Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
Apple Hardware

Apple To Buy TSMC's Entire Supply of 3nm Chips For 2023 (macrumors.com) 83

According to DigiTimes, Apple will receive all TSMC's first-generation 3-nanometer process chips this year for its upcoming devices. MacRumors reports: As early as May, Apple was known to have booked nearly 90% of the Taiwanese pure-play foundry for its upcoming next-gen devices. However, Apple is now projected to take 100% of TSMC's capacity in 2023, due to delays in Intel's wafer needs owing to later modifications to the company's CPU platform design plans. Intel's lack of orders means TSMC's sales of 3nm chips will be significantly lower this year. While TSMC is still expected to experience significant growth in the fourth quarter as it starts mass producing 3nm chips for Apple's needs, they too have been downgraded, according to DigiTimes' industry sources.

The report suggests TSMC's 3nm process output may be reduced to 50,000-60,000 wafers monthly in the fourth quarter, down from the 80,000-100,000 units previously anticipated, due to a cutback in Apple's orders. The current monthly output of TSMC's 3nm process is estimated at approximately 65,000 wafers, the outlet's sources said. Apple's upcoming iPhone 15 Pro models are expected to feature the A17 Bionic processor, Apple's first iPhone chip based on TSMC's first-generation 3nm process, also known as N3B. The E3nm technology is said to deliver a 35% power efficiency improvement and 15% faster performance compared to 4nm, which was used to make the A16 Bionic chip for the iPhone 14 Pro and Pro Max.

This discussion has been archived. No new comments can be posted.

Apple To Buy TSMC's Entire Supply of 3nm Chips For 2023

Comments Filter:
  • by Anonymous Coward

    Coincidence? I think not!

  • by Luckyo ( 1726890 ) on Tuesday August 29, 2023 @09:06PM (#63807490)

    Right now, if it's a process that you can make nvidia GPUs (well, AI accelerators, but it's a very similar chip), everything you can make is bought wholesale for a very long time in the future by nvidia.

    But TSMC 3nm isn't. Going to be interesting to see if/when nvidia will be interested in redesigning A100 or its successor for this process any time soon if any extra capacity becomes available. Since any extra capacity means money printing for nvidia right now.

    • > Going to be interesting to see if/when nvidia will be interested in redesigning A100 or its successor

      Nvidia's forthcoming 3nm GPUs, anticipated to be crafted via the TSMC 3nm chip procedure, will not be hitting the market until 2025, as reported by insider sources.

      https://www.guru3d.com/news-st... [guru3d.com]

      • But later they say 2024:

        Considering the RTX 50 series will employ the new process node (Via Hardware Times), you may expect some serious performance boost. From what we know, the upcoming RTX 50 series GPUs, codenamed Blackwell, are expected to release in 2024.

        https://www.gizmochina.com/202... [gizmochina.com]

      • by Luckyo ( 1726890 )

        I'm not talking about consumer GPUs, which is something nvidia is pivoting away from as fast as it reasonably can. I'm talking about data center GPUs. That's why I referenced A100.

        Nvidia is slowing down production of consumer GPUs where it can, and it's pretty likely that 5xxx series will suck almost as hard as 4xxx in terms of pricing. Because nvidia is trying to push same production processes to make A100s right now, as it gets far more profit out of them.

        • One thing I dont get is why they canâ(TM)t just take the existing print of the H100 or A100 and stick it in the smaller wafer maker? My basic understanding of photolithography is that the independent components are smaller. But the circuitry should still run, and give you the benefit of lower power requirements. It is an ecological gain if nothing else. Less power. Less power for heat. Looks like they need customers anyway.
          • by Luckyo ( 1726890 )

            Your understanding is very out of date. Old processes were mostly drag and drop kind.

            Modern production processes come with fairly extreme limitations on how chip itself must be designed, so chip designs target a specific process. Chip designed for one process cannot be transitioned to another directly, it must be redesigned for that specific process, then you must do test runs, fix the problems you find and keep trying until you get it right.

