Become a fan of Slashdot on Facebook

 


Forgot your password?
Close
typodupeerror
×
AMD Graphics Hardware

AMD, Nvidia Reportedly Tripped Up On Process Shrinks 230

Posted by Soulskill from the stupid-physics-getting-in-the-way dept.
itwbennett writes: In the fierce battle between CPU and GPU vendors, it's not just about speeds and feeds but also about process shrinks. Both Nvidia and AMD have had their move to 16nm and 20nm designs, respectively, hampered by the limited capacity of both nodes at manufacturer TSMC, according to the enthusiast site WCCFTech.com. While AMD's CPUs are produced by GlobalFoundaries, its GPUs are made at TSMC, as are Nvidia's chips. The problem is that TSMC only has so much capacity and Apple and Samsung have sucked up all that capacity. The only other manufacturer with 14nm capacity is Intel and there's no way Intel will sell them some capacity.
This discussion has been archived. No new comments can be posted.

AMD, Nvidia Reportedly Tripped Up On Process Shrinks More

AMD, Nvidia Reportedly Tripped Up On Process Shrinks

Comments Filter:

  • I got an idea (Score:4, Insightful)

    by Anonymous Coward on Tuesday January 06, 2015 @02:37PM (#48748303)

    Build your own fab

    • Re:I got an idea (Score:4, Funny)

      by Anonymous Coward on Tuesday January 06, 2015 @02:44PM (#48748375)

      ain't nobody got time for that!

    • Re:I got an idea (Score:5, Informative)

      by Luckyo ( 1726890 ) on Tuesday January 06, 2015 @03:09PM (#48748647)

      And get owned by intel's illegal monopolistic deals with OEMs?

      Been there, done that. Signed: AMD.

      • Dear OEM X,
        We are intrigued by your new technology Y. We have begun production of our own version we copied straight from your designs and need you to sign the attached licensing agreement giving us complete autonomy to use your patents as we see fit. We very much enjoy our relationship and would hate to see OEM X lose its ability to purchase our chips at base rate and be forced to seek them from 3rd party suppliers.

        Sincerely, Intel

    • Maybe yes, maybe no.

      Fabs are incredibly capital intensive. If you build one you sort of need to run it at full hilt to make money off of them. And then you are on the tread mill of always having to build another one to have the latest and greatest.

      I am not saying they should not build it, just that it is a completely different beast then design and marketing of their chips. You can't do it half way. FYI, almost everybody agrees that the reason why the x386 dominates today is not because Intel had the best o

      • Re: (Score:2)

        by itzly ( 3699663 )
        It's both things. The x86 is a good design, and the intel fabs are good.

        • Re:I got an idea (Score:5, Informative)

          by alexander_686 ( 957440 ) on Tuesday January 06, 2015 @03:59PM (#48749321)

          I read an interview of John Sculley, ex Apple CEO, 10 years after he left Apple. (There is a link to it somewhere here on Slashdot. ) He said that one of his great mistakes was to choose Motorola's PowerPC RISC chip over Intel's x386 design. All of Apple engineers pulled for the Motorola chip, pointing to its better architecture. And it was true, for a given bit of silicon Motorola was better than Intel.

          Except that Intel's fabs were so much more efficient than Motorola's, they were able to deliver more powerful chips at a cheaper price even with poorer design. Which allowed them to make fat profits, which they plowed back into newer fab plants, which let them sprint past Motorola.

          • Wasn't it Gil Amelio who made that comment? But at the time, Intel was at par w/ PowerPC, and slower than other CPUs, like PA-RISC and Alpha. What made Intel catch up w/ RISC CPUs was Microsoft using the NT kernel to replace the Windows 9x kernel, and as a result, having an SMP capable mainstream OS. So now Intel could toss more cores into it to catch up w/ competing silicon, and because Intel's processes too were so much ahead of everyone else, those Core CPUs were still cheaper than PA-RISCs, Alpha's

          • Intel really goes all in on R&D. It isn't just that they make a lot of money, their percentage of R&D investment is very high, and they don't cut it in down times to try and squeeze out numbers. That snowballs in the long term and is why they had a 14nm process running when others were still struggling to finalize their 20nm.

