I'm curious what precisely the IBM alliance will end up including. The article mentions potentially giving IBM access to Intel processes, but I can't imagine IBM is in a great hurry to have its processors manufactured by a direct competitor.
IBM sold their fabs to GloFo, so in that sense IBM isn't producing processors and they're not offering foundry services. But they're definitely still in the game of designing processors such as the POWER series, which are sold in some non-IBM machines and do compete against Intel's processors.
You're missing the point; IBM is (mostly - not entirely, as Inspur and RCS can attest) out of the merchant processor business, but Power systems (and to a significantly smaller extent, z systems) absolutely do compete with Xeon systems.
Merchant processor business is the only thing that matters in this context. IBM retains the right to use TSMC or whatever foundry they want if they feel that Intel is treating them unfairly. But there is no reason for Intel to do so. For the most part IBM is a software company now. Companies buying their servers do it for reasons other than processor merits (as long as they are not completely junk)
That's dumb. You do realise they'll need nearly two years to move production away, right? During that 2 years you expect IBM to do what? Sell air?
POWER10 was delayed by over a year because GloFo backed out of 7nm so they had to switch to Samsung, that was in 2018 and POWER10 was still in logic designing phase, no actual physical design even started. You think IBM would use Intel Foundry for their lifeblood? You are insane.
who said they have to use Intel process? sure you need to plan your chip for the manufacturing requirements of the process but CPUs designs are portable and if Intel foundries do not make the cut either deadline, yield, performance etc etc they can go to TSMC or Samsung and get their services, albeit they need to sign a deal with enough time margin. Remember the time Apple produced the same process at the same time at Samsung and TSMC - it wasn't ideal but it served their needs at that time to not be depended on one foundry. I think IBM knows it cannot rely just on Samsung and TSMC for various reason and this deal if successful can have big advantages.
It might also be that the US government is a major customer for IBM's systems, and several of those contracts require that the chips are made by a trusted foundry. If Intel really gets their 7 nm EUV up-and-running and fabs chips for IBM right here in the US, IBM could move to a better node and still sell to DoD or the FBI.
The US government has a pathetic concept of competition.
For instance, it announced that it would have two leading supercomputers built, one powered by Intel and the other powered by AMD.
• two companies is hardly amazing competition
• they're using the same x86 architecture, too
The government, though, seems to be very content with duopoly (which is generally a quasi-monopoly, with the second company being quite a bit smaller/weaker) serving as the ostensible competition in its ostensible capitalism.
'For instance, it announced that it would have two leading supercomputers built, one powered by Intel and the other powered by AMD.'
I forgot to mention that it specifically cited the reason for doing it that way was to ensure adequate competition (which a rather farcical claim as I pointed out already).
So what if Power systems compete with Intel systems? IBM is a software and services company first and foremost. They have been for a long time. They used to sell both x86 and Power servers. Were they competing with themselves at that time? Apple and Samsung are much bigger competitors to each other than Intel and IBM and Apple will use Samsung's foundries and components.
Apple is using TSMC foundries and is buying up 25% of their capacity. They are first ones on the 5nm process node with A14 and M1. Samsung is providing many components but their foundries have fallen behind.
IBM isn't even a good software & services company anymore (excepting RedHat) I think they could leverage Power RISC IP and Intel expertise in manufacturing and power management to create a compelling alternative to ARM if they wanted to...
POWER and Z don't directly compete with Xeon. They do similar things but the software is not easily migrated from one to the other. This is especially true of Z. If you are running Z systems you are probably leveraging a stack that has a multidecade lineage that stretches back to classic mainframes. While it is possible to engineer that to work on x86 in a Windows or *nix environment, its not trivial or necessarily cost effective to do so.
So yeah you can move between them but its not a decision easily made such as replacing Xeon servers with Epyc for example.
Depends on how you define competitor. IBM does compete in chip designs with POWER10 coming out this year. While IBM has been desperately trying to expand its user base for POWER, the majority are continued legacy customers.
IBM used to be a chip manufacturing competitor and operated as a foundry also producing 3rd party designs. (The nVidia 6800GTX was manufactured at IBM.) As pointed out by others here, IBM sold off their manufacturing line to Global Foundries years ago. The fine print is that IBM never relinquished their patents nor halted their R&D wing. The theory was that IBM would continue to fill various patents and force some sort of partnership or worse, legal arbitration with the remaining foundries on various patents. That aspect seems to be coming true but with an interesting asterisk. If I had to guess, Intel is preparing to ditch silicon in some fashion with a strong possibility of going optical in some fashion. Between Intel and IBM there was a collection of patents that would encompass optical interconnects and motherboards. IBM may also gain access to Intel's fabs for their own products instead of relaying on Samsung currently.
The last time Intel collaborated with IBM in semiconductors was in the ill-fated 450mm wafer project in Albany.
My guess is that now IBM could collaborate with Intel in GAA-FET processes.
IBM may still be a competitor to some of Intel's products, but as IBM's main client for those same chips is the US government they would certainly prefer that IBM makes those products in a US foundry rather than at Samsung or TSMC.
The biggest potential news here is x86 licensing. There are zero details, so I have doubts whether Intel will do it fairly. Does Intel have enough faith in its architectures to open up its golden goose ISA?
Intel is incredibly insecure & irrationally jealous of its x86 ISA, including when Gelsinger was a top engineer:
One big mistake about Gelsinger's praises: he was in the top divisions at Intel when Intel was at its most anti-competitive and shameless. Gelsinger was at Intel from 1989 to 2009, then *the head* of the desktop & server CPU division (then called the "Digital Enterprise Group").
1989: Gelsinger joins Intel 1991: AMD sues Intel (Intel pays $10 million in damages + forced to issue AMD x86 licenses) 1997: DEC sues Intel (Intel forced to buy $700m DEC + forced into licensing at $1.5 billion) 1997: Intergraph sues Intel (Intel pays $225m settlement) 1999: FTC investigates Intel under the FTC Act (Section 5) 2005: AMD sues Intel (Intel pays $1.25b settlement) 2009: FTC sues Intel (Intel pays $10m settlement) 2009: Intel threatens to terminate AMD's x86 cross-license 2009: Gelsinger leaves Intel
Gelsinger was literally at his prime when Intel was at its worst.
Most of their IDM customers will be doing things like cellular modems or hyper scaler in-house ARM server parts and not Intel x86 IP, so the days of their easy IP margins are over. They're pretty much also giving their customers the means to ditch their own high margin server parts too by enabling the competition. When their factories are finally done, so will TSMC's and Samsung's so this competition will further crush their once excellent margins.
They aren't going to have any IDM customers unless they can fix their fab process. Apparently people forget history, Intel's been offering fab services since 2016 but no one uses them because they are so far behind in process tech and they can't guarantee wafers because they can't even manufacture enough to meet their own internal demand.
I get they want to offer fab services but honestly until they can prove they can advance reliably past 14/12nm they aren't going to have real customers. Anyone willing to use 14nm has ample choices that will be cheaper with guaranteed wafers, anyone needing 10/7nm is going to either use TSMC or Samsung as they are the only two fabs with working processes ATM.
The only reason I can think that Intel made this announcement is for Stock reasons as there is little to no chance of this being a reality until they can demonstrate an ability to fab something at smaller than 14/12nm because they simply don't even have the capacity to fab their own processors let alone outside companies.
All of us Arsians and anyone paying attention know this will not have a big impact now.
However, it makes me happy to know that The new intel leadership has chosen a direction and is moving.
I have been gaming Intel stock price for a few months now, and it is marvelous to know that current leadership has a plan and direction that EXPANDS intel's possibilities.
