GlobalFoundries stops development of its 7-nm LP node

The frighteningly expensive race for the last shrinks of silicon processes has claimed another contestant. GlobalFoundries announced today that it is suspending development of its 7-nm LP process indefinitely in order to shift its resources to specialized, continued development of its 14-nm and 12-nm FinFET nodes. This abrupt strategy change comes just months after the company touted its confidence and the scale of its investments in leading-edge fabrication technology at its Fab 8 facility in Malta, New York.

The decision was driven by economics rather than any technical roadblocks with 7LP, said CTO Gary Patton in an interview with Anandtech. The company says its future lies in tailored process improvements to its 14-nm and 12-nm FinFET processes for radio-frequency, embedded memory, and low-power customers. GlobalFoundries will also focus development efforts on its 22FDX and 12FDX processes for use with RF, analog, or mixed-signal designs that need low power, relatively low cost, and high performance.

Anandtech also reports that future development of GlobalFoundries' 5-nm and 3-nm nodes has been terminated, and the company will cease cooperating with IBM's silicon research arm at the SUNY Polytechnic Institute in Albany, NY after the end of this year. GlobalFoundries will also lay off 5% of its staff, according to the report, and it will be forced to renegotiate its wafer supply agreements with IBM and AMD.

While this move is certainly heartbreaking for silicon geeks like yours truly, the hard truth of the matter appears to be that GlobalFoundries would have had difficulty attracting enough business in the long run to make investment in its 7-nm, 5-nm, and 3-nm processes worth the enormous sums of cash needed to realize them. By the sound of its press release, GlobalFoundries' non-AMD and non-IBM customers are more interested in continued long-term returns on the already-challenging transition to 14-nm FinFET than they are in weathering repeated and difficult transitions to future leading-edge nodes.

This move seems unlikely to harm AMD's hunger for next-generation silicon. While every Ryzen and Epyc CPU made today comes from Fab 8, the company's first 7-nm product—a Vega data-center GPUis being fabricated at pure-play competitor TSMC. The company's next-generation server processors, code-named Rome, are also set to be fabricated in Taiwan, not at GloFo's Fab 8.

In any case, TSMC, Samsung, and Intel remain the only leading-edge foundries pursuing the outer limits of silicon fabrication after GlobalFoundries' exit. Only time will tell whether the race claims another victim.

Comments closed
    • ronch
    • 1 year ago

    Should’ve asked this earlier: will AMD still release Zen 2 on 7nm next year?

      • Chrispy_
      • 1 year ago

      Yes, but there might be a delay to the end of next year because of the extra work to transition to TSMC. AMD made a press announcement about this a couple of days ago.

        • ronch
        • 1 year ago

        But even before this turn of events, wasn’t AMD originally planning to build ‘Rome’ EPYC dies at TSMC? It’s only the desktop Ryzen chips that are planned to be built by GloFo. But since I’d expect Epyc dies and Ryzen dies to be quite similar if AMD continues to build Epyc with the same Ryzen dies, this means they can simply take those TSMC Epyc dies and sell them off as Ryzen parts.

    • anotherengineer
    • 1 year ago

    We all know this is because of Trump’s Tariffs 😉

      • chuckula
      • 1 year ago

      If it hadn’t been for all those markups on imported sand they’d still be in business!

      • gc9
      • 1 year ago

      New tariffs tax processors (US) and microprocessors (China), apparently as of last week, so the timing certainly leans toward that conclusion.

      Conjecture: With tariffs it’s unprofitable to fabricate even CPU semiconductors in the US ?

      Because they get hit by tarriffs both ways, importing the silicon from the US into China (subject to Chinese tariffs) for manufacture into packages, then reimporting the packages or computer products assembled in China into the US (subject to US tariffs) for incorporation or sale ?

      (Or even if not yet tariffed, are threatened to get hit, causing uncertainty that risky billion dollar global businesses loathe.)

      Made the choice easy for AMD.

      No wonder Intel decided to delay new fabs, they might have to move to another country to remain viable.

      /s

      IAmNotATariffExpert, but

      US tariffs that went in effect in July include ([url=https://circuitsassembly.com/ca/editorial/menu-news/29579-pending-us-tariffs-to-effect-host-of-pcb-equipment-components.html<]1[/url<]) ([url=https://ustr.gov/sites/default/files/2018-13248.pdf#page=16<]2[/url<]): [list<] [*<]Machine tools operated by laser, of a kind used solely or principally for manufacture of printed circuits (8456.11.70) [/*<][*<]Machine tools operated by light or photon beam processes, of a kind used solely or principally for the manufacture of printed circuits (8456.12.70) [/*<][*<][b<]Parts & accessories[/b<] for machines of heading 8456 to 8461 used to make printed circuits or PCAs, [b<]parts[/b<] of heading 8517 [b<]or computers[/b<] (8466.93.96) [/*<] [/list<] (Discourage anyone from duplicating Chinese factory lines in the US.) US tariffs that went into effect last week include ([url=https://ustr.gov/sites/default/files/enforcement/301Investigations/Final%20Second%20Tranche.pdf#page=8<]3[/url<]): [list<] [*<]8542.31.00 Electronic integrated circuits: [b<]processors[/b<] and controllers [/*<][*<]8542.32.00 Electronic integrated circuits: [b<]memories[/b<] [/*<][*<]8542.33.00 Electronic integrated circuits: amplifiers [/*<][*<]8542.39.00 Electronic integrated circuits: other [/*<][*<]8542.90.00 Parts of electronic integrated circuits and microassemblies [/*<] [/list<] Chinese tariffs threatened as of Aug 3 (please link if you know a final Aug 24 list) include ([url=https://assets.documentcloud.org/documents/4782394/China-Tariff-List-Aug-3.pdf#page=74<]4[/url<]): [list<] [*<]413 84714120 Minicomputers [/*<][*<]414 84714140 [b<]microprocessings[/b<] [/*<][*<]415 84714910 Other mainframes presented in the form of systems [/*<] [/list<] ("microprocessings" appers to be a mistranslation for "microprocessors".)

        • Anonymous Coward
        • 1 year ago

        How much of the machinery in a fab is coming from China, or other countries which are now under tariff? Export back to China could be a concern, but then so is import into the US. Adding a few 10’s of percent on top of costs could end up being serious money.

        • derFunkenstein
        • 1 year ago

        So they’re not going to manufacture microprocessors at all anymore? That’s a silly conclusion to draw.

    • benedict
    • 1 year ago

    AMD deciding to move all their production to TSMC meant GF lost their only 7nm customer. Makes sense to stop R&D on a product you have no buyers for.

