15 ms·
Apple and Intel first to use TSMC 3nm
- 0xy 5y agoI wonder how large the king's ransom was that Intel threw at TSMC to make this happen. Seems like it was necessary given their faltering fab strategy, and faced with years of a limping 10nm and 14nm+++++++++ products.
- test_epsilon 5y agoAre you generally asking how much it costs to purchase early slots on leading edge nodes? I'm not sure if being Intel would have made TSMC jack up the price or anything. It is in their interest to get large profitable customers, and it is in their interest to take business away from Intel's fabs. TSMC might well have offered significant discounts or other special treatment to land Intel contracts.
- 0xy 5y agoTSMC are experiencing over 100% demand, they don't need to offer discounts. In fact, they stopped giving any discounts months ago even on older processes. Apple has deep pockets and spends heavily to ensure supply chain stability, but we haven't seen this behavior from Intel so I'm curious how big of a price they had to pay was.
- test_epsilon 5y ago> TSMC are experiencing over 100% demand, they don't need to offer discounts. In fact, they stopped giving any discounts months ago even on older processes. I stand by what I wrote. > Apple has deep pockets and spends heavily to ensure supply chain stability, but we haven't seen this behavior from Intel so I'm curious how big of a price they had to pay was. Would obviously highly depend on volumes, timelines they committed to, layers, materials, and specific devices and tweaks they are paying for. Could be less per transistor than Apple, I can't see hwy it would be vastly more (unless we're talking about extremely limited volums).
- smoldesu 5y ago> I wonder how large the king's ransom was that Intel threw at TSMC to make this happen. Probably similar to how much Apple paid to edge out the competition on 5nm.
- greenknight 5y agoWeren't Intel to use TSMC on 5nm in 2H 2021? -- https://www.extremetech.com/computing/319301-report-intel-will-outsource-core-i3-production-to-tsmcs-5nm-node https://www.extremetech.com/computing/319301-report-intel-wi... I doubt Intel would move any of their core designs across to TSMC. Only fringe products... Its not like their CPU designs port across natively and from my understanding, they would have to have an entirely different team that would work with TSMCs IP.
- xiphias2 5y agoYou can't have fringe products on 3nm, they have to be profitable ones. It seems like a risky, but exciting strategy. I prefer that over the old boring Intel.
- sharken 5y agoIt's a much better strategy for sure, as for AMD it seems that hard times are ahead if they can not move to 3nm after 5nm. It seems very likely that AMD has no other option than moving most of their products to Samsung 3nm. I hope that works out as the competition is needed. https://www.gizmochina.com/2021/02/02/amd-outsource-gpu-apu-production-to-samsung-semiconductor/ https://www.gizmochina.com/2021/02/02/amd-outsource-gpu-apu-...
- smoldesu 5y agoSamsung's fabs have been getting better and better over the years, we could see a surprise blowout on the 3nm node. Hell, just the other day Samsung was testing an RDNA2 iGPU that was faster than the GPU on the iPhone 12... on the 7nm node. I'm no expert here, but AMD might finally have the silicon it needs to compete with Nvidia in raw compute.
- atty 5y agoThat’s exciting news, but I can’t help but feel that at least in my field (machine learning), Nvidia is far more sticky than just their compute dominance. As long as CUDA is Nvidia proprietary, I just don’t think our team can afford to move. I’m seeing lots of very impressive movements from Tensorflow and Pytorch to support ROCm, but it’s just not at a level yet where it would make good business sense for us to switch, even if AMD GPUs were 50% faster than Nvidia. And it seems like Nvidia is improving and widening their compute stack faster than AMD is catching up.
- amelius 5y agoIs the US even aware at a political level that Taiwan is now ahead of the US and that US businesses are giving up?
- xiphias2 5y agoThey are giving incentives for TSMC to build factories in US for a good reason. Both China and US are aware how important it is.
- amelius 5y agoYeah but a Taiwanese factory on US soil still isn't US tech. Shouldn't the US invest more, while we can still catch up?
- xiphias2 5y agoTSMC is both a monopoly for the newest chips and a geopolitical risk in the case of China taking over TSMC in Taiwan. I think that if TMSC has a subsidary in US, it would be really hard for TSMC Taiwan to block TSMC US from licensing IP from it after the knowledge is transferred. Regarding the monopoly status: sure, healthy competition is good, but it's something competitors (Samsung, Intel) should tackle on the market.