So, even if it is just an announcement for shareholders, it’s the correct announcement for those of us who have been betting that Intel can come out of this period of mode stagnation as a stronger company.
Also: I feel like it’s a big deal to keep a decent volume of semiconductor manufacturing on US soil. We are a shitty country with a shitty, violent, aggressive history. It’s best we make sure it’s hard for anyone to try to make us pay for all the awful.
Violent and agressive? Get out of here with your anti american rhetoric, you trash. Only reason you have a pc to type on is because of the US. Did you never read of the horrors of china or russia? Imbecile.
Another point: licensing x86 will more quickly crush their margins by eliminating the small moat they have left in binary compatibility with tons of legacy software. I think they won't be doing any licensing and will jealously guard against the onslaught of ARM servers using emulation for x86 compatibility.
We've been hearing about "muh ARM servers" for a decade now. I was hearing these things in high school, how ARM was not laden with legacy code and how efficient it was and how it would destroy intel in the server market.
This was 2013.
Today its 2021, and I'm hearing "ARM is not laden with legacy code and how efficient it is and how it will destroy intel in the server market". Reminds me of fusion reactors.
Today ARM servers are being produced and shipped. ARM processors deliver some of the best multi-threaded performance money can buy. x86 only survives because many current processes were designed only for x86. ARM servers will be coming en masse to overthrow x86 in the coming decade because they already are.
Ah baloney. The only ARM server processors even available only fit a certain niche. ARM has made incremental improvements with each generation, but AWS's Graviton was the first viable server processors and we have no idea what kind of economics are involved because AWS could be choosing to loose money on the processor to gain market-share and experience.
The fact is you can't buy a ARM server right now that will be competitive with x86 in every category. Yes there are a couple options that work in certain situations but the TCO is still out of wack and IMO non-viable against x86, particularly with the strides AMD has made with Zen.
If servers still relied only on Intel I have no doubt ARM would be a much more viable competitor but that's simply not the case.
This is very good info to keep in mind thanks for posting the timelines. Gelsinger would ultimately have to sign off on the terrible pricing of the 11th gen high end sku's as well where at the top end Intel are trying to price an 8 core part with worse performance, thermals, and power against AMD's 12 core part which also has a slightly lower MSRP. The ultimate decision to push that absurd pricing must be signed off from the top.
Who would want to license x86 for future CPU, when ARM is available with better ISA. Is backwards compatibility in old software that important? Or would the licence just the x86-64 commands (which I think AMD has something to say in). Wouldn't someone licensing x86 get rid of the old 32 bit compatibility and just support 64 bit?
There is no ISA advantage for ARM. It's a BS talking point that has no basis in reality. x86 processors haven't used x86 internally in more than 20 years. Every x86 processor uses an internal flexible RISC core and tacks a 128k transistor translator on to handle the conversion from x86 to the internal language. That translator was a big part of the processor 20 years ago, today it's not even 0.01% of the die.
If anything, the x86 translator actually gives the advantage to x86 because every single x86 processor can have a custom designed RISC core that is optimized for each process node.
That is a rather disingenuous argument. One of the ISA's biggest impacts is on instruction decoding, and you're simply ignoring it by waving away all the stages that happen before it becomes RISC-like internally. The variable-length instructions in x86 make it exceedingly difficult to avoid longer pipelines for decoding or to achieve a greater issue rate at high frequency. You blow through power and area searching for the ends of instructions and doing that translation, and the cost of any pipeline flush ends up being higher.
One mitigating factor may be that code size (and cache pressure) is reduced there, but I'm fairly certain various papers have argued those benefits are pretty small against ISAs with a compressed format for common instructions.
Phoenix has the largest nuclear power plant in the country, it does not snow there and land is relatively cheap for being the fifth largest city in the country. It is the perfect place for fabs.
Last time I was there most of Arizona is sitting on a huge water aquifer. I could not believe paper plants were in the desert south of Phoenix but it is due to the water.
Regardless of the water issues, it seems unnecessarily risky to concentrate fabs in one area. Even if you're not worried about the national security concerns, there's the possibility of man-made or natural disasters that could wipe out production. If serious damage occurs, it could be a year+ before they're back online given the waiting list for key equipment.
That's already true to a large extent today with how many fabs are concentrated in Taiwan. And in the US, not all of the capacity is in Arizona, even after Intel's expansion -- there's a big manufacturing presence in Oregon as well. There's some risk in geographical concentration, but there's some advantages as well: it's much easier to hire good employees if you're not the only company in the area doing that type of work. And no matter where it is, a natural disaster hitting a leading-edge node foundry will affect worldwide supply -- these things are too expensive for true redundancy.
Have to confess, the prospect of US-fabs with TSMC or larger levels of wafer capacity has quite the allure. The more dominant AMD becomes, the more fab capacity Intel would have to offer.
Currently Intel fabs nearly as much silicon at TSMC as AMD does... which is mind-boggling to consider when AMD serves two console launches, plus mobile+desktop+server CPUs, and then GPUs on the side. It appears that Intel bringing more of its fabrication in house would open up almost double AMD's current wafer allotment capacity at TSMC, and AMD would certainly could put most of that to use.
Intel has a lot more to manufacture in house now than it ever had before. The Altera acquisition alone shifts a good chunk of what was expected to be built at TSMC to Intel's own in-house foundries. Due to the delays of 10 nm and 7 nm, only now is that really starting to happen.
That is also only one of Intel's recent acquisitions. There are numerous other companies Intel purchased who still have outstanding orders with TSMC that will also shift toward Intel's own fabs over time.
Kougar-->"It appears that Intel bringing more of its fabrication in house would open up almost double AMD's current wafer allotment capacity at TSMC"
Not as much as that, a good amount of what TSMC makes for Intel are chipsets and network chips which are not made with leading edge processes. Yes Intel will probably move Altera and Mobileye stuff in house, but I don't see that happening until 2024 and also starting later this year Intel will be increasing their leading edge wafers starts at TSMC as begin to make GPUs and next year CPUs there.
Aye, but AMD doesn't use leading edge TSMC nodes either, Apple gets dibs on those. Then there's the IO hub die which are still 12nm GloFo. AMD needs a very large supply of those particularly if we're speaking in terms of physical size.
Zen 4 / Genoa is confirmed to be using a new IO die thanks in part to the power issue with Milan's IO chips as well as the changes required for DDR5. I've not seen anyone mention what node or fab AMD will use yet, but as a random guess it seems unlikely they could source enough 7nm TSMC for it. But it won't stay on 12nm GloFo, Milan's IO die power consumption will make sure of that much.
Isn't Genoa compute chiplets confirmed to be manufactured on 5nm? That would free enough capacity of 7nm for 7nm IO chiplets, wouldn't it? TSMC has also been expanding their 7nm capacity.
Yes, 5nm Zen 4 is confirmed. And theoretically yes, it should... but so many things use 7nm TSMC that we won't know until it's confirmed. AMD may stick with a larger, cheaper node at TSMC or Samsung.
They probably will far into the future when it's cheap to do so, EUV machines from ASML are already backordered a full year now. But with Intel, TSMC, and Samsung each planning new fabs that backorder is going to get worse before it gets better.
In any case, there's nothing GloFo can do right now. AMD needs a big supply of shrunk IO chips no later than Q4
I miss my edit button.... For example Xilinx chips are 16nm TSMC. The Xbox One is TSMC 16FF+, as another example. Microsoft likely preplanned it that way with a die shrink at some date in the future. Probably others as well I can't think of.
