Processor Shopping / CPU-manufacturer`s worldwide

Posted:
in Future Apple Hardware edited January 2014
1-2 years ago I encouraged you to agnoledge that Intel/AMD/IBM and Motorola are not the only companies in the semiconductor-buisiness who are capable of producing CPU-silicon.



I intend to repeat this now and share a few thoughts about it because Mac-users who wonder what companies could deliver future-Mac-CPU`s only look at the 4 afore-mentioned manufacturers and that`s just as ignorant like the Win-users only know 2 CPU-manufacturers: Intel & AMD.



HP had a RISC-processor line now sold to Intel and ARM is specialized at the PDA-market and as far as I know owned by Intel.

ARM however started to be the CPU of desktop computers: the ACORN ARCHIMEDES. The Archimedes appeared in the early 1990-ies and it`s processing power was stunning and far beyond any other desktop computer.

Today ARM just like HP is surely no longer a competitor.



So what else is there?



MIPS for example. But they have low clock-rates and people would even more worry about being behind with a 450Mhz in 2002. The Playstation 2 is equipped with a MIPS-CPU but then again the PS2 also has a whole bunch of specialized co-processors. Good for a console but certainly not for a desktop machine.

MIPS is pretty uninteresting I think.



Still there are Toshiba and Samsung. I know nothing about the capabilities of Toshiba to develop and manufacture CPU`s but Jay Miner who developed the original Amiga platform (A1000.A500,A2000) designed IC`s for cardiac pacemakers before designing the Amiga`s architecture and custom chips. So there`s surely talent for that also in engineers who currently design lower performance and lower cost embedded-chips.

However I would assume that for competing in the CPU-arena the company would require a advanced manufacturing process. This might not be the case for companies who yet only design .. let`s say circuits for mobil phones or washing machines.

But take Samsung for example: The manufacturered the then-highly-acclaimed DEC Alpha CPU. Even if this has vanished it still proves they are capable. Don`t forget DEC themselves...

Apple already hmade a larger investment in Samsung (for supply of TFT`s I think).



I`m pretty sure there`s especially a bunch of Japanese and overall asian companies that could be capable enough.



What else? nVidea is looking for new markets. They just released a x86-chip-set that`s really fast, innovative and stable and they succeeded in the Graphics-card buisiness which is highly competitive just like the CPU-buisiness.. nVideas silicone has already reached transistor-counts and advanced processes that would also look good for a CPU to have.

nVidea could at one point form a joint-venture with another company -maybe via a daughter-firm- and develop a CPU.



There`s more than 4 companies. I must admit that I consider AMD the most interesting of what I know - but then again I don`t know much about the corporations other then those who are already "in the neighbourhood".

Still AMD because they have the 64-bit CPU waiting and it`s unclear if that will lead to anything because there would be compatibility probs with Windows. It would depend of Microsofts goodwill and the goodwill of the other developers. With Linux-only it won`t take off but AMD surely want to deploy the CPU somewhere.

Additionally from what I understand the AMD-chips you can buy now (Athlon / Duron) are already completely different from Intels and use a compatibility-layer..





What other comanies would be feasible or at least interesting for Apple or not from your point of view?



NOT for speculation - just theoretical.

Comments

  • Reply 1 of 20
    I'll bite. It's been kinda boring around here. Isn't July (MWNY) right around the corner?



    VIA makes a µprocessor, but I don't think they'd be interested in Apple business. I think they want to make a chipset (or board) to go with it, which they wouldn't be able to do with Apple.
  • Reply 2 of 20
    To be honest I wouldn`t be interested in VIA. I think although I could be wrong that VIA owns CYRUS (or wasn`t that the name) or at least manufactures other low-end/low-cost-x86-compatible CPU`s.

    Not very interesting.

    Apart from that I`m not even convinced that they are are good at what they do. VIA chipsets have been badmouthed and it might be founded by facts.
  • Reply 3 of 20
    I forgot to reply to your statement that MWNY is around the corner anyway:



    1. This was not ment to be a thread for speculation what will or won`t happen and when but just a look around what could be concerning the LONG TERM strategy of Apple.. maybe for the next 5-10 years.



