PPC970 in 90nm process? (news)

Posted:
in Future Apple Hardware edited January 2014
<a href="http://www.vnunet.fr/actu/article.htm?numero=10479&date=2002-12-16"; target="_blank">vnunet.fr</a>



This article says that IBM will manufacture 90nm chips for Xilinx in their East Fishkill Fab that will debute in the 2nd semestre on 300 mm wafer. This will reduce by 50 to 80% the production costs against the 130nm.



The PPC 970 will be manufactured in this fab? No?

IBM said that they would introduce the 970 with 130nm.



Can two manufacturing process coexist in the same fab plant? Why would they do different manufacturing process in a new fab plant?
«1

Comments

  • Reply 1 of 38
    <a href="http://www.eweek.com/article2/0,3959,767883,00.asp"; target="_blank">This articke @ eWeek</a> mentions that IBM will make 90 nm-porcessors for Xilinx H2'03. It states that IBM will make FPGA-chips, and not 970. But.. if they are making any chips in 90 nm the second half of next year, I'd be a happy Mac-user.





    But.. the french article says exactly the same thing as the eWeek-article. No mention of 970 what so ever.
  • Reply 2 of 38
    Yes I know. If they can produce in 90nm the 970 than the costs for Apple will be down and the margins for IBM will be higher.



    Need they work on chips to pass from one process to another?
  • Reply 3 of 38
    Maybe the other chip is less complex than the 970? Maybe the process has already been nailed for the .13 970.



    I'd hardly imagine it's something you could flit about with.



    .9 will come soon after the 970 .13. Half a year later?



    I don't think this news has any bearing on Apple getting the .13 970 any sooner.



    It looks like September next year IF we're lucky.



    Yeesh, Apple didn't have a proper motherboard ready for the current g4 'revision', so Apple Pi for the next chip? Bet's on that Apple delays IBM into powerMacs and not the other way around?



    Lemon Bon Bon
  • Reply 4 of 38
    ompusompus Posts: 163member
    The East Fishkill plant seems intended for sub 130nm chips. I also gather that IBM will move the 970 to 90nm rather quickly.



    Now I don't know the economics of fabrication, but I wonder if IBM might surprise everyone by skipping 130nm altogether. I do not know the costs of moving from 130nm to 90nm but I imagine it is significant. But why spend significant money tooling up a 130nm process if you plan to move 90nm?
  • Reply 5 of 38
    outsideroutsider Posts: 6,008member
    The 90nm move does not worry me as much as the actual availability of the 970. I hope that IBM has an aggressive schedule for the 970 on 130nm. We needed it yesterday.
  • Reply 6 of 38
    It may well be that Apple told IBM "We need this chip yesterday, so we'll be satisfied with a 130nm chip for now." Otherwise I think we would have seen a later release date and higher speeds.
  • Reply 7 of 38
    snoopysnoopy Posts: 1,901member
    It takes a while to get a new process running smoothly, so IBM and Apple do not want any unnecessary risks. On the other hand, it is not unthinkable that they would be starting the 970 on the 90 nm process in parallel with 130 nm. If 90 nm is running well, that will be the one they go with. If they take this approach, they would not say anything, in the event they must use 130 for a while.
  • Reply 8 of 38
    powerdocpowerdoc Posts: 8,123member
    The fab process is not everything for clock speed and heat issue in a chip.

    The athlon 2200 + 0,13 micron was not cooler than the old Athlon palomino made under 0,18 microm process .

    The athlon 2200 + (architecture thoroughbred a0) was a great disapointement , and hardly go beyond 1,8 ghz and produced a lot of heat.



    AMD was in a great trouble with this chip, and make a huge work in order to show the new architecture the thorougbred A1 with 300 000 more transistors and a ninth layer. Due to this refinements the chip was able to be overclocked until 2,6 ghz (500 mhz more than the old architecture of the thoroughbred A0).



    So the fab process is not everything, and it's better to have a well optimized 0,13 SOI fab process than a approximative 0,09 nm process. You will notice also that the power 5 will be produced upon 0,13 micron process (like the 970) in early 2004



    Intel is on 0,13 like AMD but without SOI.
  • Reply 9 of 38
    ed m.ed m. Posts: 222member
    Powerdoc, I tend to agree with what you say, however...