- lotsofpulp 5y agoUS should be handing out green cards to TSMC workers and their families.
- smoldesu 5y agoIf your house was built using "US tech", you'd be living an Amish-adjacent lifestyle.
- ohazi 5y agoHalf the world won't even acknowledge that Taiwan is a country.
- 5y ago
- heisenbit 5y ago> “If Intel can get back to a two-year node cadence with 2x density improvements around mid-decade they can be roughly at density parity with TSMC,” says Jones A guarded statement indeed. It is hard to see how Intel can catch up on semiconductor technology. At this point of Moore‘s law it is about financial capabilities and while Intel‘s revenues may be still a little bigger than TSMC the semiconductor related part measured in capacity is a third and Intel is not in the top-5. Intel diverting money from own fabs in order to keep its options open and maintain a tech lead position is telling. Watch what they do not what they tell you.
- MangoCoffee 5y ago>while Intel‘s revenues may be still a little bigger than TSMC the semiconductor TSMC's Q1 2021 revenues is 362.41B Intel's Q1 2021 revenues is 19.67B TSMC market cap @553.16B Intel market cap @229.20B TSMC already surpass Intel. https://www.google.com/finance/quote/INTC:NASDAQ https://www.google.com/finance/quote/INTC:NASDAQ https://www.google.com/finance/quote/TSM:NYSE https://www.google.com/finance/quote/TSM:NYSE
- greenknight 5y agoTSMCs Q1 Revenue was 12.92B USD -- https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2021-04/cf13f8bbe93d30e1c654a8711ca3aa5ebcfdcd39/1Q21Presentation%28E%29.pdf https://investor.tsmc.com/english/encrypt/files/encrypt_file... The 362.41B is in TWD
- MangoCoffee 5y agoah ok. google show b. i assumed its in USD.
- greenknight 5y agoIn any case, in this nanometer war, ASML will come out the victor.
- deleted 5y ago
- jiggawatts 5y agoI just spent a month upgrading a bunch of cloud servers to AMD EPYC, and it was such a nostalgic feeling to be able to speed up systems with a simple, risk-free hardware upgrade. Remember those days? When every server upgrade magically sped up everything by a factor of two or three? The stop button was pressed on that wonderful time for about a decade, but what seemed like an end of an era wasn't. It was just a pause, and now we're playing with ever faster tin once again. It feels almost strange for that era of ever increasing hardware speed to make such a forceful comeback after having been stuck at Intel 14nm in the server space for years. The upgrades to 7nm AMD are already pretty impressive. I hear good things about Apple's 5nm laptop chips. Server chips on 3nm should blow everyone's frigging minds. Exciting times!
- prox 5y agoBut we can’t go much smaller can we? Totally not well versed in the chip space, so what can we expect?
- MangoCoffee 5y agoadvance packaging, chiplet, 3DFabric...etc
- peheje 5y agoIf history has taught us anything it's that technology won't stop evolving. And whenever humans thinks surely we've reached the peak of technological advancements, time proves us wrong. One thing is for sure, it's going to look very different from what we can imagine today.
- Retric 5y agoTechnology stagnated all the time in history. The pattern is new approach gets perfected resulting in ever smaller gains and different tradeoffs. Look at say rings where personal skill plays a larger role than the advancements in technology. We might be better at mining gold today, but that doesn’t translate into a better ring. Include longevity as one of the points of comparison and a lot of progresses looks like a step back. Cheaper and doesn’t last as long has been a tradeoff people have been willing to make for thousands of years.
- dilawar 5y agoHow these nm based metric relates with transistor density or flops per watt?
- bullen 5y agoI think we're going to be stuck at ~2Gflops/W CPU. If we ever see 3+ on cheap low power consumer hardware I'll eat my shorts.
- sschueller 5y agoFrom what I have gathered the smallest trace spacing is 3nm but not all traces. So it would be interesting to know what percentage is and the progress in getting all or most to 3nm.
