Power5 based 975 90nm chips to ramp in May?

245678

Comments

  • Reply 21 of 148
    oldmacfanoldmacfan Posts: 501member
    Quote:

    Originally posted by Rhumgod

    That's the understatement of the century. This whole thread is really baseless and just doesn't offer any kind of realistic speculation.



    Even worse, it references Nr9's blab about dual-core 440's going into PowerBooks. When Apple releases a router called "PowerBook", I'll buy it...until then this is all just a bunch of crap!



    Power5's have been shipping for a few months, but how long were the Power4's in products before a 970 was released? Couple of years? No, IBM isn't that good/ignorant.




    Power5's have been shipping for a few months? IBM is saying they will start shipping them in June. IBM is shipping blades based on a PPC 970 (which is Power4 based) at this time with a top speed of 1.6Ghz.



    As for your Baseless comment, I have a very good base for my speculation. I just happened to be triggered to finally post by a source that is not well respected. That being said their are bits and pieces that come from other places. As for realistic, it is very realistic that Apple and IBM have been working on a Power5 based PPC.



    The reason Power4's had been shipping for so long before the release of the 970fx, was that Apple and IBM put their heads together on the project very late in that processor's life cycle. That is why they went with designing two different lines at the same time.
  • Reply 22 of 148
    programmerprogrammer Posts: 3,467member
    Quote:

    Originally posted by PB

    How do we know that this has been corrected? There has been some announcement or do you know something not publicly available?



    IBM said as much, and Xserves (even duals) are now shipping. Apple said they expected the yield problems to be corrected this quarter and while everyone seems to read this as "the last day of this quarter", I disagree.







    Its been said twice here now, but I'll say it again: the development of the 970 started very late in the POWER4 project and that caused the long lag.



    The followup to the 970 (not the process shrink to 970FX) started the day they finished the 970 or earlier, and it started from the 970 and POWER5 work that was already largely complete. IBM's use of automated design tools should also accelerate the rate at which they can develop new designs. All of this means that the next chip will show up long before you would expect based on the POWER4->970 timeline. Even simultaneous with the POWER5 wouldn't surprise me because I do not believe that they are taking the finished POWER5 and scaling it down.







    Some people seemed to have missed another point: the 970FX is a tiny little chip. It is well below the normal die size for a desktop chip and is intended for embedded, portable and high density applications (i.e. routers, notebooks and blade/rack servers). It leaves plenty of room for a stablemate that has a significantly larger die size, with transistor counts similar to Prescott. That means IBM can potentially double the transistor count of the 970FX... leaving more than enough room to double the cache size and pay the 25% core size increase that was reported (by IBM) for the POWER5 to support SMT. While they're at it they could address a few of the 970's shortcomings.
  • Reply 23 of 148
    msanttimsantti Posts: 1,377member
    Rumor has it that Apple just ordered truckloads of this:











  • Reply 24 of 148
    tinktink Posts: 395member
    I think this summers line up was / is suppose to look something like this:



    G4 --- iBook, eMac.

    G5 (970FX) --- iMac, Xserve, Powerbook (closer to fall).

    G6 (975) --- Powermacs (announced around June and WWDC2004), Xserve (maybe, depending on size and heat).



    I agree with Programer that the 3GHz chip could very well be the 975.
  • Reply 25 of 148
    pbpb Posts: 4,255member
    Quote:

    Originally posted by Programmer

    IBM said as much, and Xserves (even duals) are now shipping. Apple said they expected the yield problems to be corrected this quarter and while everyone seems to read this as "the last day of this quarter", I disagree.





    So, what was the problem, the 90 nm process or the SSOI technology, used by IBM only for now, if I am not mistaken?
  • Reply 26 of 148
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by Programmer





    . . . Some people seemed to have missed another point: the 970FX is a tiny little chip. It is well below the normal die size for a desktop chip and is intended for embedded, portable and high density applications (i.e. routers, notebooks and blade/rack servers). It leaves plenty of room for a stablemate that has a significantly larger die size, with transistor counts similar to Prescott. That means IBM can potentially double the transistor count of the 970FX... leaving more than enough room to double the cache size and pay the 25% core size increase that was reported (by IBM) for the POWER5 to support SMT. While they're at it they could address a few of the 970's shortcomings.