            Also a lot of the architectural arguments regarding Motorola v Intel were bogus. It was programmers arguing on academic issues that maybe theoretically should matter, but didn't

          • Re: (Score:2)

            by simm_s ( 11519 )

            Yes, what people do not understand is that Intel is primarily manufacturing company. Intel builds the fabs and R&D/demand fills them.

      • Re:I got an idea (Score:5, Informative)

        by unixisc ( 2429386 ) on Tuesday January 06, 2015 @05:49PM (#48750301)

        Maybe yes, maybe no.

        Fabs are incredibly capital intensive. If you build one you sort of need to run it at full hilt to make money off of them. And then you are on the tread mill of always having to build another one to have the latest and greatest.

        I am not saying they should not build it, just that it is a completely different beast then design and marketing of their chips. You can't do it half way. FYI, almost everybody agrees that the reason why the x386 dominates today is not because Intel had the best or most efficient designs for microchips but that it had the best run and most efficient fabs. Intel has publicly said it want to be the "McDonalds" on the CPU world – standardized and cheap.

        Are AMD and Navida ready to bit off such a big chuck? I doubt it.

        Fully agree w/ this - it costs a ton of cash to just build the fab, then quite some time before they are in full production, and only after they've run for a few years and been depreciated that their products are profitable. By then, it's usually time to shrink, but by now, at ~15nm, there ain't too much scope for that.

        But in addition to that, the costs of running a fab are recurring - even if there is no demand for a product, the fab has to be kept running, thereby piling on inventory. Which is a good reason why a lot of companies either started of as, or later became, fabless. The only one who is profitable is Intel, as a result of being a full 2 generations ahead of the competition. Reason? Because they make volumes of products that they can sell in the market profitably, thereby enabling themselves to put in a whole ton of cash into newer fabs, better equipment and everything else geared towards process improvements.

    • Re: (Score:3)

      by erice ( 13380 )

      Build your own fab

      While not a bad idea, it doesn't solve the problem. When you have your own fab, you are pretty much obligated to use it. Even when it is late, low on capacity, or a full node behind. You can reduce this risk by throwing a lot of money at R&D and spare capacity. However, this is more than a little bit expensive. That is why AMD doesn't have a captive fab anymore. They can't afford it.

      TSMC is in the business of making chips. They don't make money if they can't make chips. I haven't heard that App

    • Re:I got an idea (Score:5, Informative)

      by Anonymous Coward on Tuesday January 06, 2015 @03:57PM (#48749301)

      Build your own fab

      AMD did exactly that. Intel responded by raising their prices and offering "discounts" to vendors in exchange for refusing to stock any AMD products, ensuring that AMD's production capacity was wasted because customers found it incredibly difficult to find anyone willing sell them (despite having a large performance, price and temperature advantage over Intel at the time). AMD took Intel to court over this and won, but Intel managed to drag the case out so ridiculously long that AMD weren't able to keep their foundries. Their R&D took a big hit too.

  • Consumers tripped up on exclusive deals (Score:2)

    by Anonymous Coward

    Who cares about the designs when my AMD GPU doesn't run software properly because Nvidia has a special deal with that software company? And visa-versa? Stop worrying about design and fix the real problem...your current products are unreliable because of greed.

  • So, this is why we can't have nice things?

  • bean counters ruin another company (Score:5, Insightful)

    by banbeans ( 122547 ) on Tuesday January 06, 2015 @02:44PM (#48748379)

    Who didn't see not having their own fabs was going to bite them in the rear?
    Only a bunch of bean counters would not have seen this coming.

    • Re: (Score:3)

      by TWX ( 665546 )
      It's a lot simpler to sit down at a CAD workstation to do layout design than it is to manufacture that design.