      • tipoo
      • 1 year ago

      I dunno though, where does that place future fabs? Has anyone skipped a generation and successfully recovered? That’s what people are fearing.

      It’s like if Sukhoi cancels a questionably fifth gen fighter and says they’re moving straight to a sixth gen (except in this case, GloFo hasn’t even provided that guidance, just saying they’re focusing on last gen)

        • chuckula
        • 1 year ago

        [quote<]Has anyone skipped a generation and successfully recovered? [/quote<] TSMC and GloFo both effectively skipped "22nm" or the equivalent thereof and went from 28nm --> 16/14nm. But it wasn't without pain.

          • NTMBK
          • 1 year ago

          TSMC shipped millions of chips on 20nm, they definitely didn’t skip it.

            • blastdoor
            • 1 year ago

            TSMC probably would have skipped 20nm if Apple hadn’t wanted it — nobody else seemed to want it.

            But anyway… it sure sounds to me like GloFo has no plans at all to move forward with shrinks for the foreseeable future. It sounds like they’ve decided to innovate along other dimensions. Other firms have done that, too. There is a demand for microchips fabbed using older technology.

        • TheRazorsEdge
        • 1 year ago

        Skipping a generation has traditionally been the end of the line.

        Historically, you either die or make niche products cheaply without having to worry about R&D as much again.

        Look at UMC, who used to compete closely with TSMC. They just started 14nm last year, roughly two years behind Intel/TSMC/GoFo. They do not ship any high-performance designs that I know of.

        Economics will force everyone to move along with process tech, but we’re at the point where some industrial microcontrollers and most controllers for consumer electronics don’t need to get any better. The market is dividing up into performance and budget markets—and the budget markets don’t support much R&D.

      • ronch
      • 1 year ago

      But then AMD’s decision to move 7nm production to TSMC implies there’s something not quite right with GF’s 7nm.

    • Krogoth
    • 1 year ago

    Watch as 7nm parts from AMD become nearly unobtainium throughout 2019.

      • psuedonymous
      • 1 year ago

      7nm/10nm from ANYONE.

      GloFo (EUV) have dropped out completely, Intel are dealing with SAQP yields and shipping small volumes of small dies, Samsung dropped their 7LPE (SAQP) entirely and 7LPP (EUV) is due at some point in 2020, and TSMC 7nm is coming-totally-soon-honest-for-realsies-pinky-swear (and when it does, all output is likely to be eaten by the lumbering fruit corporation) and they are also wrangling with SAQP due to EUV continuing to be deferred due to the lack of pellicles.

      On top of that, gate thickness hit the 1nm limit a decade ago, and price/transistor has been going up since 28nm. 7nm/10nm is good news for low power chips for portable devices, but there’s no magical performance boost to be conferred, just as with the last few process shrinks.

    • cygnus1
    • 1 year ago

    [quote<]While this move is certainly heartbreaking for silicon geeks like yours truly, the hard truth of the matter appears to be that GlobalFoundries would have had difficulty attracting enough business in the long run to make investment in its 7-nm, 5-nm, and 3-nm processes worth the enormous sums of cash needed to realize them. By the sound of its press release, GlobalFoundries' non-AMD and non-IBM customers are more interested in continued long-term returns on the already-challenging transition to 14-nm FinFET than they are in weathering repeated and difficult transitions to future leading-edge nodes.[/quote<] I mean, everything boils down to needing a certain amount of revenue to support the project, but it's a really simple problem they have. They basically can't afford to physically displace any of their current manufacturing to give space to 7nm production. And they appear to have run out of new Oil money to build another fab or expand Fab 8. They just literally have nowhere to build the 7nm process without compromising existing, very much needed, revenue.

    • Kretschmer
    • 1 year ago

    It’s interesting how Intel’s struggles with 10nm was their death knell in the comments section last week, but GF’s (and TSMC’s eventual) struggles with new nodes are fine for everyone else.

      • derFunkenstein
      • 1 year ago

      well, this is as close to a death knell as it gets.

      • chuckula
      • 1 year ago

      From some of the posts around here, you’d think that GloFo abandoning 7nm means that AMD is guaranteed to launch Zen 2 ahead of schedule.

      Intel needs this type of hype! They need to officially cancel 10nm, which apparently magically means it will launch early.

      • blastdoor
      • 1 year ago

      Intel fabs only for Intel, and Intel’s only fab is Intel. So it’s a different dynamic.

      I’d say… GF giving up on 7nm is:
      1. clearly good for TSMC.
      2. a little bad for AMD
      3. a little good for Intel
      4. a good long term omen for trillion dollar mega-corps
      5. a bad long term omen for companies that are not trillion dollar mega-corps
      6. bad for consumers

    • kuttan
    • 1 year ago

    GF uses Samsung licensed processes right ? If that is the case AMD can depend on Samsung as a backup to TSMC when demand is too high ?

      • chuckula
      • 1 year ago

      That’s not accurate whatsoever.

      GloFo and Samsung have similar but not identical 14nm nodes, but AMD has never had a problem getting 14nm out of GloFo and has even paid GloFo hundreds of millions of dollars to [b<]not[/b<] run wafers in 14nm. There's practically zero chance that AMD is going to run to Samsung for 14nm parts now. GloFo's 7nm node was unrelated to Samsung's and was allegedly tuned to be similar to TSMC to accommodate AMD, but it wasn't identical to TSMC's process and it doesn't exist now, so that's moot. TSMC and Samsung do not share processes at all.

    • slushpuppy007
    • 1 year ago

    Can we expect more competitive clock speeds from AMD vs Intel with this transition to TSMC?

      • chuckula
      • 1 year ago

      After Intel abandons 10nm and goes to TSMC… sure!

      • Anonymous Coward
      • 1 year ago

      AMD must surely gain some clocks… but they have probably chosen a different clock target than Intel. I think choosing a speed and featureset that puts their work-per-watt out of reach of Intel (in for example servers) would be wise indeed.

    • Zizy
    • 1 year ago

    So, essentially only the TSMC has a leading edge HVM process right now? What a change a few years makes. I wouldn’t believe it if someone told me that.

    We might even see TSMC jumping to 5nm before either competitor reaches 10/7.

    • jensend
    • 1 year ago

    Wow. That’s a mind-boggling amount of R&D to just simply write off as sunk cost. And leaving their research partners and customers hanging out to dry, and basically making it clear that they won’t ever have a more advanced node unless it’s just licensed from their competitors.

