Is the current rate of PPC CPU upgrades sustainable?

245

Comments

  • Reply 21 of 96
    thttht Posts: 3,954member
    Quote:

    Originally posted by stingerman

    Dude, you got to do your own due diligence.



    I've tried finding it but my powers of diligence appear to be weak. Why don't you do me a favor and just tell me since you know which white papers they are?



    Quote:

    There is nothing to doubt, its a fact and part of the design. Enough said.



    It's a stupid design and doesn't have any benifit whatsoever, possibly actually degrading performance for single threaded apps. The interconnects need to support various cache coherency protocols for SMP and a shared state is one of them. But the shared state doesn't mean that one processor is literally using the cache of another processor, just that there will be a direct transfer from one processor's cache to another processor's cache, instead of routing through main memory.
  • Reply 22 of 96
    Well here is just a brief mention of it on the Apple site:



    Quote:

    In addition to providing fast access to main memory, this high-performance frontside bus architecture lets each G5 discover and access data in the other processor?s cache.



    http://www.apple.com/g5processor/architecture.html



    There is more detail in the developer references in the PDF documents.
  • Reply 23 of 96
    thttht Posts: 3,954member
    Quote:

    Originally posted by stingerman

    Well here is just a brief mention of it on the Apple site:





    http://www.apple.com/g5processor/architecture.html



    There is more detail in the developer references in the PDF documents.




    This sounds like a reference to a shared cache coherency protocol.
  • Reply 24 of 96
    nr9nr9 Posts: 182member
    it is heartbreaking to see such a relevant topic decay into an argument on cache coherency.
  • Reply 25 of 96
    thttht Posts: 3,954member
    Quote:

    Originally posted by Nr9

    it is heartbreaking to see such a relevant topic decay into an argument on cache coherency.



    It's a difficult question to answer because we have no data to base a rate of PPC improvements on.



    Overall, it really just depends on IBM being able to jump to 90 nm, 65 nm and then 45/50 nm every 18 to 24 months. If IBM ships a 90 nm 970 in Q2 04, we can probably expect a 4 GHz processor at 65 nm in Q1 2006 if IBM is good. The big problem is that R&D for new fabs doubles every generation. If IBM doesn't grow big enough, they won't have enough money to move on to the next node.



    Hence, IBM's current activities in getting as many customers as they can. If IBM buys out Texas Instruments' fab business or something similar, I'd probably say yes to 65 nm in 2006. If not, it'll probably be 2007+. It may be that the US gov't will prop up 2 companies and fund development for next gen nodes.
  • Reply 26 of 96
    nr9nr9 Posts: 182member
    Quote:

    Originally posted by THT

    It's a difficult question to answer because we have no data to base a rate of PPC improvements on.



    Overall, it really just depends on IBM being able to jump to 90 nm, 65 nm and then 45/50 nm every 18 to 24 months. If IBM ships a 90 nm 970 in Q2 04, we can probably expect a 4 GHz processor at 65 nm in Q1 2006 if IBM is good. The big problem is that R&D for new fabs doubles every generation. If IBM doesn't grow big enough, they won't have enough money to move on to the next node.



    Hence, IBM's current activities in getting as many customers as they can. If IBM buys out Texas Instruments' fab business or something similar, I'd probably say yes to 65 nm in 2006. If not, it'll probably be 2007+. It may be that the US gov't will prop up 2 companies and fund development for next gen nodes.




    i think IBM should just focus on architectural improvements, and leave fabrication technology to foundries.



    Both TSMC and UMC have working 9 layer copper 90nm processes, they are likely to leapfrog the rest of the industry in semiconductor processes, UMC already has plans for standard 193 light lithography down to 45nm. if they can do that, imagine what they can do with EUV
  • Reply 27 of 96
    Quote:

    Originally posted by Nr9

    i think IBM should just focus on architectural improvements, and leave fabrication technology to foundries.



    Both TSMC and UMC have working 9 layer copper 90nm processes, they are likely to leapfrog the rest of the industry in semiconductor processes, UMC already has plans for standard 193 light lithography down to 45nm. if they can do that, imagine what they can do with EUV




    IBM wants to compete as a foundry as well as a design house. They've always had an industry-leading research group, and they should be able to compete with the likes of TSMC and UMC... especially since they are offering their design services to clients of their foundries. Makes a lot of sense this then subsidizes their own processor projects.
  • Reply 28 of 96
    nr9nr9 Posts: 182member
    IBM will not be able to compete.



    for fab technlogy, specialization is much more efficient.
  • Reply 29 of 96
    Quote:

    Originally posted by THT

    To this day, I have not found any evidence that IBM ever shipped a 180 nm Power3-III. They were slated to be in IBM's Nighthawk 2 servers, but it appears the Power4 came out in time and IBM decided to stick with the Power3-II for their entry servers.



    If you can find evidence of the Power3-III in an actual product, let me know.