    Keep in mind that this is IBM we are talking about here. Unlike AMD, they aren't pressed for $$ and they don't seem to be surviving from chip release to chip release and they aren't in a head-to-head race with Intel, which means that they didn't have to push any technology faster than it should have been pushed, nor did they have to cut any corners. The 970 is vastly different from what AMD and Intel produce. It's arguably a more tested and mature chip - relatively speaking that is. Perhaps Programmer can pick up the ball at this point and better explain what I'm trying to say here. I guess what I'm trying to say is that IBM is DEFINITELY NOT AMD.



    --

    Ed M.
  • Reply 10 of 38
    [quote]Originally posted by Powerdoc:

    <strong>You will notice also that the power 5 will be produced upon 0,13 micron process (like the 970) in early 2004</strong><hr></blockquote>One reason for IBM to make the Power5 in 130 nm is its reliability. Everything about producing the Power4 is done with reliability in mind. they don't want the processors to fail while in operation. That philosophy isn't going to change with Power5.



    But.. If IBM is going to produce these Xilinx FPGAs in large volumes 2H03, then they probably have the capacity to make the 970 in volume too. But. There might be other reasons for them not to, such as the 970 must be working for Apple.. delays due to manucacturing problems are out of the question.
  • Reply 11 of 38
    outsideroutsider Posts: 6,008member
    Speculation**** IBM may intend to officially bring the 970 out on 90nm, but for Apple to have some early chips for 1H03 they will release the initial batch on 130nm.
  • Reply 12 of 38
    [quote]Originally posted by Outsider:

    <strong>Speculation**** IBM may intend to officially bring the 970 out on 90nm, but for Apple to have some early chips for 1H03 they will release the initial batch on 130nm.</strong><hr></blockquote>



    From the MPF documents, they [IBM] said that it [the 970] was going to be released on a 130nm process, going to 90nm "rather quickly". The official release of the 970 will be 130nm... what happens 6+ months from that is pure speculation.



    The point of all this being that 130nm 970 chips have been available in "one off" quantity, and they are not going to throw away a design, just so they can go with an as-of-yet untested 90nm version of the chip.



    .

    the visigothe
  • Reply 13 of 38
    Reply for ED M.



    They clearly go head-to-head against x86 and Microsoft. They want low-end servers and perhaps desktop working with Linux with their 970. If IBM come out with such a desktop then we will see the differencies between MacOS X and Linux on the same processor.
  • Reply 14 of 38
    [quote]Originally posted by Spartacus:

    <strong>Reply for ED M.



    They clearly go head-to-head against x86 and Microsoft. They want low-end servers and perhaps desktop working with Linux with their 970. If IBM come out with such a desktop then we will see the differencies between MacOS X and Linux on the same processor.</strong><hr></blockquote>





    The problem there being that Linux, by its nature, will be faster. A monolithic kernel [unless deeply flawed] will always be faster than a micro kernel architecture, like Mach/OSX.



    Will it be easier to use? Well, that is a question. Surely for testing, and desktop work, the Mac OS is head and shoulders ahead of the X-Windows universe. KDE cannot hope to compete with OSX, as it still tries to play catch-up with Windows, rather than trying to innovate the GUI.



    The other question is "will the speed be competitive" If Linux on the 970 is 3 times faster than OSX on the 970, it's not going to look too good for Apple. If the speed difference is ~10 or 15%, and the cost is also similar, then it becomes a non-issue. People will buy the environment they like... akin to the Mercedes vs. BMW debate. Both make wonderful cars, but, in the end, you're a Mercedes person, or you're a BMW person.... unless you are like my friend Bill who ownes one of each!



    .

    the visigothe
  • Reply 15 of 38
    [quote]The problem there being that Linux, by its nature, will be faster. A monolithic kernel [unless deeply flawed] will always be faster than a micro kernel architecture, like Mach/OSX. <hr></blockquote>

    Don't mean to nitpick but OSX is far from a micro kernel. Mach started out as a pure microkernel, but the version used in OSX is a hybrid with many of the characteristics of a monolithic kernel. It's much more towards the microkernel end of the spectrum compared to Linux however. But the kernel is not the main reason that OSX is so slow compared to Linux, it's a small part of it but most of the slowdown comes from overhead from the graphics display engine, unoptimized APIs, and various other unoptimized bits of a still very immature OS.