Intel recently announced a DARPA partnership where they'll use Intel 10nm for eASIC chips, rather than TSM 16nm. I wonder if that program will also be planned to use the new foundry.
What i want to see is more fab plants and also spread throughtout the world. American EU etc. Too many eggs in one basket is NOT a good thing. Any country that relies too much on foreign goods without some local balance, especially in Important field like fab plants etc leaves itself a hostage to fortune. The UK has hardly ANY Steel production left, most of it is imported and that leaves us vulnerable.
Keep in mind that Intel's TSMC manufacturing use today may not be entirely leading-node processes. Probably a fair amount of that volume is for things like networking chips, FPGAs, PCHs, etc., which don't always need to be on the most advanced process the same way CPUs and GPUs do.
The presumption from the article is that Meteor Lake will be leveraging an IO die announced at a later date. While it is clear that the IO die hasn't been formally announced, it could actually release prior to Meteor Lake: it could appear as a companion to a 2022 chip using 10 nm ESF using similar packaging techniques. Intel's 2023 move would be to slip in the 7 nm tile while heavily reusing other existing tile designs. That's the real benefit of chiplets is being able to recycle parts between generations
There are actually more opportunities in Pat Gelsigner's strategy. SiFive, a startup, said that it will work with Intel to make the RISC-V designs availabe to customers of Intel's Foundry Services.
So if Intel can meet its predictions (unlikely!!) they will have 7nm CPUs about the same time as AMD CPUs are on 3nm. Given the backorders on EUV equipment from ASML, Intel could easily find that its new fabs are unable to produce 7nm products until well after their proposed timeline. There is also a big question as to how well an Intel chiplet design will work - a lot of the current IP will have to be completely replaced (eg the ring buses). As with AMD any multiple CPU chiplet design will have much longer delays than current Intel designs for off chiplet CPU to CPU communication.
Apple will be on 3nm by 2022 and then use it for their A and M APUs and dGPUs. AMD will have to wait a while for TSMC3nm until TSMC ramps up lots of capacity or Apple moves to TSMC 2nm.
AMD will probably 2nd source TSMCs 5P with SAMSUNG 3nm GAAP on GPUs etc. but wont switch to TSMC3/2nm anytime soon and only have CDNA2 on 5 nm in 2021 and follows up with other 5nm designs in Q3/Q4 2022...
Intels 7nm SuperFin is probably comparable to TSMC 5P in density - nit sure avout other metrics.
This is a bunch of talk. It's been going on for years now. Just look one year ago when Intel was guaranteeing ice lake server by 4qtr 2020.
Intel is now four years behind AMDs chiplet design. Even with perfect execution from here on, the idea that they will be able to reverse, much less stop, AMD continuing to gain market share and revenue, is a fools dream.
I think the same. Intel will not be able to buy their way out of this mess this time. Anti-competitive behaviors will not fly this time around. They can't win against TSMC, it is over, they are the industry leaders for the next decade.
Intel as foundry ? Maybe only for trailing edge process where margins are low, and they will be competing with plenty of capacity at TSMC / Samsung / SMIC and others, so right amount of clients won't be guaranted.
Their 7nm is 'simplified', Simplifed process means that they can't afford for 'full node progress', and must rely on 'half node progress', just like TSMC and Samsung, but TSMC is two years ahead of Intel. That's why they incorporated IDM 2.0, and design CPU in two variants - highest SKU (i9, i7, i5) for TSMC, while other slower SKU for their trailing fabs. This is proof that they won't get manufacturing leadership in future.
I can see intel totally competitive as a foundry. The fabless companies would actually benefit the most by having a third cutting edge node option.
Intel is recognizing here that TSMC and Samsung are as much their competitor as AMD, NVidia and Apple.
At this point with ASML only able to ship 50 EUV tools a year then any that intel purchase is capacity that Samsung or TSMC can't add. It gets a double benefit from increasing capacity at the high end and the same with purchasing production at other fabs. Demand for semiconductor products is in an exponential growth phase and no one will be able to satisfy it.
Biggest issue for INTEL and SAMSUNG currently might be EUV capacity and ASML supply. That's why they cannot supply high volume EUV nodes as they lack EUV capacity and their last DUV nodes are 8nm Samsung and 10nm Enhanced SuperFin for Intel.
Since supply is tight and even if they could retooling all their DUV fabs would be an economic dissaster it makes it more than reasonable to use 22, 14 and 10nm for IoT, Automotive and controllers. Those nodes are enough for lots of applications there...
At current market and even in 2022 it probably doesn't matter which node is ompetitive as they can sell anything right now! So they just make crap to gold now and focus on 2023/2024 ;)
EUV pellicles are nearly ready for commercialisation. Much more competitive defect rates once these are introduced, probably, and maybe EUV equipment will become less complex after that.
new keyboard ... amd suggested for a couple of yrs that they were moving to 7N+ for Milan. Reviewers took that to mean the EUV version of 7N.
I'd guess tsm shares EUV equipment for 5nm, and will do the same for 3nm, so it isn't clear to me that apple moving to 3nm will free up that equipment for amd.
Meh. It appears that Intel is definitely no match to TSMC, hoping for more availability or cheaper ASML equipment by 2024. I reckon, the AMD vs Intel narrative will remain the same for several years.
Must be a hard pill right now for all those that dissed AMD for going the chiplet way with Zen. Once Intel gets it manufacturing fixed it’ll be good competition.
They are 5 years behind AMD when it comes to Infinity Fabric and Chiplet designs. With AMD relentless execution, Intel will never catch up unless AMD screw up big time.
supdawgwtfd of course he does, you havent noticed he praises intel, while bashes amd in most of his posts ? to say that AMD's current IF wont handle PCIe 5 bandwith, is like saying Intels DMI wont handle PCIe 7 bandwidth. to say amd OR intel wont update their respective interconnects as PCIe bandwidth also increases, is well, just plain stupid.
The real question? Where in the hell do they plan to build it in Chandler? Space is already at a premium and there's next to no land left at the Ocotillo campus, which is 10 minutes from my front doorstep. Unless they're planning on building on what little space remains and what may not be enough for another fab the size of F42, their options are truly limited.
Nice to know they’re back in the game. 2022 will be a great year, this year is just more cores, more Watts, more 14nm. Thanks for smacking Intel around AMD.
Below is a google map link to the Ocotillo site, looking at it I think they still have room on the existing land, especially if the build a parking garage.
The dirt to the west (left) of the Fab 42 location is Native American land. They're not giving that up. They'd have to build south of Chaparral Way (where the green land is) if they even own it or to the spot currently occupied by some existing structure immediately south of F42.
All smoke and mirrors. They killed real product Optane 3D X point. They sold mobile 5G to Apple. Got booted off by Apple.
Lost all the Fab lead to TSMC long back. No HEDT successor since 2 years. Mainstream RKL is DOA except for Gen 4 platform with DMI upgrade at high cost of Power consumption and lack of Core scaling upgrades. Lost to EPYC Rome and now Milan 7003 CPUs.
TSMC and Samsung now leading. Okay maybe compete with GloFo for the Semi industry for multiple contracts and all but their biggest news of IP licensing is shady and not clear at all. Why will Intel sell their top x86 IP to some corporate for SoC ? Microsoft is already with AMD 7N custom SQ1 chip. Who want to get Intel lost lead uArch and their failed 10nm ? On top non comparable to ARM since Intel does their own business with selling x86 parts while ARM doesn't. That's a major thing. This is direct competition. How does it even work ? EMIB failed with Kaby G. Foveros ? Its great but AMD is not dumb neither is Apple nor is Ngreedia. Nor is Samsung lmao everyone is having their own direction.