    That's presicely what my research dug up too. POWER3-III seems to be scrapped in favour of POWER4 witch should have blown any POWER3-III out of the water if the preliminairy specs for POWER3-III and the actual specs for POWER4 is compared. A stop gap of sorts between POWER3-II and POWER4 was the RS64-IV (S-Star) witch is a complete 32/64 bit PowerPC and 64 bit PowerPC-AS implementation running at similar performance to POWER3-III.



    The S-Star and POWER3 was the foundation on witch POWER4 was built.



    Back to the topic..



    I have every confidence that IBM can sustain its PowerPC development at least under the five year contract they have with Apple. The roadmaps that are available, the new plant and the enourmous ammount of resources that IBM invests in Linux is evidence of that. IBM seems very confident that the PowerPC design is the right way to go. And.. why take IBM's word for it? Take Nintendo's, Sony's or even Microsoft's word for it. They too are investing a lot of money and prestige in using PowerPC in the future. In the case of Sony and Microsoft, it's not entirely certain that Cell and Xbox-2 will use PPC technology but it's like a IBM rep said "We don't have any other processor technology".



    My hope for IBM itself is that they do try to get back at Intel and Microsoft, and Linux on PPC is the best way for them to do that. Everything developed in house. Every invested dime goes back into the company, everything from processors, chipsets, operating systems, servers, clients and services are developed in house. Not a single penny to the likes of Intel or Microsoft.



    Or.. that might just be _my_ dream
  • Reply 30 of 96
    powerdocpowerdoc Posts: 8,123member
    Quote:

    Originally posted by Nr9

    IBM will not be able to compete.



    for fab technlogy, specialization is much more efficient.




    Apple will not be able to compete



    for hardware technology, specialization is much more efficient
  • Reply 31 of 96
    Quote:

    Originally posted by Nr9

    i think IBM should just focus on architectural improvements, and leave fabrication technology to foundries.



    Both TSMC and UMC have working 9 layer copper 90nm processes, they are likely to leapfrog the rest of the industry in semiconductor processes, UMC already has plans for standard 193 light lithography down to 45nm. if they can do that, imagine what they can do with EUV




    I disagree, IBM is the only company I know of that actually does research and development in chip and fab technologies (SOI, that copper thingy, use of geranium) and licenses it to other chip manufacturers. Any other chip manufacturer owes part of their advancement to IBM. IBM is a specialist in this field.



    Which brings up another question. What became of that geranium technology to improve chips.
  • Reply 32 of 96
    Quote:

    Originally posted by Nr9

    IBM will not be able to compete.



    Heh. Well there are many people, some of whom have a whole lot more money to spend than you do, who seem pretty confident that not only can IBM compete but that they are a good bet going forward.
  • Reply 33 of 96
    709709 Posts: 2,016member
    Quote:

    Originally posted by Leonard

    Which brings up another question. What became of that geranium technology to improve chips.



    I beleaves that this tech is being used in IBM's plants for tele-comm chips only. So I wouldn't expect to seed this process in a desktop chip anytime soon.
  • Reply 34 of 96
    nr9nr9 Posts: 182member
    Quote:

    Originally posted by Powerdoc

    Apple will not be able to compete



    for hardware technology, specialization is much more efficient




    to some degree, this is true for price/performance



    apple sells its computers because of a superior operating system



    Also, there is not much to do in computer making as the performance is usually based on what the processor manufactuer does.
  • Reply 35 of 96
    nr9nr9 Posts: 182member
    Quote:

    Originally posted by Programmer

    Heh. Well there are many people, some of whom have a whole lot more money to spend than you do, who seem pretty confident that not only can IBM compete but that they are a good bet going forward.



    then let them make bad investments
  • Reply 36 of 96
    @Nr9



    Laconic, aren't we?
  • Reply 37 of 96
    nr9nr9 Posts: 182member
    u kno, they can put all the money they want in IBM's fabs, but with the quality of american education in recent years, they are going to work really really hard to try to surpass TSMC and UMC.
  • Reply 38 of 96
    Quote:

    Originally posted by Nr9

    u kno, they can put all the money they want in IBM's fabs, but with the quality of american education in recent years, they are going to work really really hard to try to surpass TSMC and UMC.



    Haha... a couple of new fledgeling companies think they can surpass IBM. These companies will have to catch up to IBM first. When have these companies designed a desktop CPU. What's with Taiwan these days, they think they're the next Japan. See where Japan is now... that will be Taiwan in a few years.



    Get back to us when they have developed a CPU like AMD, Intel, or IBMs current CPUs.



    By the way, first there is nothing wrong with American education, and second IBM is an international company and has more than American employees.
  • Reply 39 of 96
    bigcbigc Posts: 1,224member
    seems like someone is trying to catch some more fish....

  • Reply 40 of 96
    powerdocpowerdoc Posts: 8,123member
    Fabbing huge CPU like the PPC 970, the P4 or the Athlon 64 is the most difficult thing to fab.



    One of the more easy one is to fab memory.



    IBM is really competitive for fabbing and he cheers his indepedance. It's better to not be entirely dependant of an other companie. I hope that IBM will continue to fab his owns chips.
Sign In or Register to comment.