    Given how far along they made it sound like they were when they gave the tours, canning it is really a shock. Delaying it quite a bit – saying ‘hey we’ve decided this won’t be economical for a couple years so we’re going to keep the lights on in 7nm R&D but not spend what it would take to put it into production next year’ wouldn’t have seemed so surprising; nor would canning the 5 and 3nm nodes.

    I wonder to what extent the new competitive landscape may lead TSMC and Samsung to ease off the throttle on R&D for post-7nm processes.

      • Pwnstar
      • 1 year ago

      If they ease off, they let Intel catch up.

        • Klimax
        • 1 year ago

        Interesting typo for “get lead again”…

        • jihadjoe
        • 1 year ago

        If Glofo doesn’t launch 7nm they never took the lead from Intel in the first place.

        Foundry 12nm < 14nm++ (+++?)

          • BurntMyBacon
          • 1 year ago

          I think Pwnstar was responding to the idea that TSMC and Samsung may ease up on post-7nm development. (Last sentence of OP)

      • Srsly_Bro
      • 1 year ago

      You should check out ASC 730 to learn more about R&D handling. It’s an interesting section of you’re into that kind of thing. I’m an accountant so I find it intriguing.

      • Ryhadar
      • 1 year ago

      Agreed. It seems amazingly shortsighted to just up an throw away all of the work and partnerships that have lead to 7nm fabrication.

      That being said, good for GloFo for being able to come up with revenue that isn’t all from AMD anymore. I’ve worked at a couple of places now where entire departments are propped up by one or two customers (and as a result, business decisions are made with only them in mind). It’s not a healthy place for the company to be in.

        • K-L-Waster
        • 1 year ago

        From a technology perspective it may look short sighted, but they have to make sure that they’re going to be profitable. If they analyzed the numbers and saw that the cost of completing the development of the process was going to end up costing more than they could ever receive in revenue from it, then any way you look at it pulling the plug is the only option.

        We may be approaching peak-semi-conductor — the point at which even if it is technically possible to go to a smaller node, doing so is prohibitively expensive.

          • Ryhadar
          • 1 year ago

          I agree to the profitability extent but they also said that would be cancelling their partnerships with IBM and indirectly with AMD (AMD wants 7nm that GloFo will no longer provide). It’s probably a calculated risk since there are so few high end fabs left, but I would imagine that IBM and AMD are not at all happy. After all, GloFo’s given two of their biggest partners a reason to not work with them in the future.

            • Anonymous Coward
            • 1 year ago

            Personal grudges are for losers, partnerships in business are about money, carefully written into contracts and enforced in court as needed.

      • blastdoor
      • 1 year ago

      If customers demand post-7nm processes, they will get it.

      But this is a good reminder/warning that the future belongs to companies with trillion dollar market caps that can make massive investments/payments.

      • cygnus1
      • 1 year ago

      It’s not all wasted. From what I understand, there’s a fair amount of knowledge gained from the 7nm R&D that will be applied to improve the other processes they’re sticking with.

        • blastdoor
        • 1 year ago

        Maybe they can sell what they’ve learned to Intel…

          • Freon
          • 1 year ago

          “Shit that doesn’t work:
          [1…n]

          • Pwnstar
          • 1 year ago

          LOL

      • ronch
      • 1 year ago

      Not to mention nobody would want to trust them as a partner in the future.

      • mikepers
      • 1 year ago

      Sunk-Cost Fallacy – reasoning that additional investment is warranted due to the fact that you’ve already spent X money, time, etc – even when it’s not the best thing to do…

        • Anonymous Coward
        • 1 year ago

        I think you’re too quick to whip out the Fallacy. A bit part of the surprise has to do with the [i<]progress[/i<] that they have presumably made, it is surprising that the [i<]remaining[/i<] costs have been judged so large as to make the whole thing a losing proposition.

          • K-L-Waster
          • 1 year ago

          Do we know they were making actual progress though?

          Perhaps they were on a path they were confident would be successful but that failed in the end, resulting in them being much further from a workable solution than originally hoped.

    • Pancake
    • 1 year ago

    What happens at 5nm then 3nm? I think unless Intel turns into an open fab they won’t have the revenue and volume to afford increasingly expensive nodes. Together with increasingly 2nd tier clients like Qualcomm and AMD they’ll have to use the services of TSMC and/or Samsung.

    So, who has the volume and money to invest in the smaller nodes and build products no one can compete with? Who will be the only 1st tier company? Apple.

      • Pwnstar
      • 1 year ago

      This article says we might get to 1nm transistors, with the caveat that it might not happen at all due to unforeseen problems.

      [url<]https://semiengineering.com/transistor-options-beyond-3nm/[/url<]

    • ronch
    • 1 year ago

    Would it be possible for AMD to have Samsung as a Fab partner? I reckon GF got some of its process tech know-how from Samsung. Guess pumping oil for decades makes businessmen not really want to go through all the trouble perfecting intricate processes. All they wanna do is pump, like pumping money out of the ground.

      • Tom Yum
      • 1 year ago

      AMD has said they have qualified designs on Samsung’s 14nm process (but haven’t said what design), so I’d say they can in the future also. Qualifying designs on different processses isn’t a zero-cost exercise however, so for a cash-strapped company like AMD, I doubt they would do it unless they were extremely concerned about TSMC having insufficient capacity to meet product demand.

        • BurntMyBacon
        • 1 year ago

        Samsung’s process tech is focused on smaller (phone/tablet sized) chips. This makes sense as that is what Samsung and most of their customers need. Unfortunately, and possibly as the result of this focus, they’ve had issues fabricating larger chips in the past. Midrange GPUs should be fine, but a GV100 sized chip is probably out of their purview. I’m sure they can work out the issues eventually (and possibly already), but I would want a backup plan for anything larger than Polaris.

        [quote=”Tom Yum”<] Qualifying designs on different processses isn't a zero-cost exercise however, so for a cash-strapped company like AMD, I doubt they would do it unless they were extremely concerned about TSMC having insufficient capacity to meet product demand.[/quote<] Unfortunately, with Apple already contracted for a large portion of TSMC's 7nm production capacity (sole source for A12), and around twenty other customers (including Qualcomm, nVidia, and even AMD's other side of the house) vying for 7nm capacity, AMD may in fact find themselves in exactly the position you describe.

          • Anonymous Coward
          • 1 year ago

          I wonder if AMD could park some products on 12nm for a couple years and refine them, or if they don’t have the resources for two process techs.

          • blastdoor
          • 1 year ago

          If they think it would be profitable to make larger chips then I imagine they can do that.

          If Samsung is going to stay in the race they need all the customers they can get. They ought to be looking at this as a nice opportunity to pick up a customer. They are clearly locked out of Apple’s products for the next few years.