            This takes a very long time. Usually measured in years rather than

        • NVidia reuses many of the same does for their consumer GPUs. The 4090 is just the same die that goes in to their L40 Accelerator cards, only is has fewer CUDA cores and less VRAM. Essentially they're selling anything that has defective hardware or can't hit performance targets into the consumer market.

          That's part of the reason the pricing sucks. Normally companies would bin hardware to match market demand, but anything that can be sold to data center customers is and there's no excess consumer cards to f
          • by Luckyo ( 1726890 )

            This is very old and outdated information. They moved to different architectures for AI accelerators vs GPUs several years ago when it became clear that these will be the primary product in the future. They're still logically similar, but have significant enough differences to warrant a different development path and a different design team.

            The pricing sucks because there's insane demand, coupled with the fact that A100 has incredibly expensive tech on board like 80GB of HBM3. And that the old gen, new gen

      • > Going to be interesting to see if/when nvidia will be interested in redesigning A100 or its successor

        Nvidia's forthcoming 3nm GPUs, anticipated to be crafted via the TSMC 3nm chip procedure, will not be hitting the market until 2025, as reported by insider sources.

        https://www.guru3d.com/news-st... [guru3d.com]

        The A100 successor, i.e., the H100, is already out. The H100 uses the 4N node, which is sort of 4nm, but similar in ways to the 5nm family.

        The problem with Nvidia data center GPU supply is not wafers but CoWoS packaging. This packaging is not something that Apple uses, and that's part of the problem, sort of. Past Nvidia demand for the data center packages with HBM has been muted. So, when the sudden huge uptick in demand appeared, TSMC was not prepared to handle the increased need for packaging. That'

    • What're you going on about? TSMC and NV don't often get along, especially when NV flirts with using Samsung nodes to try to save money.

      TSMC prioritizes Apple first and AMD second. NV is sort of in the same doghouse as Intel in that if they want wafers on a new node, they'd better pay for the capacity up front.

      • by Luckyo ( 1726890 )

        Considering that nvidia is printing money right now limited only by production capacity, it could probably buy production wholesale if TSMC let them. With cash, not credit lines.

        The amount of money being thrown at nvidia right now due to ML AI boom is difficult to fathom. They don't just sell everything they produce. They have sold their production out for a very long time and there are people waving money at the back of the queue begging to get in front.

        If they could make a "A100 new production line versio

    • The latest manufacturing node always generates a pile of faults. This does not last forever, it just takes some time for the errors to be discovered and the manufacturing procedures corrected.

      This new 3nm process is best used to manufacturer ICs with small dies. This is because yields increase inversely to the die size. So using the 3nm process to make cell phone chips is a good idea. Using this process to create monster NVIDIA dies would be a waste. In a year or two things will be different.

      • by Luckyo ( 1726890 )

        It's not just that. If this is a mobile phone process, it may be a low power process fundamentally unsuitable for high power circuitry like A100.

    • Apple gets 1st dibs on any new TSMC node, as Apple funds TSMC R&D directly. Nvidia gets no priority ordering, as they've been antagonistic to TSMC recently.
      • by Luckyo ( 1726890 )

        True. But as we see here, apple is cutting back, while nvidia is desperate to secure more production.

  • Selfish (Score:1, Troll)

    by Gabest ( 852807 )

    It is unfortunate, but China must liberate the fabs for the rest of us, android users.

    • lol if you think that if China actually did invade taiwan that *anything* from either country will be available for you to buy for a very long time.

      But I dont think China will. Theres too much money in Taiwans role as essentially the way for americans/europeans to trade with china without political discomfort. Nobody wants to fuck with that cash cow, assuming Xi doesn't go full death or glory on it.