    2. I`m pretty convinced that if there will be something in that direction we won`t see it in July. Oops, now I get carried away with speculation
  • Reply 4 of 20
    [quote]Originally posted by babel-syndrome:

    <strong>To be honest I wouldn`t be interested in VIA. I think although I could be wrong that VIA owns CYRUS (or wasn`t that the name) or at least manufactures other low-end/low-cost-x86-compatible CPU`s.

    Not very interesting.

    Apart from that I`m not even convinced that they are are good at what they do. VIA chipsets have been badmouthed and it might be founded by facts.</strong><hr></blockquote>



    If I recall Via bought Cyrix and the Nat Semi's processor line, though I think they have mainly developed the Cyrix line (which if I recall was in many ways the less ambitious of the two).



    To be fair to Via part of the bad rep on their chipsets is all of the bleeding edge stuff they are rushing out before it's fully cooked to keep up with the market.
  • Reply 5 of 20
    &gt; To be fair to Via part of the bad rep on their

    &gt; chipsets is all of the bleeding edge stuff they &gt; are rushing out before it's fully cooked to keep &gt; up with the market.



    That`s no excuse - the opposite is the case: I think that`s bad policy and certainly not one Apple should follow.

    Actually I find that Apple or Motorola did just that when they released the G4. Apple could have used beefed up G3`s for some time or a G4 (or another MP-capable 604-successor) without Altivec that IBM could have procided.
  • Reply 6 of 20
    eskimoeskimo Posts: 474member
    Before you can discuss the companies capable of producing leading edge processors you must realize what is necessary. The requirements for a company to produce a performance leading microprocessor is twofold. First a company must have the resources to design the MPU. This consists of having a talented design staff and the time and resources to bring a design to market. The second requirement is the advanced manufacturing capability necessary to produce a high performance MPU. Now that we have set the basic necessities lets examine the 'players'.



    AMD, IBM, Intel, and Motorola are all known as IDMs or Integrated Device Manufacturers. The integrated nomenclature designates them as companies who have the resources to both design and manufacture their own designs.



    The two other types of companies that could be considered for this discussion are so called 'fabless' or 'fab-lite' companies which have the ability to design products but rely on 3rd party companies for the actual manufacture of their products. These 3rd party companies are known as foundries, they concentrate on manufacturing technology and have little to no design capabilities of their own.



    Now let us discuss problem with any companies besides the big 4 making a serious competitor. There are not that many design engineers in the world that have the experience and expertise to produce a leading edge design. It takes hundreds of designers to produce designs today and it takes at least 3 years of solid work to produce working samples. Few companies have the resources to takes such as risk in a market that is already dominated by the aforementioned companies.



    The design of a high performance microprocessor is much different then laying out an ASIC with standard cells or doing a semi-custom design like Nvidia does with its GPUs. Sure Nvidia's transistor counts are enormous but most of their design is simply drag and drop with previously completed components. Designers at AMD and Intel spend years huddled over their screens optimizing critical speed paths my manually resisizing transistor sizes and layout in a full custom environment. Sure they have automated tools to asist but there is a lot of human participation. That is why Nvidia and ATI can produce new designs every 18 months (Nvidia has 3 parallel design teams, that is why new products come out every 6 months) rather than AMD and Intel which require 3-5 years between generations. The only rumor I have heard that seems plausible is that there is a group of designers in Russia known as Elbrus that have supposedly developed an extraordinary design known as the E2K which would use an EPIC VLIW architecture similar to Intel's Itanium. The only reason this seems plausible is that the former Soviet Union contains many talented engineers and scientists that have remained since the collapse of that superpower. Even if they do exist they face the same problem as every other design group that isn't an IDM, how to manufacture their design.