FFS get the damn 10nmSF properly beat Zen 3 and upcoming Zen 4 (but with ADL having 8 only x86 big cores and small pathetic Non HT Atom cores) how is it possible. Ice Lake Xeon already looking outdated vs 3PYC Milan.
Why don't they STFU and prove it with a strong product ? Damn. All I wish is x86 be healthy and competitve thanks to AMD. For that since this uArch puts people in power rather than corpos in power (ARM, highly custom and need more OEM help - Qcomm CAF, Blobs and etc for Android CTS / Updates plus performance improvements vs x86 stability back compat AND DIY scene with PC market.
Intel in 2005 had an advantage though: While their american engineers were farting around with netburst, their israeli team had been refining the pentium III core into the pentium M, which was doing quite well. That eventually became core 2.
Today, in the mobile space intel does have sunny cove CPUs, but they are restricted by the 10nm node, just like their desktop parts. After 2 years we still only have quad core parts, with promises of an 8 core part.
It all coems down to fabs, unlike in 2005 where intel just had to dump a bad design, intel now has to overcome a manufacturing challenge that they have spent 7 years on with little apparent success. I imagine it will take at least another 3-4 years before either their 10nm or 7nm pan out any significant improvements.
So this is closer to intel in 2002, hurting from AMD smacking them around and about to get another handful with athlon64, I mean, zen 4.
"their israeli team had been refining the pentium III core into the pentium M"
I think credit needs to be given to Intel Israel more often. Usually, we think of Intel as one abstract entity, and those engineers are forgotten. Besides Pentium MMX and M, they designed Sandy Bridge, Skylake, and Sunny Cove. Perhaps Core/Conroe too. Anyhow, Conroe was based strongly on the Pentium M. Quite likely, the successor to Sunny Cove is on their drawing boards as we speak, if it hasn't been taped out already.
Since Golden Cove-based Adler Lake is supposed to ship this year (Willow Cove-based Tiger Lake already replaced Sunny Cove-based Ice Lake for U-series chips, but Sunny Cove will still be in Ice Lake SP server CPUs; Sapphire Rapids is Willow Cove), I think it's safe to say it's taped out already. Ocean Cove (the successor to Golden Cove) is supposed to tape out later this year and launch in Meteor Lake CPUs Q1 2023.
You're right, but when I wrote that, I meant Sunny as the latest microarchitecture, perhaps wrongly, because all Willow did was alter the caches, as far as I know. Concerning Golden Cove, is it just a microwaving of Sunny/Willow or is it a proper revision of the architecture? Can't seem to find much data on it.
intel statement is that Optane plans are not changed by Micron's exit. Micron will complete their supply obligations for gen2. Intel and Micron were already split on gen3 and beyond.
I guess this is "The Empire Strikes Back", or, at least, announces that it intends to do so (see fine print for disclaimers and details). On a more serious note, took them long enough. Intel's strength for the last two decades came also from being better at fabbing chips than the competition, and without that, their competitive edge has dulled a lot.
Probably pretty easily if TSMC doesn't have any spare capacity, as has frequently been the case recently. Especially if it's also cheaper (which remains to be seen).
TSMC has being holding the mature nodes market firmly before 2010, when it was far less competitive and sophiticated than Samsung or Intel as for advanced ones. Mature nodes are the foundation of the TSMC empire we know today, I don't think intel can have an edge over them or even GF in this market.
Sure, he gets it. The problem remains if they can get their foundry problems in order enough to close the gap of the moving target and then maintain cadence with their arch-rival who is led by an equally brilliant CEO.
He can't compare to what Lisa did. She is literally... THE SAVIOR... of AMD. She didn't just turned around a company on the verge of bankruptcy, she literally turn it around to the industry standard.
He get it, but will not be able to change ANYTHING... TSMC and AMD will continue their relentless execution which will force Intel to difficult choices...
Keeping their current fab strategy is a huge mistake. Basically, they are going to be the dog for now on.
Exactly. Intel might have a world-class CEO now at the helm, but this may be a case of too little, too late. AMD was able to get ahead not of Intel only because they had an amazing CEO, but because Intel had lousy management who meanwhile squandered their time away. Think of this in terms of a timeline of getting two world-class triathletes. AMD got theirs when Intel had a sluggish sloth in the race and they darted into the lead and kept running faster and faster miles into the distance. Intel finally has gotten their star player but AMD still has theirs, and she has gotten her team a good 10 to 20-mile lead and is not letting up. I am not counting out anything here because the unexpected has happened so many times throughout the history of the tech industry. However, all Gelsinger may end up doing is preventing Intel from sinking anymore than actually ever restoring market leadership, but that is only one possibility of many. Heck, Sega used to have the world's best gaming graphics a whole five years ahead of the PC when they released Daytona USA to arcades in 1993. But by the 2000s, Sega's console business was deader than a doornail and they were using embedded x86 systems that were a bit behind the best in PC graphics. We just never know what might happen with how volatile the tech industry is.
And TSMC will have 3nm by 2023, and Intel will be utilizing their nodes. That's a slap in the face to their manufacturing leadership. It proves that they are no longer the world leader in silicon manufacturing. That crown goes to TSMC.
more like AMD persevered despite being on the verge of bankruptcy.
Intel was always flush with cash yet they couldn't bring 10nm to market in 2016 out their original roadmap. let's not forget their 14nm was delayed twice before that.
Intel fabs are what dragging them behind, not the contrary. The only reason why they stick to their fabs is because they can't get that kind of capacity on a better process. If they were going for the better product, they would lose half of their sales and cut their margins since they need to pay for taping out their chips instead of doing it in-house.
The only reason why Intel CEO is now going into the fab business is so they can buy waffer at TSMC and maintain their fab business... it is a pure admission of failure.
After the debable of texas having to cut power to their chip plants do these companies have to invest more in standby power. Can Arizona handle the peak loads that may come with all these new plants. America is renowed for failing to plan ahead. Rather pay dividends than invest in infrastructure.
That is Texas, which is a pathetic joke of a state. Texas choose to not be part of the federal electrical grid because it did not want to comply with "onerous" regulations associated with cold weather reliability.
Arizona is part of the federal electrical grid and did not and will not suffer the problems Texas had.
Because freak weather effects never shut down power ins tates like california, right? Teh rest of the US never has freak weather and power outages, no siree, its all perfectly fine, ONLY texas has weather outages.
The Texas power outage was a replay of what happened in 2011. After the 2011 outage it was identified that the cause of much of the issues was lack of winterization. The Texas power industry decided it wasn't worth the cost of winterizing the system and since they were unregulated they didn't have to.
Some 70 people died and millions were without some combination of heat, water and power for...but per the likes of Rick Perry..“Texans would be without electricity for longer than three days to keep the federal government out of their business,” which he was saying at the same time Governor Abbott complaining Binden hadn't approved enough federal disaster relief.
Each day of the outages they were billing their domestic customers at the same rate as they get in a year, ~$10B. ERCOT kept the price too high for too long which resulted in an overbilling of $16B, but whatever we can't fix that after the event.
Many wind providers will go bust, as they had to buy power at the high rates to cover the intermittency clauses in their supply contracts.
Yet the carbon based producers and the fuel suppliers that profited hugely as a result of their inability to operate have no sanctions, only giant C suite bonuses.
Ain't lack of regulation great, if you are a member of the good old boys club.