    Your comment about a small die size of the 970FX should mean lower price too, yes? More chips per wafer means lower cost, once the 90 nm processes is ironed out and yields are very high. I would say your analysis is right on for the longer range uses for these two chips. I'd include things like a new iMac in your list of potential uses, no?



    Yet, it's interesting that the 970FX could be used in the Power Mac for the next revision or speed bump. I projected a power dissipation of about 55 Watts at 3 GHz for the 970FX running at 1.3 Volts, assuming it will run that fast. Some of the power dissipation data we have read about is at 1.1 Volts I believe.
  • Reply 27 of 148
    Quote:

    Originally posted by snoopy

    ...once the 90 nm processes is ironed out and yields are very high...



    Just sos you all know.



    IBM isn't there yet.



  • Reply 28 of 148
    whoamiwhoami Posts: 301member
    Quote:

    Originally posted by Transcendental Octothorpe

    Just sos you all know.



    IBM isn't there yet.









    that's pretty obvious!

    would you care to elaborate?

    i would love to hear what is really going on at Fishkill with the .09....
  • Reply 29 of 148
    programmerprogrammer Posts: 3,467member
    Quote:

    Originally posted by PB

    So, what was the problem, the 90 nm process or the SSOI technology, used by IBM only for now, if I am not mistaken?



    We don't know. The most detailed rumour was that it was neither, it was a new packaging technology they tried using on the 970FX.



    Another thing to keep in mind is that there isn't one 90nm process, there is a different one per manufacturer. They all have their crown jewels in the form of trade secrets.





    TO's point should be well taken -- we have virtually no evidence as to what the current state of affairs really is. We also don't for sure what the problem is/was, or even its nature... rumours indicate that it was apparently some form of gating problem (stopping all speeds equally), but rumours being what they are we are still in the dark. Keep your fingers crossed.





    Re: cost & heat of 970FX. Yes, if IBM can get decent yields out of Fishkill then the prices should be low. If they can't then the small size of the 970FX might be the only thing that saves them. If they can get yields even that high.



    As for power & heat, I believe dissapation increases nonlinearly with clock rate. Its roughly a squared or cubed relationship, isn't it? Going from 40W @ 2 GHz would give 90+ watts @ 3 GHz.
  • Reply 30 of 148
    Quote:

    Originally posted by Programmer

    4) It is debateable whether the 970FX is going to reach 3 GHz, and it is possible that Steve's promise of 3 GHz was based on the 975 from the beginning. Since the 975 is only hitting production next month (supposedly) it will never suffer from the initial troubles the 970FX had. The 90nm problems basically "soaked up" the 970FX's lead time.



    This as always been my secret hope too, that the 970fx was only ever intended for the now aborted Jan/Feb 2.5 GHz Rev. Bs with the 975 as the 3GHz part for the Rev. Cs at WWDC 04. In other words, when the 970fx fell on its face early this year, Steve said screw it, we'll just sit tight with the current Power Macs and go straight to the 3GHz 975 models in June.



    This scenario would allow for the inclusion of PCIe and other mobo improvements and most importantly let Steve make a big deal out of the G5's one year anniversary at his keynote. If he can hold a bakeoff with a Dual 3Ghz G5 with a PCIe native 256MB RV420 then all will be instantly forgiven.
  • Reply 31 of 148
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by Programmer





    . . . As for power & heat, I believe dissapation increases nonlinearly with clock rate. Its roughly a squared or cubed relationship, isn't it? Going from 40W @ 2 GHz would give 90+ watts @ 3 GHz.




    I consider two parts of power dissipation. One part is due to leakage current, which is roughly constant at a particular operating voltage. The second is due to switching, which increases linearly with clock frequency at a particular voltage. If you can get data points for two fairly separated clock rates, at the same voltage, then you can calculate both the leakage current and switching power portions for that voltage. If the clock rates are too close together the errors can get big. The switching portion will be expressed as so many Watts per Gigahertz, or Megahertz if you prefer.