          Perhaps Samsung could make a deal with AMD to move their consumer 7nm products over to Samsung ‘for free’ (that is, Samsung pays the fixed cost of the switch) in exchange for a small equity stake in AMD.

      • fyo
      • 1 year ago

      GloFo doesn’t resemble what AMD spun off way back in 2009. GloFo has gobbled up fabs like crazy since then and now has 10+ completely separate fabs. Some came with the acquisition of Chartered Semi, some with the acquisition of IBM Semiconductor, and some have been built since (or are being built).

      No two fabs appear to use the same process. The 14 nm process tech purchased from Samsung is currently only used in Fab 8 in New York. Other process technologies deployed by GlobalFoundries include 22FDX, which is based on a collaboration with STMicroelectronics. A new version of this on 12 nm called 12FDX is being implemented in GloFo’s Dresden fab (Fab 1).

      So this isn’t a question of not wanting to “go through all the trouble perfecting” processes (they are very much doing that), it’s more a question of realizing that spending > $10 billion on a move to 7 nm (which they claim was working great) wasn’t going to yield a good return, especially with AMD not being tied to them for CPU manufacturing anymore. Basically, there just aren’t that many clients interested in a leading-edge design.

      Look at what UMC has done. They bowed out of the leading-edge race after being tied with TSMC for years and years. They are still very much alive and kicking and introduced their 14 nm FinFET node about two years ago (so 2-3 years after Intel and right before TSMC’s 12 nm node).

        • Anonymous Coward
        • 1 year ago

        Nice summary of things. Its a bit surprising that there aren’t too many clients interested in being cutting edge, I guess that must be due to it also being expensive for chip designers to keep on top of it. (Or is the fabrication cost too high?)

        • BurntMyBacon
        • 1 year ago

        It was an odd situation though. Given how close they were to the targeted production date and taking them at their word that things were going great, I would expect that the majority of the research costs has already been incurred. They already have two ASML Twinscan NXE machines installed in Fab 8 that they may or may not be able to repurpose. Even if they decided to go no further into 7nm, I would have thought it would make sense to try to recoup some of the already sunk cost of getting their current process to the point it is at. AMD still had the WSA and they are pushing hard on 7nm, so there was some guaranteed demand right off the bat.

        I figure probably one of two things are going on here (though their are certainly other possibilities)
        1) The decision makers at GF must be convinced that their other offerings will be profitable enough to make up for losing their investment in 7nm.
        2) They intend to move on 7nm when the associated costs come down to some extent.

          • fyo
          • 1 year ago

          The problem is that getting a fab up and running 7nm apparently costs $10 billion. And any EUV equipment already acquired wouldn’t even be used initially, since the first iteration of GloFo’s 7nm process called for the same 193nm ArF immersion scanner used with their current process.

            • BurntMyBacon
            • 1 year ago

            I was just suggesting that it looks like they already had some space available and designated for 7nm fabrication. I can’t imagine they would dedicate space to EUV equipment that won’t be used until much later and not designate at least as much space for the DUV quad mask process that was in late stages of development and/or transitioning to production. I assume it would take an upgrade or new facilities to produce large quantities of 7nm chips. However, these two options are no where near equal in cost. It think I remember it being $10B for a complete EUV fab upgrade, $20B + significant time to build for a new EUV fab. Rather than a whole facility upgrade, upgrading a few existing lines for DUV Quad Mask should be cheaper (especially if the line was already DUV capable). Given how close they are to their targeted production date, I’m somewhat confident that their initial plan probably didn’t involve a new facility early on. Assuming I’m wrong on that, I am curious as to where the new facility was going to be built.

            Smaller quantities may even be doable in the space that was already allocated to 7nm in Fab 8 (purely speculation as I don’t know exactly how much area got allocated). Again, if we take their word that things were coming along well, then the majority of the cost is already sunk. Getting some returns is better than getting none.

            It easier to understand why they don’t want to continue on to EUV 7nm, though. Standing up an EUV line would incur significant additional research costs and new facilities costs. According to GF, it is unlikely that they would recoup that cost, so it is in their best interest to reject that path before those costs are incurred.

            [i<]Edit: Added example for clarification.[/i<]

            • fyo
            • 1 year ago

            I agree that it is hard to reconcile the complete cancellation with GloFo’s claim that there are no technical issues and that revenue shipments would have occurred this year. I would wager that while both of these claims might be (have been) technically correct, yields and throughput could have been so low that no money would be made anytime soon.

      • BurntMyBacon
      • 1 year ago

      GF did license its initial 14nm node from Samsung, but I believe the derivative 14nm processes were made in collaboration with IBM’s fabrication research team. IIRC they actually bought the IBM team (who’ve had influence on the new nodes going forward as well). There will be significant differences between their 7nm processes. The GF designs could be reworked for Samsung, but not necessarily any more easily than reworking the designs for TSMC .

        • fyo
        • 1 year ago

        GloFo have two completely different 14 nm processes. One, 14LPP, is the one they use for foundry customers (bulk FinFET). This process was acquired from Samsung. The other, 14HP, is a completely different beast and only used for IBM processors (huge 600mm2 + things running 5Ghz). It is BASED on the 14nm process that was spec’ed by IBM Microelectronics prior to its acquisition by GloFo and is a SOI FinFET process.

        However, 14HP as implemented by GloFo, was extensively modified. Basically, GloFo kept IBM’s front-end-of-line, but replaced the rest with something close to their 14LPP (Samsung) process. The 14HP has a massive 17 metal layers, compared to 13 for the 14LPP process.

        Then there is 12LP, yet another GloFo process. It’s basically a half-node with feature sizes shrunk around 10-15 percent. Notably, it also offers a new design library (7.5T), allowing more compact, lower power standard cells, but perhaps at the expense of some switching speed.

        The latest AMD processors are the only ones being produced on 12LP, AFAIK.

        There’s also a different 12 nm process called 12PDX, currently ramping at F ab 1 in Dresden. That’s a completely different beast and not targeted at large high-performance chips. It’s a planar (non-FinFET) fully depleted SOI process with a lot of promise (but not for processors or GPUs).

          • BurntMyBacon
          • 1 year ago

          Yes. The 14HP and 12LP are what I consider Samsung derivative processes since they started with Samsung’s licensed technology and made (largely IBM research driven changes). For instance, it was my understanding that to get to 14HP, they started with the Samsung process and replaced the front-end-of-line with IBM’s process and made some refinements through the rest of the process. Even if they started with IBM’s process and replaced everything but the front-end-of-line with their Samsung licensed technology, the end result is the same. Their initial 14nm offering (14LPP) was Samsung licensed and their derivative processes (14HP and 12LP) include both Samsung and IBM (and GF) technology. Though, I suppose in the second case it would more accurately be considered an IBM derivative process.