Intel announces a lot of things. Doesn't change the fact that:
- 14nm delayed by 1 year. - 10nm delayed 2 years (was due 2017), though more like 4 years since 10nm still not mature enough for desktop. - 7nm delayed 2 years (was due 2021, projected 2023)
A bit off topic, but this being an article on foundries, if I may, I'd like to ask how exactly do these processes even work at <= 5 nm, where I believe quantum tunnelling starts to get out of hand. Have they mitigated it, using special techniques? Or are they just working round that? I suppose physics is showing that Intel, AMD, and the rest will have to switch to something else soon.
Intel will lose against TSMC. They will never beat their nodes. TSMC will be on 3nm in 2023... and let me remind you that 7nm is garbage. Bob Swan said the following...
For perspective, rival foundry TSMC plans to be on the 3nm node in the same time frame as Intel's new schedule for 7nm. Intel clearly isn't pleased with its execution on the 7nm node, as an embattled Swan remarked that "And we feel pretty good about where we are, though we’re not happy. I’m not pleased with our 7nm process performance" at the end of the call after a bruising question and answer session with analysts. Swan also said "we have root-caused the [7nm] issue and believe there are no fundamental roadblocks," and that the company would provide further updates at an upcoming Architecture Day.
Same old Intel...what the company needs very, very badly are new architecture x86 CPUs, but of course Intel is throwing tons of money out to see what happens--to see what the proverbial, distant, 7nm cat drags in. Intel's idea of progress is doubling, tripling its immediate overhead with seemingly no recognition of the fact that by the time Intel brings its FABs to parity with where TSMC is right now, 7nm, TSMC will be heavily into 5nm if not 3nm, and Intel will still be far behind. Typical Intel--does not know how to do much more than throw good money after bad (Gelsinger remembers Itanium well, I'm sure) and to make stupid assumptions about its invulnerability and longevity. AMD would simply not exist today if it was crippled by the same fiscal imperatives that drive Intel. AMD cannot afford to make the mistakes Intel has made and continues to make--which is probably exactly why Intel has dropped behind AMD to the massive extent that it has today. Bringing old dudes out of retirement to relive Intel's halcyon years is about as desperate as it gets, imo.
He's also a very honorable person unlike 99% of the scummy folks I have seen running these companies. He is the last person that would be having an extramarital affair. He also is a donor to many humanitarian agencies, not a corrupt money-laundering, self-serving foundation like you see Bill Gates and others in Silicon Valley using (e.g. Gates donates and invests tax-free money to the Gates Foundation, who, in turn, purchases Microsoft products and services to use to provide "free" charitable services to impoverished nations).
Thank you. I just read them now, and am beginning to like Pat even more. What a person. I just hope that some of his values can filter down through the rest of Intel.
'with Intel extending cooperation with the State of Arizona, as well as the current administration’s target of improving semiconductor manufacturing inside the country.'
Unless these fabs are going to be used to manufacture ARM CPUs or other non-x86 semiconductors, I don't see it making much of an impact on the need to rely on China / Taiwan.
$20B into new fabs, with Samsung considering as well. Translation: We expect China to take over Taiwan (i.e.,TSMC), and we're going to take the business.
EUV scanners are very complex machines with advanced technologies and parts that are difficult to manufacture do they can deal with the unfavorable physics of EUV. China has very low chance to develop this kind of machine on their own, the best they can do is try to steal the technology and even then making working machine will be difficult task.
So, the answer is: Yes... China will become equivalent to ASML at some point.
What we don't know is how long that will be.
This means the US and the EU need to change the situation by investing more to keep the edge. In fact, the US both have already lost the foundry edge and the US doesn't have an ASML.
LOL. Deleting my posts now...comic. NO swear words, just two links posted. I guess AMD portal site still...You guys hate Patrick Moorhead or something? Posting an article from competition blocked or what? Patrick worked at AMD for a decade...NO respect? pfft..
2023 ,maybe too late, for Intel's 7nm. But better late than never. As Intel has been practicing lately. 7nm at Intel's fabs will probably be equivalent to TSMC's 5nm. Only difference is that 5nm is on production now, not in 2023. If TSMC doesn't slow down, Intel will still have to compensate for lagging a node behind by offering a superior architecture. In order for Intel to be like 2006, it needs to produce at the same nm as TSMC, because Intel's nanometers tend to be a "node" better.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
159 Comments
Back to Article
SarahKerrigan - Tuesday, March 23, 2021 - link
I'm curious what precisely the IBM alliance will end up including. The article mentions potentially giving IBM access to Intel processes, but I can't imagine IBM is in a great hurry to have its processors manufactured by a direct competitor.lilo777 - Tuesday, March 23, 2021 - link
IBM is not a competitor to Intel. They are not producing processors for external customers (not for a long time now).Billy Tallis - Tuesday, March 23, 2021 - link
IBM sold their fabs to GloFo, so in that sense IBM isn't producing processors and they're not offering foundry services. But they're definitely still in the game of designing processors such as the POWER series, which are sold in some non-IBM machines and do compete against Intel's processors.SarahKerrigan - Tuesday, March 23, 2021 - link
You're missing the point; IBM is (mostly - not entirely, as Inspur and RCS can attest) out of the merchant processor business, but Power systems (and to a significantly smaller extent, z systems) absolutely do compete with Xeon systems.lilo777 - Tuesday, March 23, 2021 - link
Merchant processor business is the only thing that matters in this context. IBM retains the right to use TSMC or whatever foundry they want if they feel that Intel is treating them unfairly. But there is no reason for Intel to do so. For the most part IBM is a software company now. Companies buying their servers do it for reasons other than processor merits (as long as they are not completely junk)dotjaz - Sunday, March 28, 2021 - link
That's dumb. You do realise they'll need nearly two years to move production away, right? During that 2 years you expect IBM to do what? Sell air?POWER10 was delayed by over a year because GloFo backed out of 7nm so they had to switch to Samsung, that was in 2018 and POWER10 was still in logic designing phase, no actual physical design even started. You think IBM would use Intel Foundry for their lifeblood? You are insane.
Eliadbu - Thursday, April 22, 2021 - link
who said they have to use Intel process? sure you need to plan your chip for the manufacturing requirements of the process but CPUs designs are portable and if Intel foundries do not make the cut either deadline, yield, performance etc etc they can go to TSMC or Samsung and get their services, albeit they need to sign a deal with enough time margin. Remember the time Apple produced the same process at the same time at Samsung and TSMC - it wasn't ideal but it served their needs at that time to not be depended on one foundry. I think IBM knows it cannot rely just on Samsung and TSMC for various reason and this deal if successful can have big advantages.eastcoast_pete - Wednesday, March 24, 2021 - link
It might also be that the US government is a major customer for IBM's systems, and several of those contracts require that the chips are made by a trusted foundry. If Intel really gets their 7 nm EUV up-and-running and fabs chips for IBM right here in the US, IBM could move to a better node and still sell to DoD or the FBI.duploxxx - Wednesday, March 24, 2021 - link
exactly the one and only reason....StoneFish - Friday, March 26, 2021 - link
IBM has a huge and important government businessOxford Guy - Tuesday, April 6, 2021 - link
The US government has a pathetic concept of competition.For instance, it announced that it would have two leading supercomputers built, one powered by Intel and the other powered by AMD.
• two companies is hardly amazing competition
• they're using the same x86 architecture, too
The government, though, seems to be very content with duopoly (which is generally a quasi-monopoly, with the second company being quite a bit smaller/weaker) serving as the ostensible competition in its ostensible capitalism.