      Trouble is, at the rate we're going, the manufacturers won't need us for very much longer, especially if economies in other countries that don't respect IP become strong enough that those manufacturers can sell unlicenced production there, even if they can't sell it here. I'm thinking of that happy little picture of the world with a circle over portions of South and East Asia and the large islan

      • Re: (Score:2)

        by bug1 ( 96678 )

        I'm thinking of that happy little picture of the world with a circle over portions of South and East Asia and the large islands in the Indian Ocean where half of the population of the planet lives.

        Madagascar ?

        I dont see the connection, maybe i need to research geographic sterotypes...

    • Re: (Score:2, Insightful)

      by gstoddart ( 321705 )

      Pretty much this.

      The people who own that fabrication capacity are selling it to the highest bidder. Or the one who invested the money to build it.

      If you don't control your own production, don't be surprised when you have to wait in line.

      But don't act all shocked that the 3rd party who actually builds your products can't guarantee your stuff gets built.

      This gets a big "duh".

      • If NVidia build a fab, it would be great for them if/when their fab had cutting-edge processes and NVidia chips were the most profitable thing to run. But... what happens when their fab is out of date? NVidia chip designers would likely be forced to design for the NVidia fab anyway, and their hardware would fall behind. Or... what happens when their fab is updated? If they are one of the few on a new process, assuming they aren't sued by Intel for patent violations, should the NVidia fab lose out on pot

      • Don't be silly. It's not as if GlobalFoundries is putting AMD at the back of the line.
        They just haven't been able to keep up with TSMC technologically (GF couldn't promise to deliver 20nm or 16nm when TSMC could), and they probably wouldn't have been able to if they were still a part of AMD.

        The only big difference is that it would have been a pretty hard sell to have your GPUs be produced by a third (technologically more advanced) party instead of in your own (struggling) fabs.

      • If you don't control your own production and aren't willing to pay more for it than anyone else, don't be surprised when you have to wait in line.

        FTFY. nVidia and AMD could easily jump ahead of Apple and Samsung in TSMC's queue - they just have to offer to pay more than Apple and Samsung are willing to pay. All that's happening is that they've made a conscious decision that it's more cost-effective to pay less and have their 16nm/20nm products come out later and deal with unhappy customers, than to pay m

    • Re:bean counters ruin another company (Score:4)

      by DarkOx ( 621550 ) on Tuesday January 06, 2015 @02:55PM (#48748479) Journal

      How are they bitten exactly. Neither one of them can get chips. There is therefore no real competitive disadvantage. They only way I see them getting bitten is if Intel decides they really want a slice of the high end GPU market; ups their game on the design side AND allocates their 14nm facilities to GPUs in favor of cranking out more CPUs.

      • Re:bean counters ruin another company (Score:5, Funny)

        by gstoddart ( 321705 ) on Tuesday January 06, 2015 @02:57PM (#48748511) Homepage

        How are they bitten exactly. Neither one of them can get chips.

        Well, then it's a good thing there's only two of them.

        A third company who owned some fabrication facilities would completely screw things up.

        • Re: (Score:3)

          by DarkOx ( 621550 )

          Well, then it's a good thing there's only two of them.

          I think the bean counters might deserve some credit though. Don't you think somebody asked "what happens to our business if things go wrong at TSMC?" and the answer was well our chief competition also uses them so there will probably be little impact to our market share, they won't be able to supply anyone with chips either.

          • Re: (Score:2, Flamebait)

            by gstoddart ( 321705 )

            You attribute far more long-term thinking to the idiots who run companies than I do.

            And are you really saying that if I really wanted to destroy large US companies all I'd have to do is to bomb factories in China? Or sow dissent? Or cut off their electrical supply?

            Because that sounds like something out of a Bond film.

            Hell, build your own fabrication plant and then bomb the ones who build for your competitor; instant monopoly. :-P

            • Re: (Score:2)

              by Fwipp ( 1473271 )

              You know that bombing factories is, like, illegal, right? And that most people have, you know, ethics?