    [quote]The other question is "will the speed be competitive" If Linux on the 970 is 3 times faster than OSX on the 970, it's not going to look too good for Apple. If the speed difference is ~10 or 15%, and the cost is also similar, then it becomes a non-issue. People will buy the environment they like... akin to the Mercedes vs. BMW debate.<hr></blockquote>

    The 970 may be powerful enough that OSX's speed issues could be rendered largely moot, which would certainly be nice. Linux may still be 3 times faster, but if OSX already gets things done faster than we can accurately perceive on the 970 then it doesn't really matter does it?
  • Reply 16 of 38
    The performance advantage of a monolithic kernel compared to a micro kernel is probably going to be mitigated in SMP hardware designs, and the world is going much more SMP than it has in the past. Also, as stated above, the MacOS X kernel has a whole lot more functionality in the kernel space than a traditional micro kernel normally does.



    As Apple continues to optimize OS X and address its current weak points then speed gap will drop quickly. Linux isn't going to get much faster as its pretty well optimized already.



    I don't know what the speed differences are right now -- and there are lots of different things which can be measured in terms of OS performance. I doubt that anything really important is three times slower in MacOS X. There might be bits and pieces here and there which are that much slower, but they're probably a small part of any particular task. If 10% of the work is 3 times slower and the other 90% is the same speed then overall it'll only be about 20% slower. I've heard there are a couple of parts of MacOS X which are actually faster than the competition. The advent of Jaguar, gcc 3, and Quartz Extreme has helped a lot. And keep in mind that in many cases the speed of the OS is irrelevant because its the application that is taking all the time.
  • Reply 17 of 38
    That post is totally wrong. Linux IS highly optimised, but has lots of room to grow. Just look at the scalability and performance in the 2.5 (bleeding edge, devel) kernel. New block I/O, New <a href="http://kerneltrap.org/node.php?id=517"; target="_blank">0(1) scheduler</a>, new <a href="http://kerneltrap.org/node.php?id=336"; target="_blank">kernel preemption</a> makes thing like the GUI super snappy, regardless of how much CPU usage is going on. (so like, compiling six kernels simultaneously will not slow down menus and window resizes... seriously!).



    I was blown away moving from the 2.2 to 2.4 kernel. 2.6 is going to be k-radder still. It is computer science research in action, with awesome new technology being incorporated and refined every day.



    [ 12-18-2002: Message edited by: 1337_5L4Xx0R ]</p>
  • Reply 18 of 38
    [quote]Originally posted by 1337_5L4Xx0R:

    <strong>That post is totally wrong. Linux IS highly optimised, but has lots of room to grow. </strong><hr></blockquote>



    No, the post is not totally wrong. The wrong statement is (given that you are right) is that linux isn't going to be much faster. Maybe it is, but since it is highly optimized compared to OS X, linux isn't going to see the same amount of speed increase as with OS X, since OS X has a much larger headroom for optimization.



    If all those feats/optimizations can be implemented in linux, they sure can be in OS X too.
  • Reply 19 of 38
    I was also referring to the current bleeding edge of Linux development. The rate at which you can make software faster decreases as you work on it... the good old law of diminishing returns.
  • Reply 20 of 38
    Actually, I was unclear, but the wrong statement above was:

    [quote] The performance advantage of a monolithic kernel compared to a micro kernel is probably going to be mitigated in SMP hardware designs...<hr></blockquote>



    Whereas, SMP performance is directly tied to Scheduler performance, and the scheduler is in the kernel. All of the uniprocessor gains in 2.5 linux kernels I gush about above are a result of (originally devised for SMP, for scalability) scheduler improvements, and on SMP machines, those same improvements greatly improve performance (obviously). More intelligent scheduling plus greatly reduced overhead.



    Really cool stuff. Sorry for the off-topic posts. BTW, not suggesting 'linux rocks, osx sucks' type bs, just pointing out that scheduling, SMP, speed and kernels are all interrelated.



    [ 12-18-2002: Message edited by: 1337_5L4Xx0R ]</p>
Sign In or Register to comment.