- rob74 5y agoI'm not an expert, but from what I have been reading these "node names" have been pretty much decoupled from actual trace dimensions for a few years already. So I wouldn't be surprised to find out that most of the structures are in fact bigger than 3nm. To quote Wikipedia (on the 5nm process!): The term "5 nanometer" has no relation to any actual physical feature (such as gate length, metal pitch or gate pitch) of the transistors. It is a commercial or marketing term used by the chip fabrication industry to refer to a new, improved generation of silicon semiconductor chips in terms of increased transistor density, increased speed and reduced power consumption.
- intrasight 5y agoMy understanding is that it's more like the marks on a ruler. And the process shrinks the whole ruler. So it does ultimately relate to the feature size.
- lincpa 5y agoIt is predicted that Intel will also switch to my "warehouse/workshop model" to integrate multiple cores (workshop) and memory (warehouse) in the chip, which requires 3nm. Apple M1 has adopted my "warehouse/workshop model" and succeeded. My "warehouse/workshop model": https://github.com/linpengcheng/PurefunctionPipelineDataflow#The-unification-with-Computer-Hardware-Architecture https://github.com/linpengcheng/PurefunctionPipelineDataflow... Disscussions: https://www.reddit.com/r/programming/comments/o0gxy3/prediction_intel_will_use_riscv_plus_x86/ https://www.reddit.com/r/programming/comments/o0gxy3/predict...
- nly 5y agoOne has to wonder how long it will be before Apple build their own fabs
- eyesee 5y agoI wouldn’t count on it. Apple is vertically integrated, but doesn’t do their own manufacturing. It’s not necessary when They can dominate the supply chain and hold a practical monopsony on the best production capacity anywhere. This is why everyone else was lagging to produce 5nm parts -- Apple simply bought 100% of TSMC’s capacity.
- narrator 5y agoThe amount of centralization of fab tech is astonishing. You've got all the sub 7nm fab tech that the whole world depends on for the next generation of technology coming out of a few fab plants in Taiwan. Software is easy. Material science is hard. There is one company that makes the machines that the Fab uses. They cost several hundred million each, have a multi-year backlog and only TSMC has really mastered integrating them into a manufacturing process that can scale. Thankfully, things are getting more diversified. TSMC just started construction of one of several new plants in Arizona last month: https://www.phonearena.com/news/construction-of-tsmc-5nm-fab-in-phoenix-begins_id132524 https://www.phonearena.com/news/construction-of-tsmc-5nm-fab...
- davidf18 5y ago> They cost several hundred million each The machines you are referring to are the ASML EUV which go for $100 to $120 million.
- cwizou 5y agoCan't help but feeling it's a bit of a diversion to make everyone forget the just announced delays of their 10nm server chips. The original source, Nikkei, points to their own sources that claim that both Apple and Intel are currently doing early testing on 3nm. The article doesn't imply they are the only ones and according to Daniel Nenni, AMD did get the PDK too. Then it's about who's ready first and who bought the most allocation. Rumours points to Intel buying a lot of it, although when that allocation comes is a bit unclear. Nobody should expect anyone but Apple to have first dib at anything at TSMC. Plus it is Intel's first use of a bleeding edge PDK from TSMC, which is dramatically different from the older nodes they used so far (EUV and all). But it's been a long coming commitment (pre Gelsinger) from them to outsource their high perf DT and server chips for this node. Their volume needs for this are high (not quite iPhone high, but still far above non-Apple customers) and it will be interesting to see what, and how soon they can launch in 2023. I would probably expect Intel's lineup, especially the articulation around their so called HEDT (High End Desktop) segment to be shaken up a bit. So would their margins. https://www.intel.com/content/www/us/en/newsroom/opinion/updates-next-gen-data-center-platform-sapphire-rapids.html https://www.intel.com/content/www/us/en/newsroom/opinion/upd... https://asia.nikkei.com/Business/Tech/Semiconductors/Apple-and-Intel-become-first-to-adopt-TSMC-s-latest-chip-tech https://asia.nikkei.com/Business/Tech/Semiconductors/Apple-a... https://semiwiki.com/forum/index.php?threads/apple-and-intel-become-first-to-adopt-tsmcs-latest-chip-tech.14410/ https://semiwiki.com/forum/index.php?threads/apple-and-intel...
- ac29 5y agoIntel is already shipping 10nm server chips. See, for example: https://www.anandtech.com/show/16594/intel-3rd-gen-xeon-scalable-review https://www.anandtech.com/show/16594/intel-3rd-gen-xeon-scal... The article you linked is about their next generation server chip (which is also 10nm).