    Now the switching power increases by the square of the operating voltage, so if you have it for one particular voltage you can calculate it for any other voltage. I don't believe there is a reliable rule for how leakage current change with voltage. To find it, however, you only need one data point at the new voltage, since you have calculated the switching power already.
  • Reply 32 of 148
    programmerprogrammer Posts: 3,467member
    Quote:

    Originally posted by snoopy

    I consider two parts of power dissipation. One part is due to leakage current, which is roughly constant at a particular operating voltage. The second is due to switching, which increases linearly with clock frequency at a particular voltage. If you can get data points for two fairly separated clock rates, at the same voltage, then you can calculate both the leakage current and switching power portions for that voltage. If the clock rates are too close together the errors can get big. The switching portion will be expressed as so many Watts per Gigahertz, or Megahertz if you prefer.



    Now the switching power increases by the square of the operating voltage, so if you have it for one particular voltage you can calculate it for any other voltage. I don't believe there is a reliable rule for how leakage current change with voltage. To find it, however, you only need one data point at the new voltage, since you have calculated the switching power already.




    Ah. Well clearly you know more about it than I do. 55W sounds a little low to me though... you confident in that figure?
  • Reply 33 of 148
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by Programmer

    . . . 55W sounds a little low to me though... you confident in that figure?





    I may have calculated 59 Watts, since I remembered saying it was just under 60 Watts. I figured 55 Watts was close enough when I posted this time. A calculation is no better than the data used, and I don't know whether those early published estimates from IBM are still meaningful.
  • Reply 34 of 148
    where did MOSR get the info about this '975' chip anyway?... i have been following their rumor pattern and for the most part this mysterious 975 chip is only spoken about by them. they are even claiming shipping times...



    not sure what's up.
  • Reply 35 of 148
    Quote:

    Originally posted by Ensign Pulver

    This as always been my secret hope too, that the 970fx was only ever intended for the now aborted Jan/Feb 2.5 GHz Rev. Bs with the 975 as the 3GHz part for the Rev. Cs at WWDC 04. In other words, when the 970fx fell on its face early this year, Steve said screw it, we'll just sit tight with the current Power Macs and go straight to the 3GHz 975 models in June.



    This scenario would allow for the inclusion of PCIe and other mobo improvements and most importantly let Steve make a big deal out of the G5's one year anniversary at his keynote. If he can hold a bakeoff with a Dual 3Ghz G5 with a PCIe native 256MB RV420 then all will be instantly forgiven.




    This is my understanding of the situation based on information from friends at IBM. The most recent emails I've received indicated that Apple had ordered large quantities of 2 and 2.5GHZ 970FX chips, but IBM was unable to provide much of anything except a few 2GHZ chips. They've gotten enough out for Apple to start getting xServes out. I have a feeling that Apple had hoped to have 2.5GHZ PowerMacs and xServes. They also cancelled a large order for 1.6 GHZ 130nm chips before manufacture when this news arrived. My guess would be those may have been meant for the rumored iMac G5s. I haven't heard if they've re-ordered a large quantity of those or not.



    However, the Power5 derivative is on schedule and was the planned 3GHZ chip.



    Also of note, the VX project was cancelled in January (mislabeled as the 750VX). This project was funded by Apple to produce a new consumer end chip to replace Motorola's 7400 series chips. Apparently, Apple is reluctant to cut ties with Moto. They like some of the chips they are presently working on and have regained confidence in them since the semiconductor division split was announced - coupled with the opening of the Crolles fab. IBMers in Fishkill have heard that Apple expects Moto to hit 90nm this summer in quantity - coupled with some other processor improvements. As to what they are, I don't know. I would guess we'd see a L2 cache increase.



    I haven't heard, but I expect the 90nm Power5 derivatives (they are not badged officially yet) to have 1MB of L2 cache. I fully expect them to be in the next PowerMac revision - and I expect that announcement at WWDC.
  • Reply 36 of 148
    snoopysnoopy Posts: 1,901member
    This report sounds real to me. IBM's problems with the 970FX killed plans for a speed bump in the Power Mac line, but we will see the Power 5 derivative in new Power Macs sometime after WWDC. The cancelled 1.6 GHz 970 order likely means the new iMac will have the 970FX, and Apple is confident that IBM will solve their manufacturing problem soon. The 970FX should be cheaper than the 970, and with the Power 5 derivative in the Power Macs there will be no competition in performance. Good news, but not great news.