          I don’t include the 12PDX as a derivative as (like you said) it is non-FinFET and built on FD-SOI (not bulk) silicon, so I’m not sure how much (or little) Samsung licensed tech actually made its way into the process. My understanding is the SOI planar processes are developed separately from the FinFET processes at GF.

          In any case, reworking a design for Samsung’s process is possible, but not necessarily any easier than reworking it for TSMC’s process. I’m still curious as to what Samsung’s reticle limit is for their 7nm process and how close they can get to it with cost effective yields.

            • fyo
            • 1 year ago

            I don’t think it’s fair to consider the 14HP process a Samsung derivative. Yes, GloFo combined at least some of the BEOL (and MEOL) Samsung process with the original IBM spec process, but even then there are significant differences. For example, for the first three metal layers, the 14HP process uses double patterning like everyone else, but unlike everyone else (including Samsung) they use Litho-Etch-Litho-Etch and not Self-Aligned Double Patterning.

            And, as mentioned previously, the 14HP process has 17 metal layers, not 13 like the Samsung process. The top-most layers quite different than any other GloFo process, with a 2400nm pitch. This is because the process is made exclusively for IBM’s 700 mm2 die size Power chips. To see die sizes like that on 14nm, you have to look at Intel’s 28 core “Extreme Core Count” chips, but Intel mainly produces much smaller chips, so their process isn’t optimized for those rare chips.

            Even without those differences, considering that 14HP is the only SOI FinFET process out there, that alone makes it quite different from anything Samsung (or anyone else) has.

    • Wonders
    • 1 year ago

    EUV ETA: TBD

      • enixenigma
      • 1 year ago

      TBD?? More like DNF, amirite?

    • Star Brood
    • 1 year ago

    Maybe GF and Intel can team up in an ironic twist. Intel has money but needs a shrink, GF has the shrink but needs money.

    • tipoo
    • 1 year ago

    The drop in leading edge fabs is remarkable, and not a little disheartening.

    [url<]https://pbs.twimg.com/media/DlpQz6UXcAETwhZ.jpg:large[/url<]

      • Anonymous Coward
      • 1 year ago

      A remarkable trend to be sure. Capitalism and the free market have delivered the goods, and it seems are about to enter, shall we say, [i<]a new phase[/i<].

        • jihadjoe
        • 1 year ago

        I think that for products that are almost exactly the same allowing a completely free market does tend toward consolidation in the long run. It’s how we ended up with AT&T taking over telephones, and the handful of ISPs that basically own the internet.

        Maybe CPUs/GPUs are diverse enough for a consumer to prefer one over another (similar to liking Five Guys over McDonald’s), but this probably isn’t the case.

      • blastdoor
      • 1 year ago

      Nice graph!

      Fixed costs are rising with every node, requiring higher volume to justify the investment. Soon, maybe now, it will require higher prices, too. That is, new nodes might not reduce the cost per transistor, they might increase the cost per transistor. So the only people who move to new nodes are those that really, really value performance/watt.

      So, if you plan to sell 100 million AR glasses, you’ll pay for 5nm. But if you are buying servers to sit on a rack plugged into a wall, maybe 5nm isn’t worth it. That’s why I’m not sure Intel will be able to stay on the cutting edge.

        • Anonymous Coward
        • 1 year ago

        I sure hope that there is still room for other fabs making products a little behind the cutting edge, because its hard to see how 2 or 3 players in the whole world is going to be sufficient for competition. How far ahead will these new 5nm or 7nm processes be? Perhaps there will be real potential to choose slower fab processes and remain relevant in the market.

          • K-L-Waster
          • 1 year ago

          Inevitably you’ll get to the point where there won’t be competition anymore — it will literally be too expensive for all but the biggest companies to compete.

          Odds are that means Intel and Samsung, simply because they’re the only ones left that are big enough to keep writing checks that large.

            • blastdoor
            • 1 year ago

            I agree with the first paragraph, not the second.

            Unless we find out that the A12 in the next iPhone isn’t fabbed on 7nm after all, then TSMC is the only high volume manufacturer at the latest (7nm, or 10nm under Intel’s nomenclature) node. So long as TSMC has customers willing and able to pay for them to make the investments needed to hold that lead (and so long as they don’t screw up), then they’ll hold the lead. I think Apple is definitely willing and able to pay, and TSMC has done a good job at not-screwing-up for Apple.

            Samsung has deep pockets, but if they can’t win the work necessary to continue making the investments, they’ll stop. Same for Intel.

      • ronch
      • 1 year ago

      Only real men [s<]have fabs[/s<] can keep up with process technologies.

      • DeadOfKnight
      • 1 year ago

      It’s still early to say 3 of these companies will compete all the way to 2020.

    • ronch
    • 1 year ago

    GF is such an unreliable partner to AMD. Far from being the only time they screwed up AMD’s plans, plus the WSA that keep killing AMD. Scrap the WSA.

    • Pwnstar
    • 1 year ago

    Looks like AMD is going to TSMC. Only game in town.

      • Anonymous Coward
      • 1 year ago

      Seems like TSMC is going to have a little room to boost their margins in the next few years.

    • chuckula
    • 1 year ago

    AMD is of course saying that everything’s fine, but they probably didn’t have much lead time on this. From AMD’s [url=https://seekingalpha.com/article/4190245-advanced-micro-devices-amd-q2-2018-results-earnings-call-transcript?page=6<]Q2 conference call[/url<] on July 25: [quote<]Yeah, sure, John. So a couple of years ago, we did amend the wafer supply agreement. It was a very strategic agreement for us as we look over the long term. So at 7-nanometer, we are engaged with both TSMC and GLOBALFOUNDRIES. I would say that we do have, on a product-by-product basis, the choice between the foundries and we make those decisions on a product-by-product basis.[/quote<] So as of a month later AMD is [b<]not[/b<] engaged with GloFo on 7nm and they do [b<]not[/b<] have a choice between foundries at this point.... unless you think Su was lying in an official earnings call, which wouldn't be good for avoiding a lawsuit and potential SEC sanctions.

      • Pwnstar
      • 1 year ago

      Lisa isn’t stupid. It would have been very dumb to lie in a way that could be so easily proved. I think it makes more sense that AMD didn’t know.

        • Leader952
        • 1 year ago

        You should realize that AMD CEO’s have blatantly lied before.