      • Re: (Score:1, Interesting)

        by Anonymous Coward

        Taiwan is a golden goose, and even Xi (hopefully) knows the fable of the goose that laid the golden eggs. There is a lot of tension, but ironically, it seems to be backfiring on China. If China laid off the parades and July 4 celebrations off Taiwan's coast, there wouldn't be a CHIPS act, and there wouldn't be a global push to have fabs distributed around the world.

        Hopefully China can come back to their senses. They just need to wait until January 20, 2025, when Trump takes office (hopefully not, but rig

        • Re: (Score:2, Insightful)

          by phantomfive ( 622387 )
          Trump and Biden agree on Taiwan/China [csis.org]. Trump increased sanctions on China, and so did Biden. Actually Biden and Trump are similar in a lot of ways, Trump just throws tantrums more publicly.
          • Trump created a diplomatic problem with China where there was none previously. Before, we might have simply asked China not to invade. In fact, that could well be the reason why China has not done so for decades now.

            After Trump, we're left in an antagonistic position and our future governments can't just go back to being friendly. We're trying, slowly, that was the point of Blinken's visit in June, but those are baby steps. It's going to take a long time to repair that relationship.
            • Re: Selfish (Score:5, Insightful)

              by Junta ( 36770 ) on Wednesday August 30, 2023 @06:36AM (#63808034)

              As much as I dislike Trump and he did create diplomatic problems, China was becoming a problematic relationship before he got going. He might have accelerated things, but there was already tension.

              Basically as they grew from providing cheap labor to American companies to making their own Chinese companies, we started caring about their governance, human rights problems, and economic manipulations. We were cool so long as those served American business leaders' purposes.

              • I mean... they're also commies. At least in name. And there was a trade deficit, and they were making some moves in the South China Sea. And yet we were still on pretty good terms until Trump's pointless tariffs and trade war.

                I'm not claiming that everything would have been marvelous, but there's a lot that you can do with diplomacy. And we just threw that away for no particularly good reason.

                That trade war seems to have generated a lot of ill will among the general populace in China, not just the pol
                • by Junta ( 36770 )

                  I suppose the biggest sign is that when Biden took office, there was a conscious effort to rewind a lot of the diplomatic disasters that Trump created.

                  However, conspicuously, there didn't even seem to be an attempt to rewind the China moves. So I suspect that *perhaps* a democratic administration might have broadly disagreed on the timing and aggressiveness, they do not seem to be in disagreement on the general sense of things. If they fundamentally disagreed, I think we would have seen a more conscious e

                • I do kind of wonder how much of the tariffs and useless trade war that accomplished absolutely nothing contributed to the inflation we saw over the last two years. It's not like Chinese companies are going to say "oh well snap, they passed a bunch of import tariffs. Guess we'll just leave our prices exactly the same and eat the tariffs ourselves instead of passing it on to the buyer!"

                  So basically it's the old GOP dirty-pool playbook once again: create the shitty economic situation with bad policy, get pun

                  • Tariffs are taxes on the people who are doing the importing (i.e.: American resellers), they're always passed on to the buyer. So yes, they contribute to inflation in a direct way.

                    Chinese companies do lose out from this in the form of decreased business, since the tariffs effectively make their products more expensive. But they operate under thin margins anyway, it seems very unlikely that they would reduce their prices to accommodate American tariffs. However, I wouldn't attribute this to Machiavellian
                  • Most of the inflation can be readily explained by the increase in money supply. As another way of looking at it (because it's good to try to look at things from multiple angles), if the inflation was mainly from Chinese tariffs, I wouldn't expect to see the food inflation that we've seen.

                    I'm not saying there's no inflation from the tariffs, just that it seems to be primarily from other causes.
        • I'm not receiving the NPC patches, making it difficult for me to keep up with the current day positions.

          Trump was racist against China when he wanted to close the borders, similarly when he called for moving away from doing business with them. After he left office those same positions became acceptable when the other guy came in, and now Trump is pro-China?

        • (hopefully not, but right now, it is likely)

          Assumes... something... that doesn't intersect with easily observable reality.