    Which brings us to the other side of the coin. There are only a few foundry companies in the world that have the manufacturing facilities capable of producing high end parts. And fewer still that have the proprietary knowledge and processes necessary to ensure high speed operation. The big 2 among the foundries are TSMC and UMC. Neither have demonstrated the ability to manufacture high speed complex logic in the past. TSMC just this past week was able to allow Via to announce a 1GHz Cyrix processor on a .13 micron process. Remember AMD and Intel were able to hit 1GHz on .25 and .18 processes respectively. Since the Cyrix design is not a high IPC one but one intended for frequency scaling we can only surmise that the process knowledge that TSMC has is not viable for high speed logic. This is the reason why you now hear about TSMC and UMC announcing joint ventures and collaborations with companies who do have the process know how to produce these parts. UMC has announced joint ventures with Infineon and AMD. TSMC has announced collaboration with STMicro and Toshiba.



    I wish I had more time but in truth all indications are that in the future there will be fewer companies involved in high end MPUs and not more. My prediction is that Motorola will be the first to drop out of the race, having already put in place a fab-lite policy and reducing their capital expenditures in an effort to stem the hemmoraging of their cash flow.



    [quote]If I recall Via bought Cyrix and the Nat Semi's processor line, though I think they have mainly developed the Cyrix line (which if I recall was in many ways the less ambitious of the two).



    <hr></blockquote>



    Cyrix was National's processor line, Via also bought out IDT and Centaur who both designed low performance x86 processors.
  • Reply 7 of 20
    zosozoso Posts: 177member
    [quote]Originally posted by babel-syndrome:

    <strong>What else? nVidea is looking for new markets. They just released a x86-chip-set that`s really fast, innovative and stable and they succeeded in the Graphics-card buisiness which is highly competitive just like the CPU-buisiness..</strong><hr></blockquote>



    Innovative? Yes, AFAIK they were the first to come out with a HyperTransport implementation, and also, their dual channel DDR approach is very interesting--think of the benefits that dual channel RAM configurations brought to the whole RamBus camp... But this doesn't excuse the fact tha nVIDIA chipsets are not top performers at all. Interesting and innovative ideas, that's for sure. But if I were looking for performance I'd go KT333 without a thought. Which introduces fairly well the next issue:



    [quote]Originally posted by babel-syndrome:

    <strong>Apart from that I`m not even convinced that they are are good at what they do. VIA chipsets have been badmouthed and it might be founded by facts.</strong><hr></blockquote>



    Angry because every time they introduce a new chipset they have to release a .a version fixing all the issues a couple of months later? Ok, I totally agree with you. But--and this happened two times in these past years, with the 133SDR and the 266DDR chipsets--once they fix it it ROCKS! Both times those revised chipsets became performance leaders in their respective markets, so the assupmtion that VIA ain't good at what it does is completely wrong. Even now, if you want the best AMD system around you'd be crazy to choose anything than VIA's KT333. In a six months time frame? Well, take a look at the refernce Hammer mobos from VIA: AGP 8x, USB 2, SerialATA... Not bad from a company that doesn't have a clue, uh?



    So, does this mean VIA should be the way to go to manufacture a new revision of a PPC design? No, not by far.

    As has been pointed out, you need more than a good fab to get a really performing CPU out of the door. A good thing to start with would be a good design, and the MIPS could be a good hypothetical candidate--since it is a licensed CPU design, just like, IIRC, ARM is (you need to find a fab to build it for you after you get a license).



    As for Apple going 100% IBM with the next CPU--I don't know... I suppose that in order to include SIMD on their future products IBM will have to change the manufacturing in at least one fab quite a bit. Would this make sense for IBM? Would Apple's demand for these CPUs justify the production costs? I'm just asking, I'd like to know that from somebody with some real knowledge of the biz...



    [quote]Originally posted by babel-syndrome:

    <strong>Still AMD because they have the 64-bit CPU waiting and it`s unclear if that will lead to anything because there would be compatibility probs with Windows. It would depend of Microsofts goodwill and the goodwill of the other developers. With Linux-only it won`t take off but AMD surely want to deploy the CPU somewhere.</strong><hr></blockquote>



    Err, M$ has already announced it's going to offer a native x86-64 version of Windows. And I thik Suse was the first distro to announce support from the Linux camp. Furthermore, it looks like that with a 40% performance advantage on an Athlon XP at the same clock speed executing non-optimized 32 bit code AMD has plenty of time to get developers to optimize their code for the Hammer platform...