Oxford Guy - Tuesday, April 6, 2021 - link
'For instance, it announced that it would have two leading supercomputers built, one powered by Intel and the other powered by AMD.'I forgot to mention that it specifically cited the reason for doing it that way was to ensure adequate competition (which a rather farcical claim as I pointed out already).
brunis.dk - Monday, April 12, 2021 - link
Very much like their political system, an illusion of democracy, two shitty parties who both want endless war.Yojimbo - Wednesday, March 24, 2021 - link
So what if Power systems compete with Intel systems? IBM is a software and services company first and foremost. They have been for a long time. They used to sell both x86 and Power servers. Were they competing with themselves at that time? Apple and Samsung are much bigger competitors to each other than Intel and IBM and Apple will use Samsung's foundries and components.Bluetooth - Thursday, March 25, 2021 - link
Apple is using TSMC foundries and is buying up 25% of their capacity. They are first ones on the 5nm process node with A14 and M1. Samsung is providing many components but their foundries have fallen behind.Yojimbo - Thursday, March 25, 2021 - link
That's irrelevant. The point is that "Apple will use Samsung's foundries", as in they are willing to.FullmetalTitan - Thursday, March 25, 2021 - link
And do so currently, just not for M1 specifically. Every display driver IC Apple has used for the last 5 years was made by a Samsung fab.jakemonO - Thursday, March 25, 2021 - link
IBM isn't even a good software & services company anymore (excepting RedHat) I think they could leverage Power RISC IP and Intel expertise in manufacturing and power management to create a compelling alternative to ARM if they wanted to...zmatt - Wednesday, March 24, 2021 - link
POWER and Z don't directly compete with Xeon. They do similar things but the software is not easily migrated from one to the other. This is especially true of Z. If you are running Z systems you are probably leveraging a stack that has a multidecade lineage that stretches back to classic mainframes. While it is possible to engineer that to work on x86 in a Windows or *nix environment, its not trivial or necessarily cost effective to do so.So yeah you can move between them but its not a decision easily made such as replacing Xeon servers with Epyc for example.
Kevin G - Tuesday, March 23, 2021 - link
Depends on how you define competitor. IBM does compete in chip designs with POWER10 coming out this year. While IBM has been desperately trying to expand its user base for POWER, the majority are continued legacy customers.IBM used to be a chip manufacturing competitor and operated as a foundry also producing 3rd party designs. (The nVidia 6800GTX was manufactured at IBM.) As pointed out by others here, IBM sold off their manufacturing line to Global Foundries years ago. The fine print is that IBM never relinquished their patents nor halted their R&D wing. The theory was that IBM would continue to fill various patents and force some sort of partnership or worse, legal arbitration with the remaining foundries on various patents. That aspect seems to be coming true but with an interesting asterisk. If I had to guess, Intel is preparing to ditch silicon in some fashion with a strong possibility of going optical in some fashion. Between Intel and IBM there was a collection of patents that would encompass optical interconnects and motherboards. IBM may also gain access to Intel's fabs for their own products instead of relaying on Samsung currently.
Arsenica - Tuesday, March 23, 2021 - link
The last time Intel collaborated with IBM in semiconductors was in the ill-fated 450mm wafer project in Albany.My guess is that now IBM could collaborate with Intel in GAA-FET processes.
IBM may still be a competitor to some of Intel's products, but as IBM's main client for those same chips is the US government they would certainly prefer that IBM makes those products in a US foundry rather than at Samsung or TSMC.
ikjadoon - Tuesday, March 23, 2021 - link
The biggest potential news here is x86 licensing. There are zero details, so I have doubts whether Intel will do it fairly. Does Intel have enough faith in its architectures to open up its golden goose ISA?Intel is incredibly insecure & irrationally jealous of its x86 ISA, including when Gelsinger was a top engineer:
https://www.tomshardware.com/news/amd-intel-x86-cp...
One big mistake about Gelsinger's praises: he was in the top divisions at Intel when Intel was at its most anti-competitive and shameless. Gelsinger was at Intel from 1989 to 2009, then *the head* of the desktop & server CPU division (then called the "Digital Enterprise Group").
1989: Gelsinger joins Intel
1991: AMD sues Intel (Intel pays $10 million in damages + forced to issue AMD x86 licenses)
1997: DEC sues Intel (Intel forced to buy $700m DEC + forced into licensing at $1.5 billion)
1997: Intergraph sues Intel (Intel pays $225m settlement)
1999: FTC investigates Intel under the FTC Act (Section 5)
2005: AMD sues Intel (Intel pays $1.25b settlement)
2009: FTC sues Intel (Intel pays $10m settlement)
2009: Intel threatens to terminate AMD's x86 cross-license
2009: Gelsinger leaves Intel
Gelsinger was literally at his prime when Intel was at its worst.
Sources: https://en.wikipedia.org/wiki/Intel#Competition,_a...
https://www.ftc.gov/enforcement/cases-proceedings/...
ikjadoon - Tuesday, March 23, 2021 - link
Whoops, missed the DEC settlement source: https://www.nytimes.com/1997/10/28/business/intel-...Ironically, this 1997 lawsuit and Intel's absorption of DEC -> 2-3 individuals -> launch of the Apple M1 in 2021.
Dan Dobberpuhl (DEC -> PA Semi) -> Jim Keller (PA Semi -> Apple) -> Gerard Williams III (Arm -> Apple)
Raqia - Tuesday, March 23, 2021 - link
Most of their IDM customers will be doing things like cellular modems or hyper scaler in-house ARM server parts and not Intel x86 IP, so the days of their easy IP margins are over. They're pretty much also giving their customers the means to ditch their own high margin server parts too by enabling the competition. When their factories are finally done, so will TSMC's and Samsung's so this competition will further crush their once excellent margins.rahvin - Thursday, March 25, 2021 - link
They aren't going to have any IDM customers unless they can fix their fab process. Apparently people forget history, Intel's been offering fab services since 2016 but no one uses them because they are so far behind in process tech and they can't guarantee wafers because they can't even manufacture enough to meet their own internal demand.I get they want to offer fab services but honestly until they can prove they can advance reliably past 14/12nm they aren't going to have real customers. Anyone willing to use 14nm has ample choices that will be cheaper with guaranteed wafers, anyone needing 10/7nm is going to either use TSMC or Samsung as they are the only two fabs with working processes ATM.
The only reason I can think that Intel made this announcement is for Stock reasons as there is little to no chance of this being a reality until they can demonstrate an ability to fab something at smaller than 14/12nm because they simply don't even have the capacity to fab their own processors let alone outside companies.
TerberculosisRobocop - Thursday, March 25, 2021 - link
All of us Arsians and anyone paying attention know this will not have a big impact now.However, it makes me happy to know that The new intel leadership has chosen a direction and is moving.
I have been gaming Intel stock price for a few months now, and it is marvelous to know that current leadership has a plan and direction that EXPANDS intel's possibilities.
So, even if it is just an announcement for shareholders, it’s the correct announcement for those of us who have been betting that Intel can come out of this period of mode stagnation as a stronger company.
Also: I feel like it’s a big deal to keep a decent volume of semiconductor manufacturing on US soil. We are a shitty country with a shitty, violent, aggressive history. It’s best we make sure it’s hard for anyone to try to make us pay for all the awful.
buufball - Sunday, March 28, 2021 - link
Violent and agressive? Get out of here with your anti american rhetoric, you trash. Only reason you have a pc to type on is because of the US. Did you never read of the horrors of china or russia? Imbecile.Raqia - Tuesday, March 23, 2021 - link
Another point: licensing x86 will more quickly crush their margins by eliminating the small moat they have left in binary compatibility with tons of legacy software. I think they won't be doing any licensing and will jealously guard against the onslaught of ARM servers using emulation for x86 compatibility.TheinsanegamerN - Wednesday, March 24, 2021 - link
The *onslaught*?We've been hearing about "muh ARM servers" for a decade now. I was hearing these things in high school, how ARM was not laden with legacy code and how efficient it was and how it would destroy intel in the server market.