          Do not mistake nationwide popular polls more than a full year before the election with anything that resembles accuracy. Not a single primary vote has been cast yet, and the guy you think is likely to win has a trial that could lock him up for the rest of his life starting the day before Super Tuesday.

          If you think that the independents and moderates that rejected his bullshit 2.5 years ago are going to buy into it now that it's e

      • lol if you think that if China actually did invade taiwan that *anything* from either country will be available for you to buy for a very long time.

        But I dont think China will. Theres too much money in Taiwans role as essentially the way for americans/europeans to trade with china without political discomfort. Nobody wants to fuck with that cash cow, assuming Xi doesn't go full death or glory on it.

        Analytics said the same about Putin invading Ukraine without having the resources to do so successfully. Turns out that paranoid dictators do not always take the rational approach.

        • by gtall ( 79522 )

          This. And especially so if China's economy really goes south and the people start wondering if they wouldn't be better off without the CCP parasite sucking the life out of them. Xi is certainly stupid enough to invade Taiwan regardless of the consequences because he won't see the consequences beforehand. Even if told them, he'll find a way to ignore them to himself because he's learned the fine art of lying works in a "world" leader, especially lying to oneself. Of course, he was trained by the CCP so it is

          • by jeremyp ( 130771 )

            Xi is certainly stupid enough to invade Taiwan regardless of the consequences

            Citation needed. I don't think he is.

          • I don't think he's "stupid enough" but rather I could very much see him getting backed into a corner where he'll either have to pull that trigger, or face severe personal consequences with the home crowd. In which case, the consequences become roughly the same for the individual that gives no fucks about anything or anyone else - a thin chance of success in which case he's going to look like the leader of the next great Chinese dynasty, or he'll be dead / jailed.

            When the only other option is a casket or ja

  • Not surprising (Score:4, Informative)

    by Zuriel ( 1760072 ) on Tuesday August 29, 2023 @10:33PM (#63807624)
    Defect rates mean that new manufacturing processes tend to struggle with big chips like desktop CPUs or GPUs. Smaller, high margin chips that are extremely power-focused like phone chips tend to be the first customers of a new node.
    • Intel paid good money up front for TSMC to build out extra capacity. That TSMC is selling that capacity to Apple first is extraordinary in many respects.

      • Depends on what you consider extraordinary. Considering that Intel delayed their 3nm orders from TSMC [eenewseurope.com] until 4Q2024 after Intel already delayed the orders from 2022 to 4Q2023, TSMC is a manufacturer trying to keep their factories running after a customer's orders fall through. You could say that Intel's delays are extraordinary except for the fact they have missed shipments of their own chips for many years.
        • The delays from Intel, and the reasoning behind those delays, is what I consider extraordinary

          • It would appear from the outside that there is something very wrong at Intel. It was one thing when they had trouble making 10nm chips in sufficient yields for years. They were after all at the leading edge so delays were not unexpected. However not being ready with designs for another fab to make the chips is troubling.
            • They had initially intended for all of their compute tiles to come from internal nodes, which is not working out as well as initially anticipated. For example they have killed every product on their upcoming Intel 4 node except Meteor Lake compute tiles, and even that is heavily-constrained since Meteor Lake will only ship in 6P+8E configs and below (basically, mobile configurations) as opposed to 8P+16E configs for desktop.

              Or in other words, Intel cancelled Meteor Lake desktop altogether.

              They also killed

      • Intel may have paid for some portion of the expansion of capacity, but if they don't have any designs to build with it, would they expect TSMC to just idle the place and furlough everyone that works there until Intel is ready?

        Sorry, that's not how it works. TSMC will build someone else's shit until Intel is ready, and if they are in mid-production-run, Intel can wait because they're the ones that didn't make the date. I don't know for certain, but I'd bet good money that TSMC would have had a deadline dat

    • Defect rates mean that new manufacturing processes tend to struggle with big chips like desktop CPUs or GPUs.