        Remember these gems:

        $2 dollars is doable (yearly earnings) : this is in reference during an earnings conference call in the dot.com time. $2 was not even close and those who believed it lost a bundle when the stock crashed later that year

        We will never be more than 500 MHz behind Intel: this was stated during another earnings call and referencing intel’s 1.7 GHz P4 at that time when the athlon was only able to do 1.2 GHz. Needless to say intel shot all the way up to near 4 GHz and AMD couldn’t get higher than the low 2 GHz’s. AMD’s market share crashed because of this and the stock crashed again.

        These statements were from AMD CEO’s. Hector was the biggest liar-in-chief.

          • Pwnstar
          • 1 year ago

          I’m talking about Lisa. She’s very smart.

          • Shobai
          • 1 year ago

          [quote<]We will never be less than 500 MHz behind Intel... intel shot all the way up to near 4 GHz and AMD couldn't get higher than the low 2 GHz[/quote<] I'm... not sure I see a problem there.

            • Leader952
            • 1 year ago

            AMD did not keep the promise and fell way behind intel. Stock holders who believed the AMD CEO lost a lot of money.

            AMD CEO lied yet again on an earnings call and shareholders got reamed.

            • Shobai
            • 1 year ago

            It sounds like you meant to say “AMD said they’d never be [b<]more[/b<] than 500 MHz..." What you've written is logically equivalent to AMD saying [quote<]We will always be 500 MHz, or perhaps more, behind Intel[/quote<]

            • Leader952
            • 1 year ago

            You are right. Fixed now.

          • kuttan
          • 1 year ago

          And that low 2 GHz AMD CPU was able to beat Intel’s then 4Ghz CPU with ease. Check reviews 😉

            • chuckula
            • 1 year ago

            Really? Show me that review of Threadripper clocked down to 2GHz destroying Skylake X with ease.

            • kuttan
            • 1 year ago

            He was talking about P4s and then AMD CPU which is Athlon64. You get triggered wherever someone talk ugly facts of Intel 😀

            • chuckula
            • 1 year ago

            OK, so in ancient history AMD had an IPC advantage. Oh, and that was back when AMD had real fabs too. Funny how AMD’s magical stock price of $26 is still close to $20/share lower than it was back when it had fabs and even back when big-bad Evil Intel was being so “mean” by not letting AMD sell CPUs at cut-rate prices to Dell so AMD was “forced” to earn more money back then before it sold a single GPU and before RyZen.

            Since 15 years ago is fair game, how about less than 2 years ago: 4 core Intel parts destroy 8 core AMD parts!

            You are a well-known shill, but you could at least pretend to be competent.

            • kuttan
            • 1 year ago

            Look who is talking like a shill ? AMD share prices reached record high of $25!! All your hard efforts all these years gone down the drain. You’re required to improve your AMD hate troll tactics :DD

            • chuckula
            • 1 year ago

            [quote<]AMD share prices reached record high of $25!! [/quote<] Only in crackpot delusional fanworld is a price of $25 a "record" when the same shares without stock splits used to be worth over $44 over 18 years ago in 2000. Not to mention that AMD has basically never paid a dividend to its shareholders.

            • kuttan
            • 1 year ago

            Now you are on a full troll mode without the mask lol.

            • K-L-Waster
            • 1 year ago

            Not sure how you managed to get “record high” when they were at $44 in June 2000.

            Just ‘cus it’s ancient doesn’t mean it never happened….

            • NTMBK
            • 1 year ago

            Leader952 was complaining about one very specific period of history (Pentium 4 era), Kuttan continued to talk about the exact same period. What the hell are you complaining about?

            • kuttan
            • 1 year ago

            Chucku will simply get triggered when someone talks bad of Intel.

            • chuckula
            • 1 year ago

            If you had bothered to read the story that leads this article, AMD just lost the fab that was supposed to make your precious desktop Zen 2 and all the APUs.

            I’m sure that the drugs you take make you think that this is good news for AMD and that Intel is doomed.

            These arguments work great on some other planet. Your problem seems to be that this is earth.

            • kuttan
            • 1 year ago

            You didn’t received love from anybody that is why you were here 24×7 attacking anyone who expresses their opinion which is against your Intel. I feel sympathetic to you.

            • Leader952
            • 1 year ago

            [quote<]And that low 2 GHz AMD CPU was able to beat Intel's then 4Ghz CPU with ease[/quote<] I'm sorry but that was not the case. That is why AMD developed the XP rating trying to show that their 2.2 GHz was equivalent to an intel P4 3.2 GHz. The problem was that it really wasn't even close to the intel P4 3.0 GHz. Also intel got the P4 all the way up to 3.8 GHz. AMD Athlon XP - Xperience Performance Beyond the Gigahertz Rating [url<]http://www.informit.com/articles/article.aspx?p=339099[/url<] List of AMD Athlon XP microprocessors [url<]https://en.wikipedia.org/wiki/List_of_AMD_Athlon_XP_microprocessors[/url<] From Scott Wasson's review of the Athlon 3200+ right here on The Tech Report: [quote<]Next, I hate to say it, but AMD's credibility on this model-number rating system is showing some strain with the 3200+. Clearly, the Pentium 4 3.0GHz processor combined with the Intel 875P chipset is faster than the Athlon XP 3200+ in a slight majority of our tests, and sometimes by a notable margin. Yes, I know AMD's rating system is based not on a competing Intel processor but on the performance of a mythical Athlon "Thunderbird" processor running at the clock speed corresponding to the rating. But we all know how the ratings get used in practice, and in practice, the 3200+ model number is a bit of a reach. [url<]https://techreport.com/review/5126/amd-athlon-xp-3200-processor/12[/url<] [/quote<]

            • blastdoor
            • 1 year ago

            The Athlon 64 performed in line with its model number:

            [url<]https://techreport.com/review/5683/amd-athlon-64-processor/9[/url<] But it's still the case that a 2 GHz Athlon 64 did *not* beat a 4 GHz P4.

            • Leader952
            • 1 year ago

            Athlon XP is where the quote came from not the Athlon 64.

            • Tom Yum
            • 1 year ago

            By the time Intel got the P4 to 3.8Ghz (Nov 04), the Athlon 64 FX55 had been out for 6 months and was running at 2.6GHz, so you are stretching the truth a bit comparing the absolute tail end of Athlon XP with the future state of P4 (not to mention ‘faster in a slim majority of tests =/= not even close). Athlon 64 was released only 3 months after that review, so you’ve cherry picked a small window where the performance gap was in Intel’s favour. They remained in lockstep with each other until Conroe, and then Intel blew the barn doors open with Sandybridge.