          • Current videocards work great. You can play all kinds of games in plenty of beautiful detail. So to get someone to buy a new video card, well you have to offer them a reason. If your new card isn't enough faster, or more efficient, or enough new features, or whatever then people will say "Nah, I'm good," and stick with what they got.

            So they do have a need to move forward on tech, if they want to stick around.

      • Does anyone know how good, bad, or awful Intel's GPUs are these days? I know that the performance of the GPUs they bother to sell is poor, that much is obvious; but GPU components parallelize nicely: pretty much all market segmentation for AMD, NVIDIA, and Intel (for a given generation of design) is a matter of how many of that manufacturer's GPU-sub-element you get. The most expensive one gets as many as the largest die will hold, the next most expensive gets the same die but with 25% or whatever disabled,

        • Does intel even make a discreet GPU? I thought they were all baked into their newer CPUs.

          • They don't offer any GPUs not packaged with CPUs; but I'm working on the assumption that the same basic GPU block could be turned into a discrete GPU if Intel thought that it was worth doing.

        • Re: (Score:3)

          by Guspaz ( 556486 )

          Intel's integrated GPUs are pretty decent these days. At this point, they tend to outperform lower end discrete nVidia/AMD GPUs of a similar generation. Intel's Iris Pro 5200 GPU, for example, performs slightly faster than an nVidia GeForce 640. Each generation has seen Intel's chips creep slightly higher up the nVidia/AMD product line: it's not long since they creeped past the x30 tier of nVidia GPUs.

        • The Iris Pro is a competent gaming chip, but it only matches AMD's iGPU at best. And yes, DDR3 is a severe limitation. As are Intel GPU drivers. (garbage) But as silicon goes, they are actually good GPU tech and plow through OpenCL.

      • Re: (Score:2)

        by LWATCDR ( 28044 )

        Simple gamers do not buy new cards until you can get new cards.
        No one buys new cards until there are new cards to buy...

    • Re: (Score:2)

      by itzly ( 3699663 )
      It does you little good to keep a fab, if you're not able to compete with intel on the process shrink rate.

    • Re: (Score:2)

      by bws111 ( 1216812 )

      Yep. That is why no other companies use 'suppliers'. Oh, wait. EVERY company is dependant on its suppliers. Does Intel operate its own mines? Does it create all of the tools, etc in its fabs? No? The stupid bean counters at Intel!

    • Re: (Score:3)

      by tlhIngan ( 30335 )

      Who didn't see not having their own fabs was going to bite them in the rear?
      Only a bunch of bean counters would not have seen this coming.

      Fabs are expensive. As in, REALLY expensive. They cost $1B to build, and easily $10+B to equip (or more). So you're already in the hole many billions of dollars and you haven't produced anything yet.

      Oh, and there's a clock ticking away, because you're going to need to spend another $10+B to buy ALL NEW EQUIPMENT to handle the upcoming node shrink.

      The only way to do that i

    • Considering the astonishing rate that AMD was losing money on its fabs, and the fact that upgrading a single fab to a new process node would cost more than AMD's entire market capitalization, I'm going to have to side with the bean counters here.

      AMD coundn't even keep its production facilities running. How could they possibly have kept up with TSMC - the world's premier foundry operator?

    • Except that in this case they will be quickly out of business if they actually tried to build a fab, which is a 10 billion dollar investment which will go obsolete in 2~3 years.

      The fundamental problem is that there isn't enough demand for discrete GPUs, so they are getting a hard time hitting the economy of scale. What I see is that mobile chips are getting the latest and greatest process technologies these days - I believe the market size of mobile processors is order of magnitudes larger than discrete G

  • Actually...about Intel... (Score:3)

    by fuzzyfuzzyfungus ( 1223518 ) on Tuesday January 06, 2015 @02:56PM (#48748493) Journal
    I certainly wouldn't expect to see it happen(well, maybe with a very low probability); but it wouldn't surprise me if someone at Intel Legal has written up an "AMD/Foundry Contract Opinion.doc" and squirreled it away somewhere.