      Yeah, there won't be an M3 Max for years to come. \s. But the real reason why AMD will be served later: Apple has design goals, so does AMD, and their goals are not quite the same. Apple prioritises low power before clock speed, AMD does the opposite. So the technology that is used for Apple wouldn't be unusable for AMD, but it wouldn't be what they really want.

  • by erice ( 13380 ) on Tuesday August 29, 2023 @10:34PM (#63807626) Homepage

    The grammar in both TFS and TFA is confusing but that is really what is going on. Only two companies had orders for 3nm from TSMC: Apple and Intel. Intel postponed their order. That meant available capacity for other customers in queue but there weren't any so TSMC will just reduce production.

    More generously, perhaps there are other customers who like to 3nm chips made but the Intel delay isn't long enough for them for another customer to slide in.

    • by Anonymous Coward

      They are in the chip making business, why can't they make 'em themselves? If you have to go to a smaller competitor for product, you might as well close shop and pack it in. You are done for.

      • Their internal nodes suck.

      • Because Intel's own R&D on chip fabrication processes have fallen way behind to the point that in order to stay competitive, they have to contract out the fabrication.

        I'm sure they are working on it, but they dug a hole that is going to take them years to get out of, and billions of dollars of research and development.

    • Kinda wish people would stop calling N3 "3nm" when TSMC dropped the nm for a very good reason. The name is just marketing at this point.

      That aside, N3B kind of sucks and also Intel is behind schedule on a lot of their designs. The only things they have coming out in 2023 are Raptor Lake Refresh and Emerald Lake on Intel 7. Well there's also gonna be a token launch of Meteor Lake but Intel never publicly stated they would use N3 for any of the tiles on that (GPU tile is on an older TSMC process).

      Intel wil

    • Supposedly Intel has delayed their orders until 4Q2024. Any company that wants to jump ahead of Intel and behind Apple will have to be ready very quickly with their designs.
      • Yeah, outside of Apple no one else would be able to use the wafers unless they had designs finished ahead of schedule. Apple probably got a good deal on the additional capacity and they make so many different devices that the additional capacity isn't going to hurt them at all.
    • -1, uninformative
      "other customers in queue but there weren't any"
      You present no factual basis for the claim that no other customers than Apple and Intel want N3 fab capacity.

      "TSMC will just reduce production"
      You present no factual basis for the claim that TSMC is reducing capacity for lack of customers.

      What is actually consistent with this and all preceding information is that a) Apple has the financial ability to outbid almost everyone, and b) TSMC doesn't have sufficient yield to offer more capacity.

      • More than that, they don't need to actually reduce capacity at all - they have so many chips ordered from Apple that they can just work on filling the order faster, freeing up capacity later in the year after they've filled Apple's orders.

        It's not like Apple has declared a maximum quantity of processors that they'll take delivery on in any given period of time. If they get them faster, they can just store them on-site where the phones are being manufactured and smile knowing that one less logistical headac

        • "they have so many chips ordered from Apple that they can just work on filling the order faster, freeing up capacity later in the year after they've filled Apple's orders."

          This is an excellent point I hadn't even considered.

  • they sell wafers.

    Angry old chip designer

  • If you can hoard a year of the top and largest fab for yourself alone for a year, you have some kind of monopoly on a bigger market.

    • by Hadlock ( 143607 )

      Or manufacturing capacity dramatically is underbuilt compared to demand

    • by necro81 ( 917438 )

      If you can hoard a year of the top and largest fab for yourself alone for a year, you have some kind of monopoly on a bigger market.

      You ought to re-read the article then. Apple isn't monopolizing all of TMSC's capacity, they just happen to be the only customer for N3B at the moment. If they're the only one buying, then ipso facto they are going to receive 100% of production. Intel was the other major customer, but they have delayed their order due to design issues and worries about N3B yield.

      But if

Crazee Edeee, his prices are INSANE!!!

Working...