        • freebird
        • 1 year ago

        I doubt she is “allow” to disclose such information until GF would announce the discontinuing of 7nm production. There are such things as confidentiality agreements and such. That is why people say we are “engaged” with partner X, it doesn’t commit them to anything. It really sounds like GF ran out of resources at FAB 8 (space/tools/wafer processing capacity) and would’ve needed to expand the plant, but deep pockets weren’t there to foot the bill…

        They probably can fill FAB with 14nm, 12nm & FD-SOI production to try an make some money and cut 7nm research for now. Maybe in 5 years after the big boys are on another node they can license it cheaper. In the end the owners of GF want to see a return on their money, not a black hole that keeps sucking in money.

        GF may try to be the “go to” company for lead FD-SOI, they seem have a lot more customer interest in that than 7nm.

        Kinda mirrors AMD 3-4 years ago… had to forgo leading edge CPUs (competing with Intel) and clean up the books/balance sheet and plan for a return.

      • NovusBogus
      • 1 year ago

      A month ago, she was probably aware that GloFo was fubared but not yet ready to throw in the towel. And, per her statement, their agreement gives them a non exclusive choice between the two, only one of them happens to not be interested in taking them up on it a month later.

      • jarder
      • 1 year ago

      From the same call, AMD said that the currently sampling 7nm EPYC chip is being fabbed by TSMC:
      [url<]https://seekingalpha.com/article/4190245-advanced-micro-devices-amd-q2-2018-results-earnings-call-transcript?part=single[/url<] [quote<]As for the 7-nanometer Rome that we're currently sampling, that's being manufactured at TSMC. [/quote<] So it looks like the choice of who was going to make the next generation of CPUs had already been finalized. Historically AMD had favored Glo-Flo for it's CPUs (unlike it's GPUs) so the writing had been on the wall for quite some time. Furthermore, Glo-Flo had already indicated that it was not going to try to be at the leading edge back in May where it signaled it was going to stop trying to be at the cutting edge and settle for being a "fast follower": [url<]https://www.eetindia.co.in/news/article/18051601-new-globalfoundries-ceo-outlines-targets[/url<] Also, it's not like Glo-Flo was misleading customers on their 7nm plans, they had apparently got as far as installing 7nm EUV machines in their fab: [url<]https://www.extremetech.com/computing/276169-amd-moves-all-7nm-cpu-gpu-production-to-tsmc[/url<]

      • K-L-Waster
      • 1 year ago

      Most likely, she was aware that GF was seriously considering it but the decision had not been finalized at that time.

      Assuming that’s true, the answer she gave would literally be the only way to handle it. She couldn’t announce that GF was looking at abandoning 7nm because announcing something like that about your partner *before they announced it themselves* would most certainly violate confidentiality agreements.

      (I’m not sure what the SEC requirements are about revealing information related to your partners rather than your own business, but I suspect that the AMD CEO announcing a major strategic direction change for GF before GF themselves announce it would not be wise from an SEC perspective…)

      • derFunkenstein
      • 1 year ago

      That quoted text is all so vague. I have a feeling that AMD was under obligation to say that. Even if they knew, could they really contradict GloFo’s public statements? What’s the likelihood that AMD had agreed to not share anything they might know about their foundry partners’ status? Pretty high, I say.

      • Sahrin
      • 1 year ago

      Every post you write is like a guidebook in “how to exhibit your lack of experience in corporate America.”

      You imagine business decisions are made with far more forethought and planning than they actually are.

    • Demetri
    • 1 year ago

    Would this allow AMD to back out from the WSA? That would really suck if they still had to pay GF for every wafer produced with TSMC (which by the end of next year may be everything).

      • chuckula
      • 1 year ago

      Assuming Saul wasn’t too hungover when negotiating the contract, presumably AMD got a deal where it doesn’t have to pay if GloFo’s entire 7nm process never materializes.

    • derFunkenstein
    • 1 year ago

    It seems like just yesterday Jeff was touring the joint. And now it’s dead.

      • chuckula
      • 1 year ago

      We here at Intel would like to not invite Kampman in to our fabs!

        • derFunkenstein
        • 1 year ago

        That dude is the kiss of death, obvi.

    • tsk
    • 1 year ago

    RIP GloFo, you won’t be missed.

      • Gadoran
      • 1 year ago

      Why RIP??.
      Semiconductor market is pretty large, an the bulk of customers not even think to 7nm.

      For example in Europe GF is a big supplier of SOI devices.

      • ronch
      • 1 year ago

      RIP GloFo. Thanks for all the one-time charges!

    • Chrispy_
    • 1 year ago

    GloFo seem like they were entirely supported by AMD’s [i<]many[/i<] various 'one-time' payments.

      • blastdoor
      • 1 year ago

      No, they are mostly supported by people trying to diversify out of petroleum

        • chuckula
        • 1 year ago

        Bad move. We all know that we passed peak silicon in the 70s.

    • blastdoor
    • 1 year ago

    Wow.

    So now we are down to TSMC, Intel, and Samsung as the only manufacturers attempting to stay on the cutting edge. Any bets on who will be the next to fall? I’d say “not TSMC”, but between Intel and Samsung it’s a tough call.

      • chuckula
      • 1 year ago

      Despite what we’ve heard about 7nm miracles and flowers, TSMC has been [b<]cutting[/b<] its revenue forecast while Intel has been raising its own forecasts... and that's after the 10nm delay: [url<]https://asia.nikkei.com/Asia300/TSMC-cuts-2018-sales-target-on-weak-iPhone-demand[/url<]

        • JustAnEngineer
        • 1 year ago

        Chuck, you should look at a comparison plot of INTC vs. TSM vs. AMD share prices.

          • chuckula
          • 1 year ago

          Let’s look at PE ratios.

          Intel is way lower than TSMC.

          And both of them are insignificant against AMD.

            • blastdoor
            • 1 year ago

            Let’s look at capital letters in their names. AMD has three times as many as Intel, TSMC has four times as many!

            • tipoo
            • 1 year ago

            Can I be that guy and just say none of the previous four comments necessarily show a strong correlation?

            A technically competent timely fab, share price, P/E, and capital letters all show about as much correlation as the lowest common denominator, yes the share price and hence P/E will take a hit on bad fab news, but positive revenue guidance can undo that regardless, and lastly the number of capital letters in VIA is WAY TOO DAMN HIGH.

            • blastdoor
            • 1 year ago

            [quote<]Can I be that guy and just say none of the previous four comments necessarily show a strong correlation?[/quote<] I think you might be on to something there...

          • Pwnstar
          • 1 year ago

          AMD is up 20% today.