    I have mixed feelings about cancelation of the Mojave VX project, IBM's G4.
  • Reply 37 of 148
    Quote:

    Originally posted by Mr. MacPhisto

    This is my understanding of the situation based on information from friends at IBM. The most recent emails I've received indicated that Apple had ordered large quantities of 2 and 2.5GHZ 970FX chips, but IBM was unable to provide much of anything except a few 2GHZ chips. They've gotten enough out for Apple to start getting xServes out. I have a feeling that Apple had hoped to have 2.5GHZ PowerMacs and xServes. They also cancelled a large order for 1.6 GHZ 130nm chips before manufacture when this news arrived. My guess would be those may have been meant for the rumored iMac G5s. I haven't heard if they've re-ordered a large quantity of those or not.



    However, the Power5 derivative is on schedule and was the planned 3GHZ chip.



    I haven't heard, but I expect the 90nm Power5 derivatives (they are not badged officially yet) to have 1MB of L2 cache. I fully expect them to be in the next PowerMac revision - and I expect that announcement at WWDC.




    this is a very interesting report (?).. it can also lead one to consider that possible the 970 chips may totally be sidelined. if there is a better design in the POWER 5 derivative then producing all chips based on it might be the plan. just because it could scale to 3 Ghz doesn't mean it can only be clocked as that. if it is a good reliable design it could be utilized in products running at lower clock rates.
  • Reply 38 of 148
    programmerprogrammer Posts: 3,467member
    Quote:

    Originally posted by ghiangelo

    this is a very interesting report (?).. it can also lead one to consider that possible the 970 chips may totally be sidelined. if there is a better design in the POWER 5 derivative then producing all chips based on it might be the plan. just because it could scale to 3 Ghz doesn't mean it can only be clocked as that. if it is a good reliable design it could be utilized in products running at lower clock rates.



    The 970FX is still a better chip for portable/embedded applications because it is smaller. At the same clock rate it would therefore generate less heat and consume less power. The "975" will be a much larger beast with more cache and possibly SMT, thus inappropriate for portables even at 90nm. I'd expect the high end chip to only show up in the PowerMac.



    I'm not holding my breath for all this to come true, but it would be nice.
  • Reply 39 of 148
    The Power5 derivative should be able to run in the PowerBooks and the PowerMacs. After it moves over to 65nm a new dual-core chip should become available, coupled with multi-threading.



    I have a feeling that FreeScale (the spunoff Moto semiconductors unit) may be able to provide some impressive chips for the consumer market in the near future. The people in IBM are well aware of them and consider them to be very serious competition in the embedded market - but are aware that Apple also is impressed with proposed future offerings. I can't provide anything firm because I don't know anyone over there - but persistant rumors indicate they will move to 90nm this summer and there's a 7400 series dual core chip in the works. I hope they've got something good in the near future, but I won't hold my breath. The cancellation of the VX project (which could have been fabbed in volume in March - though the 90nm problems would have delayed that as well) indicates to me that Apple has some kind of good reason to continue relations with Moto/FreeScale. Apple spent millions developing that chip with IBM. Some of that development will go into the system-on-a-chip project that IBM is working on for Apple. Even so, that's a lot of time and money for something that won't see regular production.
  • Reply 40 of 148
    programmerprogrammer Posts: 3,467member
    Quote:

    Originally posted by Mr. MacPhisto

    The Power5 derivative should be able to run in the PowerBooks and the PowerMacs. After it moves over to 65nm a new dual-core chip should become available, coupled with multi-threading.



    Lets get 90nm going before we start talking about 65nm. I know its coming, and I know what they're planning for it, but first they have to deliver on their current plans.



    I hope Freescale has gotten its act together. Dual core doesn't interest me, but an on-chip memory controller and HT or RIO I/O bus with a fast (i.e. 2 GHz) G4 core would be a very compelling PowerBook chip.
Sign In or Register to comment.