    My dreams are populated with POWER4s and Hammers running OS X--anything else will just be a pitiful compromise... Oh, I hear another name for that is "reality"...



    ZoSo
  • Reply 8 of 20
    roosterrooster Posts: 34member
    Not to forget is Texas Instruments currently manufacturing the SPARC cpu for Sun.



    Rooster
  • Reply 9 of 20
    junkyard dawgjunkyard dawg Posts: 2,801member
    The G5 is going to suck, that's reality. It's been in the design phase for far too long....originally it was planned for 1-2 Ghz, which would have been great last year, but now it would only be catching up to Intel's CURRENT Pentium, and within a year Apple will be right back where they are now, except there will be no "G6" to look forward to.
  • Reply 10 of 20
    The next proc in Apple's lineup will be IBM's 'Cell' project.



    Dajjal is no Jet Powers
  • Reply 11 of 20
    roosterrooster Posts: 34member
    My prediction is that Motorola will be the first to drop out of the race, having already put in place a fab-lite policy and reducing their capital expenditures in an effort to stem the hemmoraging of their cash flow.



    Motorola joins Philips and ST in the 300mm wafer Crolles2 project, significantly enhancing the strength of the existing alliance. This expands the agreement announced last month by Philips, ST and TSMC to jointly develop CMOS process technology. The combined resources of the four companies will be dedicated to development of future generations of CMOS technology from the 90 nanometer node down to 32 nanometers over the next five years. Such reduced circuit geometry sizes will allow the partners to continue to meet customer demand for integrating more intelligence into smaller packages.



    the whole article is here:

    <a href="http://www.motorola.com/mediacenter/news/detail/0,1958,1308_988_23,00.html"; target="_blank">http://www.motorola.com/mediacenter/news/detail/0,1958,1308_988_23,00.html</a>;



    rooster
  • Reply 12 of 20
    Here is a <a href="http://www.geocities.com/Athens/6270/vendors.html"; target="_blank">list</a> of super computing platform venders. You will see many framiliar names. But, You will also see some lesser known names there.
  • Reply 13 of 20
    &gt; The two other types of companies that could be considered for this discussion are so called 'fabless' or 'fab-lite' companies



    Which is becoming hip lately and I think this trend is just starting off. Fabs could sell their capacities on a stock-echange-like market and therefore maximize usage. That again would reduce cost for the fab itself and the customers. Problem might be that the companies must to some extend have a guaranteed manufacturing capacity - long-term.... But if the trend becomes a development and market reality then that could improve chances for new innovative companies to get into the market-place.





    &gt; but in truth all indications are that in the future there will be fewer companies involved in high end MPUs and not more.



    That`s what I thought.. Yet, fewer manufacturers and therefore 3rd-party manufacturing becoming common-place even for yesterdays heavyweights could maybe result in a bigger variety of design-companies; because it`s then more realistic to achieve a working product if you must only develop it while ruling out that a your product although finalized cannot be manufacturered because you don`t have a fab that`s capable. Just mind the design and then rent-a-fab.





    &gt; The only rumor I have heard that seems plausible is that there is a group of designers in Russia known as Elbrus



    I heard that before and it was less a rumor but fact 2-4 years ago. The last thing I heard of them is that they where looking for venture capitalists. Maybe the guys got hired by other companies by now.

    The ARM-group however is british and there are also many technically skilled people in the scandinavian countries.

    I might add: Don`t underestimate India. They are already building cars now and the programming skills are well known. Maybe there will be computer hardware at some poit. I wouldn`t be too surprised.



    &gt; But this doesn't excuse the fact tha nVIDIA chipsets are not top performers at all.



    According to german Keyboards-Magazin it IS indeed performing very good. It was compared being equal to the best AMD-chipsets and equal to Intel-chipsets.



    So making CPU`s more complicated seems to increase possibilities for someting to go wrong. Maybe a very simple design which is scalable quicker would be something to think about (like what RISC is for code). While we`re talking of simplifying: why not let Apple`s next CPU be built by Fisher Price? I`m sorry but reading of the Power4 (unaltered) in a Mac I think the legoMac is just as likely.
  • Reply 14 of 20
    I must add to this statement of mine:



    &gt; Fabs could sell their capacities on a stock-&gt;

    &gt; echange-like market and therefore maximize &gt;

    &gt; usage.