        • blastdoor
        • 1 year ago

        I can post links too:

        [url<]https://www.digitimes.com/news/a20180828PD210.html[/url<] But neither really is relevant. These stories are talking about relatively minor +/- deviations from quarterly revenue expectations. What’s relevant to the discussion at hand is the massive amount of money being spent by Apple and invested by TSMC on next gen nodes. That is a reality that is not really affected by these little blips.

      • Klimax
      • 1 year ago

      Unless Intel leadership gets hit with big dumb, it won’t be Intel either.

        • blastdoor
        • 1 year ago

        I’m not so sure. Intel doesn’t really participate in the mobile market, which is where people might be willing to pay a higher price per-transistor in order to get better performance/watt. Intel is getting a larger share of its profits from computers sitting on racks plugged into a wall. Are Intel’s customers really going to be willing to pay a higher price per transistor? It might make more sense for them to just lower the clock speed a bit and throw in more cores.

    • LocalCitizen
    • 1 year ago

    if it weren’t for their extremely deep pockets, 10 nm could have killed intel.
    i hope this won’t be used as an excuse to raise chip prices.
    who am i kidding? of course this will lead to higher chip (high end cpu, gpu, etc) prices

      • chuckula
      • 1 year ago

      If you look at the graph in Anand’s article you’ll see something very interesting: As the 14nm process at GloFo ramped up in sales from 2016 –> 2017 with AMD GPUs and CPUs, GloFo recorded a large increase in revenue.

      They also recorded a large increase in… [b<]losses[/b<] while revenue was going up. Don't think for a second that TSMC has to play that game with AMD, who is nowhere near the #1 customer for 7nm silicon. Unlike with GloFo, AMD is going to have to pay top dollar for 7nm silicon, and don't expect fire-sale prices on future RyZen parts like we've seen in the last few months.

        • LocalCitizen
        • 1 year ago

        yup, 2016-2017 were/are great years for computing. going downhill from here 🙁

        • Srsly_Bro
        • 1 year ago

        Would AMD be better off sitting on inventory or selling for a price that would meet sales expectations in a given period?

          • Beahmont
          • 1 year ago

          AMD would be better off having a product that their fab partner didn’t have to sell to AMD at a loss for AMD to have any kind of profit margin.

            • freebird
            • 1 year ago

            Stupid as stupid does…

        • derFunkenstein
        • 1 year ago

        You get what you pay for, though. AMD might be paying more but they’ll get stuff that comes out on time.

          • Beahmont
          • 1 year ago

          Yeah, but how does the AMD value proposition work out if AMD has to charge $20-50 dollars more per chip just to maintain the anemic margins they have now, let alone actually have decent margins that will get their debt under control?

            • derFunkenstein
            • 1 year ago

            Enthusiasts are going to have to get over AMD taking one for the team and underselling their capabilities. ¯\_(ツ)_/¯

            • _ppi
            • 1 year ago

            Their CPU margins are not that bad, definitely not when compared to nVidia consumer products, so Ryzen will be probably OK as long as it performs vs. Intel.

            Compare e.g. ~200 mm2 Ryzen die and prices it sells for vs. 1080Ti (471 mm2) or 2080Ti (nearly 800 mm2) and their sales prices.

            Sure, Ryzens go as low as ~$150, but
            a) do not need a full board (incl. board vendor margin), expensive DRAM, as big cooler and higher cost of transport (box size).
            b) their top models sell well

            Note that they have one die for everything (except APUs), which drives cost down a lot.

            And on GPU side, they will be on equal footing with nVidia with regards to wafer cost. RTX is certainly intersting tech, but rumors are that Sony had some input on what Navi should have, so it might surprise us as well.

        • blastdoor
        • 1 year ago

        I’m not sure you can draw conclusions about what AMD will be paying TSMC from anything we learned today, except perhaps that GloFo can not compete with TSMC on price. Presumably AMD didn’t pick TSMC over GloFo because TSMC charges more.

        TSMC has a big advantage over GloFo — a customer willing to fund new nodes and commit to huge volume purchases.

        That’s also an advantage TSMC has over intel

          • chuckula
          • 1 year ago

          [quote<]Presumably AMD didn’t pick TSMC over GloFo because TSMC charges more.[/quote<] Evidence of that unsubstantiated conclusion? Considering that AMD intentionally chose TSMC for all of its high-end expensive stuff, every piece of evidence literally says the opposite of your conclusion: AMD was more than willing to throw money at a 7nm product with the hopes that TSMC would deliver on time. As for GloFo, every product that it was supposed to deliver on 7nm was supposed to be a cheaper consumer-grade product. You know, desktop RyZen 2 and those 2020 7nm APUs? Those were were supposed to be the cheap parts on GloFo. And now they aren't.

            • techguy
            • 1 year ago

            I think his point is that AMD went with TSMC because their processes are superior to GloFo. Since they’re arguably within striking distance of Intel’s parts in terms of per core performance, it might just be worth paying a bit more for a process that can buy say another 500MHz (random number).

            • blastdoor
            • 1 year ago

            That’s what I should have said 🙂

          • derFunkenstein
          • 1 year ago

          Neat guess but I think the only conclusion of any substance is that AMD picked TSMC because TSMC can make stuff that Global Foundries can’t. Doesn’t matter what something costs if it can’t be built.

        • kuttan
        • 1 year ago

        TSMC process likely costs more than GFs but considering the size of AMD it will still be able price considerably lower than what Intel can.

        • cynan
        • 1 year ago

        While I agree TSMC will have no incentives to cut AMD a special deal on 7nm. Some of the “fire sale” stock may also have been a result of the Wafer Supply Agreement stipulating penalties to be paid if failing to meet a minimum number of wafer orders. Further, according to the Anandtech article, AMD had to pay for wafers purchased at non GloFo foundries.

        So while TMSC might not be cutting them a price per wafer as competitive as GloFo might have for 7nm, this might be offset considerably if GloFo’s cancellation of 7nm nullifies some of the WSA penalties to GloFo if buying 7nm from TSMC.

      • PrincipalSkinner
      • 1 year ago

      No, GPU prices won’t increase because it will hurt the gamers…

        • K-L-Waster
        • 1 year ago

        I lol’d…

      • K-L-Waster
      • 1 year ago

      Why would it be an excuse? If component costs increase, so do final product costs.

      Or did you expect suppliers and vendors to deliberately lose money just so consumers would be happy?

        • chuckula
        • 1 year ago

        [quote<]Or did you expect suppliers and vendors to deliberately lose money just so consumers would be happy?[/quote<] Generally? No. But GloFo was an exception apparently. Emphasis on "was".

Pin It on Pinterest

Share This