This was 2013.
Today its 2021, and I'm hearing "ARM is not laden with legacy code and how efficient it is and how it will destroy intel in the server market". Reminds me of fusion reactors.
Otritus - Wednesday, March 24, 2021 - link
Today ARM servers are being produced and shipped. ARM processors deliver some of the best multi-threaded performance money can buy. x86 only survives because many current processes were designed only for x86. ARM servers will be coming en masse to overthrow x86 in the coming decade because they already are.rahvin - Thursday, March 25, 2021 - link
Ah baloney. The only ARM server processors even available only fit a certain niche. ARM has made incremental improvements with each generation, but AWS's Graviton was the first viable server processors and we have no idea what kind of economics are involved because AWS could be choosing to loose money on the processor to gain market-share and experience.The fact is you can't buy a ARM server right now that will be competitive with x86 in every category. Yes there are a couple options that work in certain situations but the TCO is still out of wack and IMO non-viable against x86, particularly with the strides AMD has made with Zen.
If servers still relied only on Intel I have no doubt ARM would be a much more viable competitor but that's simply not the case.
lmcd - Monday, March 29, 2021 - link
Licensed Atom Cores will crush low-end ARM, and high-end ARM hasn't actually proven to be efficient.nfineon - Thursday, March 25, 2021 - link
This is very good info to keep in mind thanks for posting the timelines. Gelsinger would ultimately have to sign off on the terrible pricing of the 11th gen high end sku's as well where at the top end Intel are trying to price an 8 core part with worse performance, thermals, and power against AMD's 12 core part which also has a slightly lower MSRP. The ultimate decision to push that absurd pricing must be signed off from the top.Bluetooth - Thursday, March 25, 2021 - link
Who would want to license x86 for future CPU, when ARM is available with better ISA. Is backwards compatibility in old software that important? Or would the licence just the x86-64 commands (which I think AMD has something to say in). Wouldn't someone licensing x86 get rid of the old 32 bit compatibility and just support 64 bit?rahvin - Thursday, March 25, 2021 - link
There is no ISA advantage for ARM. It's a BS talking point that has no basis in reality. x86 processors haven't used x86 internally in more than 20 years. Every x86 processor uses an internal flexible RISC core and tacks a 128k transistor translator on to handle the conversion from x86 to the internal language. That translator was a big part of the processor 20 years ago, today it's not even 0.01% of the die.If anything, the x86 translator actually gives the advantage to x86 because every single x86 processor can have a custom designed RISC core that is optimized for each process node.
proflogic - Friday, March 26, 2021 - link
That is a rather disingenuous argument. One of the ISA's biggest impacts is on instruction decoding, and you're simply ignoring it by waving away all the stages that happen before it becomes RISC-like internally. The variable-length instructions in x86 make it exceedingly difficult to avoid longer pipelines for decoding or to achieve a greater issue rate at high frequency. You blow through power and area searching for the ends of instructions and doing that translation, and the cost of any pipeline flush ends up being higher.One mitigating factor may be that code size (and cache pressure) is reduced there, but I'm fairly certain various papers have argued those benefits are pretty small against ISAs with a compressed format for common instructions.
Shorty_ - Tuesday, March 30, 2021 - link
this is precisely how it was explained to me by someone who built processors for a living.quiksilvr - Tuesday, March 23, 2021 - link
Phoenix has the largest nuclear power plant in the country, it does not snow there and land is relatively cheap for being the fifth largest city in the country. It is the perfect place for fabs.FreckledTrout - Tuesday, March 23, 2021 - link
As long as there are not to many to overwhelm water needs.JCB994 - Wednesday, March 24, 2021 - link
Last time I was there most of Arizona is sitting on a huge water aquifer. I could not believe paper plants were in the desert south of Phoenix but it is due to the water.sing_electric - Wednesday, March 24, 2021 - link
Regardless of the water issues, it seems unnecessarily risky to concentrate fabs in one area. Even if you're not worried about the national security concerns, there's the possibility of man-made or natural disasters that could wipe out production. If serious damage occurs, it could be a year+ before they're back online given the waiting list for key equipment.tomatotree - Wednesday, March 24, 2021 - link
That's already true to a large extent today with how many fabs are concentrated in Taiwan. And in the US, not all of the capacity is in Arizona, even after Intel's expansion -- there's a big manufacturing presence in Oregon as well. There's some risk in geographical concentration, but there's some advantages as well: it's much easier to hire good employees if you're not the only company in the area doing that type of work. And no matter where it is, a natural disaster hitting a leading-edge node foundry will affect worldwide supply -- these things are too expensive for true redundancy.Kougar - Tuesday, March 23, 2021 - link
Have to confess, the prospect of US-fabs with TSMC or larger levels of wafer capacity has quite the allure. The more dominant AMD becomes, the more fab capacity Intel would have to offer.Currently Intel fabs nearly as much silicon at TSMC as AMD does... which is mind-boggling to consider when AMD serves two console launches, plus mobile+desktop+server CPUs, and then GPUs on the side. It appears that Intel bringing more of its fabrication in house would open up almost double AMD's current wafer allotment capacity at TSMC, and AMD would certainly could put most of that to use.
Kevin G - Tuesday, March 23, 2021 - link
Intel has a lot more to manufacture in house now than it ever had before. The Altera acquisition alone shifts a good chunk of what was expected to be built at TSMC to Intel's own in-house foundries. Due to the delays of 10 nm and 7 nm, only now is that really starting to happen.That is also only one of Intel's recent acquisitions. There are numerous other companies Intel purchased who still have outstanding orders with TSMC that will also shift toward Intel's own fabs over time.
ilt24 - Tuesday, March 23, 2021 - link
Kougar-->"It appears that Intel bringing more of its fabrication in house would open up almost double AMD's current wafer allotment capacity at TSMC"Not as much as that, a good amount of what TSMC makes for Intel are chipsets and network chips which are not made with leading edge processes. Yes Intel will probably move Altera and Mobileye stuff in house, but I don't see that happening until 2024 and also starting later this year Intel will be increasing their leading edge wafers starts at TSMC as begin to make GPUs and next year CPUs there.
Kougar - Wednesday, March 24, 2021 - link
Aye, but AMD doesn't use leading edge TSMC nodes either, Apple gets dibs on those. Then there's the IO hub die which are still 12nm GloFo. AMD needs a very large supply of those particularly if we're speaking in terms of physical size.Zen 4 / Genoa is confirmed to be using a new IO die thanks in part to the power issue with Milan's IO chips as well as the changes required for DDR5. I've not seen anyone mention what node or fab AMD will use yet, but as a random guess it seems unlikely they could source enough 7nm TSMC for it. But it won't stay on 12nm GloFo, Milan's IO die power consumption will make sure of that much.