    Given that AMD isn't terribly threatening anymore, we aren't in the Netburst vs. A64 beatdown era now, Intel is probably saved a fair amount of unpleasant antitrust inquiry(US and abroad) by AMD at comparatively limited cost in product margins or lost design wins. If it came to it, selling them foundry services would probably be preferred to letting them die.

    • Re:Actually...about Intel... (Score:4, Insightful)

      by bill_mcgonigle ( 4333 ) * on Tuesday January 06, 2015 @03:26PM (#48748859) Homepage Journal

      it wouldn't surprise me if someone at Intel Legal has written up an "AMD/Foundry Contract Opinion.doc" and squirreled it away somewhere.

      Or they can take advantage of the situation to acquire nVidia's market position in GPU computing and patent arsenal (to annihilate the trolls - probably not gunning for AMD). Intel wants a viable AMD, to keep DoJ off its back. nVidia is just a market competitor with a manufacturing problem.

  • Goodbye AMD (Score:2, Insightful)

    by Anonymous Coward

    Your shortsighted business practice of selling off your foundry has just killed you. It's gonna suck having Intel as the monopoly for the X86 CPU market.

    • Your shortsighted business practices of getting fucked over by Intel's almost completely unpunished illegal monopoly abuse has just killed you.

      FTFY.

      • Re: (Score:2, Insightful)

        by arbiter1 ( 1204146 )
        o please, intel has superior cpu, straight out. AMD hasn't had a decent cpu since athlon xp/64. Blaming intel for AMD failure to innovate their cpu is straight up retarded. Intel has the market share they do cause they have made cpu's people want, not slow power hungry piles of silicon.

        • Re: (Score:2, Troll)

          by Carewolf ( 581105 )

          o please, intel has superior cpu, straight out. AMD hasn't had a decent cpu since athlon xp/64. Blaming intel for AMD failure to innovate their cpu is straight up retarded. Intel has the market share they do cause they have made cpu's people want, not slow power hungry piles of silicon.

          Go back to the time period we are talking about. The fabs were sold off in 2008 which was when AMD was still a top dog performance wise, but losing money due to Intel illegal deals with OEMs.

  • no way??? (Score:3)

    by sribe ( 304414 ) on Tuesday January 06, 2015 @03:21PM (#48748807)

    The only other manufacturer with 14nm capacity is Intel and there's no way Intel will sell them some capacity.

    Riiiight. Because Intel never ever sells fab capacity.

    Oh wait, they started doing that in 2010. Oops.

    • Re:no way??? (Score:5, Funny)

      by drinkypoo ( 153816 ) <drink@hyperlogos.org> on Tuesday January 06, 2015 @03:27PM (#48748873) Homepage Journal

      Riiiight. Because Intel never ever sells fab capacity.

      AMD going to Intel for fab capacity is like the Palestinians going to the Israelis for rocket technology.

    • Riiiight. Because Intel never ever sells fab capacity.

      The sentence doesn't say that. It says it wouldn't sell AMD or Nvidia, the companies referred to by the "them" in that sentence, some of their capacity.

    • They only have one 14nm fab right now, and it only makes Intel chips. Maybe they'd sell some 22nm, I dunno, and that might be of interest for a GPU, but they aren't selling 14nm at this time near as I know.

      • Re: (Score:2)

        by sribe ( 304414 )

        They only have one 14nm fab right now, and it only makes Intel chips. Maybe they'd sell some 22nm, I dunno, and that might be of interest for a GPU, but they aren't selling 14nm at this time near as I know.

        Their announced plans for 14nm fabs are vastly more than what they need to supply their own chips--in fact, more than all the world's other fab production combined. Doesn't sound to me like they're planning on hoarding it forever.

  • Just one. Nothing out of control. Intel has something like a dozen of them. Just build one. Then you have the ability to expand your own capacity using Intel's model of "EVERYTHING EXACTLY THE SAME" where in Intel copies their fabs exactly down to the furniture.