- alkonaut 5y ago> If Intel can get back to a two-year node cadence with 2x density improvements around mid-decade If I cut my kilometer time in half every six months I’ll do that 3h marathon next year!
- jonplackett 5y ago> It is thought that the iPad will be first to get 3nm chips Is anyone really pushing the current iPads tech specs? Until they massively beef up iPadOS I’m not sure this will benefit anyone particularly. The battery already lasts all day.
- gameswithgo 5y agoyou could shrink the battery
- Tempest1981 5y agoHigh-end games
- elorant 5y agoAre there triple A games for iPad?
- jonplackett 5y agoExactly. And if there were is that a reason anyone’s buying an iPad?
- veselin 5y agoI was wondering why AMD didn't finally get access to the top notch process at TSMC. They can have some of the best margins in their products, just because they ended being the best now and they certainly have the volume already. They didn't even start addressing notebooks or the lower ends of the market with Zen 3. Zen 3 is probably the fastest CPU on the market, it is not even scaled down to N5 or N6 which could get power consumption down to the best we have seen for notebook market. Instead the rumors are that in 2022 the improvements will go mostly into packaging extra cache. And now the next processes are already booked by others.
- selectodude 5y agoBecause AMD doesn’t have any money. Intel can fire off a 1 billion dollar wire transfer tomorrow. That would wipe out half of AMD’s cash reserves. Money talks.
- colinmhayes 5y agoIntel has 20 billion in profit a year. AMDs revenue is half that. AMD can't compete with apple and intel's cash flow.
- the_lucifer 5y agoBut tbh Apple has been consistent in their TSMC orders so far. Intel is the dark horse here. It looks like Intel's trying to throw a wrench in AMD's future plans while they try hard to catch up.
- swiley 5y agoWe reached the point a long time ago where the freedom of the platform is far more important than performance for all but the most demanding tasks (really just video editing at this point.)
- ghaff 5y ago>performance for all but the most demanding tasks Client-side performance is increasingly irrelevant except for, as you say, video editing and a vanishingly small slice of gaming (and that is at least as much about GPUs anyway). But then, clients outside of phones are also increasingly commodities anyway. However, performance on servers remains incredibly important. Of course, you can just throw more hardware at the problem which increases costs but is otherwise perfectly doable for most applications.
- zozbot234 5y ago> Client-side performance is increasingly irrelevant except for, as you say, video editing and a vanishingly small slice of gaming I really don't see it that way. Web browsing used to be considered a very light task that almost any hardware could handle, but the performance demands have been steadily climbing for quite some time.
- ghaff 5y agoPerformance and memory requirements have doubtless gone up but I use a five year old MacBook Pro and it's perfectly fine for browsing. Performance isn't really irrelevant of course but browsing generally doesn't push anywhere close to the limits of available processors.
- jnjj33j3j 5y agoMeanwhile, zero innovation in Europe... No wonder considering that even in Germany (country with highest SW engineer salaries in Europe), an average developer will only make 2500 per month after tax... For that money, you will only be able to buy a bread and butter, nothing else.
- RealityVoid 5y agoLast time I checked ASML is an European company. Regardless, things are... Different in the EU and there are plenty of complaints regarding innovation, but this tired talking point about EU stagnation lacks nuance and is not true.
- akvadrako 5y agoThe machines that power TSMC are exclusively from Europe. Also the average programmer salaries might be low, but there is plenty of full-time freelance work in Europe with competitive rates; a decent gig in NL is around €100/hour.
- moooo99 5y ago> For that money, you will only be able to buy a bread and butter, nothing else. Maybe you're trolling or maybe you're just actually clueless, but when looking at how much people can afford, its not sufficient to convert the euro to dollar amounts
- cma 5y agoAt this point ASML has a higher market cap than Intel.
- ksec 5y agoI thought it was worth pointing out, Nikkei, ( the original source of this pieces ) and DigiTimes has exceptionally poor track record on anything TSMC. So exceptional that every prediction or rumours have been wrong. Everything they got right was restating what TSMC said. My suggestion for the HN community is whenever you see TSMC in the headline read it with the biggest grain of salt. Unless it is coming from a fairly reputable site. ( Anandtech for example )