in the low end, and Intel will have won the war of attrition. Or maybe none of it will matter because the action moves to something like Cell. Robert.
From: Del Cecchi` on 23 Dec 2008 22:43
Robert Myers wrote: > On Dec 23, 8:45 pm, Del Cecchi` <dcecchinos...(a)att.net> wrote: > >>Robert Myers wrote: > > >>>They didn't get what they expected from strained silicon. I'm sure >>>that if I wanted to read up on the subject, I could get a clue as to >>>why, but that would be energy wasted for this group. Intel's design >>>and process technology flopped at 90nm--a double whammy. That still >>>doesn't mean that Intel should have gone with SOI. No one here has >>>enough information to say that. >> >>Sorry, 90 nm is pretty much ancient history. But I don't recall a >>process flop. >> > > All recorded in this august group for posterity. Andy Grove showed up > at a Transmeta rollout, apparently looking for clues to make things > better at 90nm. This discussion has been going on here for years. > And 90nm was AMD's opening (helped along in a big way by IBM), ancient > history or not. Transmeta had no process technology. I am not sure which foundry they used. A quick search says the first was TSMC at 130 nm, followed by a switch to Fujitsu at 90. > > >>As for polemics and knowledge of solid state physics etc, I resent that >>statement. I have some knowledge and have yet to engage in polemics. >> > > I spoke of knowledge as shown in csiphc. You generalized and did not qualify. > > >>Everyone has problems in their fabs now and then, going from development >>to production. But why the dig at IBM? Trying to provoke a polemic? >> > > Polemicize away if the spirit moves you. You've carved this down to > csiphc, where there are no Intel employees to write to me to say, > "Thank God you're here." Instead, I'm surrounded by current and > former IBM'ers whose envy of Intel is exceeded only by their loathing > of Microsoft. I don't envy Intel, and never did. Please leave me out of this. > >> >>>GM's stock value is estimated by analysts to be zero. GM made an >>>awful lot more cars than anybody else for an awfully long time. >> >>I thought you were talking about Intel, not GM. What does GM have to do >>with it? >> > > I'm pretty picky about what I read up on (read the best books first, > as Ezra Pound said, you may not get a chance to read any others), and > I'm not likely to spend my time reading about more microprocessor > industry self-aggrandizement. To quote you: > > >>> And the lesson of the learning curve is that the more >> >>>>you make the better you get at it and the less they cost. So the fact >>>>that Intel makes maybe 5 times as many processors as AMD does gives them >>>>a cost advantage over and above that bestowed by spreading fixed chip >>>>design and process development costs over a larger base. >> > > That applies differently to Intel or to IBM than it does to GM? The > handling characteristics of a product with so many moving parts as a > car can easily be duplicated from one copy to another? It's not even > true of the manufacture of pianos. Ask any Steinway salesman. Both > the human ear and the grip of rubber on the road are subtle and > complicated. Just ask about the tires at last summer's Indy 500. > Don't get me wrong. Microprocessors are nothing short of amazing. In > the end, though, manufacturing is manufacturing and every industry has > its bag of tricks and mysteries. Being bigger doesn't necessarily > mean that your bag is better or even necessarily bigger. I think the learning curve applies to Semiconductors more than to automobiles or grand pianos. > >> >>>They're six months to a year behind in scale shrinks. >> >>That is what lack of money and market share will do. > > > As my track coach used to say, saying "Yeah but they had all the > horses" doesn't change the outcome of the meet. > > >>The real question >>is are they cost and performance (including power) competitive? > > > That's determined by design wins. AMD ain't doing so hot, and arguing > here isn't going to change that. It isn't determined by design wins. It is determined by the behavior of the part and stuff like die size. Design wins are influenced by many things, and technical superiority is only one factor. > > >>At some >>points moving to a smaller node increases the chip cost. It is only >>after you get down the learning curve a ways that crossover occurs. >> > > We *are* headed toward monopoly. IBM will hang onto it's Power > business somehow or other, someone else, maybe Asian, will fill in the > low end, and Intel will have won the war of attrition. Or maybe none > of it will matter because the action moves to something like Cell. > > Robert. Could be. Or it could be that the industry moves more towards a netcentric or client server or thin client model or some other arrangement where the processor architecture and operating system are of less importance. IBM had a monopoly in punch cards and almost one in the 360 era and we know how much help that was 20 years later. del |