    • Fab facilities are tremendously expensive. This isn't something you can throw together in your basement. A fab is going to cost at least a billion dollars - and that's not even for state-of-the-art 14nm stuff. TMSC's fab cost ten billion dollars to build. That's just the construction cost - semiconductor tech is constantly changing, so if you want to make the latest goodies like high-performance GPUs there's also the need to constantly puchase new and better equipment.

      • Intel maintains many of them. AMD is perfectly capable of maintaining ONE.

        • Re: (Score:2)

          by brxndxn ( 461473 )
          No, it isn't. The entire market cap of AMD is less than the cost of building a new fab. Intel royally fucked AMD in 1999-2003 and AMD has been in a dismal financial position ever since. Intel does not play fair - and until Intel is forced to play fair, the best you can expect from AMD is for them to stay alive. Intel was forced to pay a fine that was about 5% of where AMD should have been financially.
    • AMD had fabs. They had several. The cost of operating them was a heavy burden on their bottom line; so, they spun them off in 2009 (that would be Global Foundries) and decided to go fabless. Now, they're experiencing the consequence of being dependent on 3rd parties to do their manufacturing for them. TSMC is building more fab capacity -- they just can't get it online fast enough (it takes several years to build and qualify one of these fabs)

    • Just get a 3D printer and make your own chips. Jeez, cheapskates.

    • Re:Build your own fab (Score:5, Informative)

      by hendrips ( 2722525 ) on Tuesday January 06, 2015 @04:09PM (#48749403)

      As of September 2014:

      -AMD's available cash: $950 million
      -AMD's market capitalization: $2.6 billion
      -AMD's credit rating: Absolute garbage
      -Cost of a new Intel/TSMC style fab: $7 billion - $10 billion

      It's a nice thought, but the reason that so many companies, including huge companies like Apple, IBM, and Qualcomm, have gone fabless is that fabs are astonishingly, mind-blowingly, expensive.

    • With what money do you expect them to build a fab?

  • Intel's ahead, but it kind of looks like they have a lot of difficulty too. There are so many different part #s, that it makes you wonder if they all tried to be the same thing, but different wafers passed different tests, and .. viola, diverse line of products!

  • Nokia the ultimate outsourcing warning (Score:5, Interesting)

    by jphamlore ( 1996436 ) <jphamlore@yahoo.com> on Tuesday January 06, 2015 @03:56PM (#48749287)
    The ultimate story about the dangers of outsourcing is how Nokia destroyed its mobile phone business. Once upon a time Nokia and Texas Instruments had a very close working arrangement with TI being Nokia's fab partner. The two together had a complete phone solution. So how does Nokia treat TI in the mid 2000s: They decided to diversify their wireless chipset providers away from working with TI. Only Nokia forgot one thing: TI don't play in markets where it cannot be overall #1. TI will as fast as possible get out of business segments where it cannot lead. And so TI said to Nokia, bye by 2012. In 2009. By then Nokia had decided it wanted to get its ARM SoCs and wireless modems from the same supplier, and there was one natural candidate, especially since they were, and still are, the leaders in LTE: Qualcomm. Only there was one big problem: Nokia had been caught in a patent war with Qualcomm for years trying to put Qualcomm out of business. It was Nokia that wound up having to settle for billions of US dollars, and suddenly it was at the mercy of the company to whom it had been an existential threat. Oops.

  • Former Semiconductor Furnace Engineer Here (Score:2, Insightful)

    by Anonymous Coward

    If you want the biggest bang for your buck, a single 12" Vertical Semiconductor furnace shipped and installed - $900K - $1m USD (low end guess). I've been to Taiwan into another fab, UMC, and they have hundreds of these furnaces. But with wafers that big, you need an automated transfer system (because you can't trust people to carry $10,000 worth of substrate) Wafer boats, typically holding 13 wafers, are $2000-$3000 each. the substrate itself is incredibly expensive. And not to mention the electrical bills

Slashdot Top Deals

To be is to program.

Close