    If thought of that as for IC manufacturers in general. The high-end manufacturers will surely not need that for lack of competition.
  • Reply 15 of 20
    Just to clarify two points :



    HP did not sold it's cpu division to Intel. They begun working on their next gen. chip (that was supposed to be pa-risc 3.0) and Intel found that their job was cool, so they decided to make their next gen chip be HP's one : it's the IA-64 architecture (still called pa-risc 3.0 by most engineers within hp ). So the processor division was not sold, it just disappeared because they let Intel do the finitions, the fab and the marketing.



    As for MIPS, they make all the processors for SGI workstations, and they do fine, thank you



    If you want to see how pa-risc and mips do, have a look at this graph :

    <a href="http://www.aceshardware.com/SPECmine/index.jsp?b=0&s=3&v=4&if=0&r1f=0&r2f=0&m1f=0&m2f=0 &o=0&o=1" target="_blank">http://www.aceshardware.com/SPECmine/index.jsp?b=0&s=3&v=4&if=0&r1f=0&r2f=0&m1f=0&m2f=0 &o=0&o=1</a>



    as for the g4, they lack good frequencies, but they are very MHz efficient...





    oh, and btw : this is my first post (been lurking a while...)



    [ 06-10-2002: Message edited by: DaMouloud ]</p>
  • Reply 16 of 20
    bigcbigc Posts: 1,224member
    [quote]Originally posted by There is no g5:

    <strong>The next proc in Apple's lineup will be IBM's 'Cell' project.



    Dajjal is no Jet Powers</strong><hr></blockquote>



    Yeah, cell been metioned before. hmmm

    <a href="http://www.siliconstrategies.com/story/OEG20020322S0068"; target="_blank">http://www.siliconstrategies.com/story/OEG20020322S0068</a>;
  • Reply 17 of 20
    Eskimo, are you certain AMD hit 1GHz on .25um? I seem to remember them needing until .18um, like Intel.



    From memory:

    [quote]AMD Athlon "K7": 500-700MHz, Slot A, .25um

    AMD Athlon "K75": 700MHz-1000MHz, Slot A, .18um

    AMD Athlon "Thunderbird": 800MHz-1400MHz, Socket A, .18um

    AMD AthlonXP: 1333MHz-1733MHz (1500+ - 2100+) on .18um; new high-end 1800MHz (2200+), .13um<hr></blockquote>



    [ 06-10-2002: Message edited by: TheAlmightyBabaramm ]</p>
  • Reply 18 of 20
    spartspart Posts: 2,060member
    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>Eskimo, are you certain AMD hit 1GHz on .25um? I seem to remember them needing until .18um, like Intel.



    From memory:





    [ 06-10-2002: Message edited by: TheAlmightyBabaramm ]</strong><hr></blockquote>



    um? Dontcha mean µm?
  • Reply 19 of 20
    [quote]um? Dontcha mean µm?<hr></blockquote>



    Yes, but I'm too lazy to use the proper combination of keys to make the proper symbol. Even if I knew it, I probably wouldn't use it, because I don't care that much.
  • Reply 20 of 20
    eskimoeskimo Posts: 474member
    [quote]Originally posted by TheAlmightyBabaramm:

    <strong>Eskimo, are you certain AMD hit 1GHz on .25um? I seem to remember them needing until .18um, like Intel.



    From memory:





    [ 06-10-2002: Message edited by: TheAlmightyBabaramm ]</strong><hr></blockquote>



    Damn, I'm sorry, I was wrong.

    /me hangs head in shame



    I was getting the switch to .18 confused with the fact that we switched designs (K75 to Tbird) and process (but still at .18) when we reached 1GHz. That was when we phased out the Athlon classic and shipped Tbird and coincidentally when I first started working at AMD We change to .18 for the release of the 750MHz Athlon classic.
Sign In or Register to comment.