Rudde - Wednesday, March 24, 2021 - link
Isn't Genoa compute chiplets confirmed to be manufactured on 5nm? That would free enough capacity of 7nm for 7nm IO chiplets, wouldn't it? TSMC has also been expanding their 7nm capacity.Kougar - Wednesday, March 24, 2021 - link
Yes, 5nm Zen 4 is confirmed. And theoretically yes, it should... but so many things use 7nm TSMC that we won't know until it's confirmed. AMD may stick with a larger, cheaper node at TSMC or Samsung.adelio - Wednesday, March 24, 2021 - link
Are Global founderies not going to eventually move to a smaller process?Kougar - Wednesday, March 24, 2021 - link
They probably will far into the future when it's cheap to do so, EUV machines from ASML are already backordered a full year now. But with Intel, TSMC, and Samsung each planning new fabs that backorder is going to get worse before it gets better.In any case, there's nothing GloFo can do right now. AMD needs a big supply of shrunk IO chips no later than Q4
tomatotree - Wednesday, March 24, 2021 - link
No: https://www.anandtech.com/show/13277/globalfoundri...Kougar - Wednesday, March 24, 2021 - link
I miss my edit button.... For example Xilinx chips are 16nm TSMC. The Xbox One is TSMC 16FF+, as another example. Microsoft likely preplanned it that way with a die shrink at some date in the future. Probably others as well I can't think of.tomatotree - Wednesday, March 24, 2021 - link
Intel actually started porting the Altera stuff to 10nm internal back in 2019 (https://www.anandtech.com/show/14149/intel-agilex-... and if this Mouser listing is to be believed, the devkit will be shipping in about a month (https://www.mouser.com/ProductDetail/Intel-Altera/...JayNor - Wednesday, March 24, 2021 - link
Intel recently announced a DARPA partnership where they'll use Intel 10nm for eASIC chips, rather than TSM 16nm. I wonder if that program will also be planned to use the new foundry.adelio - Wednesday, March 24, 2021 - link
What i want to see is more fab plants and also spread throughtout the world. American EU etc.Too many eggs in one basket is NOT a good thing.
Any country that relies too much on foreign goods without some local balance, especially in Important field like fab plants etc leaves itself a hostage to fortune.
The UK has hardly ANY Steel production left, most of it is imported and that leaves us vulnerable.
tomatotree - Wednesday, March 24, 2021 - link
Keep in mind that Intel's TSMC manufacturing use today may not be entirely leading-node processes. Probably a fair amount of that volume is for things like networking chips, FPGAs, PCHs, etc., which don't always need to be on the most advanced process the same way CPUs and GPUs do.JayNor - Wednesday, March 24, 2021 - link
Intel has stated the eyeq5 chips are TSM 7nm. I believe either Intel or Habana announced a next generation of their training chip will be on TSM 7nm.Kevin G - Tuesday, March 23, 2021 - link
The presumption from the article is that Meteor Lake will be leveraging an IO die announced at a later date. While it is clear that the IO die hasn't been formally announced, it could actually release prior to Meteor Lake: it could appear as a companion to a 2022 chip using 10 nm ESF using similar packaging techniques. Intel's 2023 move would be to slip in the 7 nm tile while heavily reusing other existing tile designs. That's the real benefit of chiplets is being able to recycle parts between generationsRanFodar - Tuesday, March 23, 2021 - link
There are actually more opportunities in Pat Gelsigner's strategy. SiFive, a startup, said that it will work with Intel to make the RISC-V designs availabe to customers of Intel's Foundry Services.https://www.zdnet.com/article/intels-tilt-to-found...
Duncan Macdonald - Tuesday, March 23, 2021 - link
So if Intel can meet its predictions (unlikely!!) they will have 7nm CPUs about the same time as AMD CPUs are on 3nm.Given the backorders on EUV equipment from ASML, Intel could easily find that its new fabs are unable to produce 7nm products until well after their proposed timeline.
There is also a big question as to how well an Intel chiplet design will work - a lot of the current IP will have to be completely replaced (eg the ring buses). As with AMD any multiple CPU chiplet design will have much longer delays than current Intel designs for off chiplet CPU to CPU communication.
drothgery - Tuesday, March 23, 2021 - link
AMD won't be on TSMC 3 nm until at least a year after Apple, and probably closer to two years if 5nm is a reasonable predictor.Matthias B V - Wednesday, March 24, 2021 - link
Apple will be on 3nm by 2022 and then use it for their A and M APUs and dGPUs. AMD will have to wait a while for TSMC3nm until TSMC ramps up lots of capacity or Apple moves to TSMC 2nm.AMD will probably 2nd source TSMCs 5P with SAMSUNG 3nm GAAP on GPUs etc. but wont switch to TSMC3/2nm anytime soon and only have CDNA2 on 5 nm in 2021 and follows up with other 5nm designs in Q3/Q4 2022...
Intels 7nm SuperFin is probably comparable to TSMC 5P in density - nit sure avout other metrics.
Bagheera - Thursday, March 25, 2021 - link
transistor density projections:https://semiwiki.com/semiconductor-manufacturers/i...
outsideloop - Tuesday, March 23, 2021 - link
This is a bunch of talk. It's been going on for years now. Just look one year ago when Intel was guaranteeing ice lake server by 4qtr 2020.Intel is now four years behind AMDs chiplet design. Even with perfect execution from here on, the idea that they will be able to reverse, much less stop, AMD continuing to gain market share and revenue, is a fools dream.
JayNor - Wednesday, March 24, 2021 - link
Intel Sapphire Rapids already sampling with PCIE5, CXL, DDR5. Where are AMD's zen4 samples?Bagheera - Thursday, March 25, 2021 - link
Intel makes a big deal about 300K Ice Lake shipped, and conveniently forgets to mention Ice Lake SP has been "sampling" since 2018.mdriftmeyer - Thursday, March 25, 2021 - link
So trueoutsideloop - Tuesday, March 23, 2021 - link
Pat Gelsinger's "dreams" are no match for AMD and TSMCs design and fab execution.eva02langley - Wednesday, March 24, 2021 - link
I think the same. Intel will not be able to buy their way out of this mess this time. Anti-competitive behaviors will not fly this time around. They can't win against TSMC, it is over, they are the industry leaders for the next decade.TristanSDX - Tuesday, March 23, 2021 - link
Intel as foundry ? Maybe only for trailing edge process where margins are low, and they will be competing with plenty of capacity at TSMC / Samsung / SMIC and others, so right amount of clients won't be guaranted.Their 7nm is 'simplified', Simplifed process means that they can't afford for 'full node progress', and must rely on 'half node progress', just like TSMC and Samsung, but TSMC is two years ahead of Intel. That's why they incorporated IDM 2.0, and design CPU in two variants - highest SKU (i9, i7, i5) for TSMC, while other slower SKU for their trailing fabs. This is proof that they won't get manufacturing leadership in future.
COtech - Tuesday, March 23, 2021 - link
I can see intel totally competitive as a foundry. The fabless companies would actually benefit the most by having a third cutting edge node option.Intel is recognizing here that TSMC and Samsung are as much their competitor as AMD, NVidia and Apple.
At this point with ASML only able to ship 50 EUV tools a year then any that intel purchase is capacity that Samsung or TSMC can't add. It gets a double benefit from increasing capacity at the high end and the same with purchasing production at other fabs. Demand for semiconductor products is in an exponential growth phase and no one will be able to satisfy it.
Matthias B V - Wednesday, March 24, 2021 - link
I agree:Biggest issue for INTEL and SAMSUNG currently might be EUV capacity and ASML supply. That's why they cannot supply high volume EUV nodes as they lack EUV capacity and their last DUV nodes are 8nm Samsung and 10nm Enhanced SuperFin for Intel.
Since supply is tight and even if they could retooling all their DUV fabs would be an economic dissaster it makes it more than reasonable to use 22, 14 and 10nm for IoT, Automotive and controllers. Those nodes are enough for lots of applications there...
At current market and even in 2022 it probably doesn't matter which node is ompetitive as they can sell anything right now! So they just make crap to gold now and focus on 2023/2024 ;)
Sub31 - Wednesday, March 24, 2021 - link
EUV pellicles are nearly ready for commercialisation. Much more competitive defect rates once these are introduced, probably, and maybe EUV equipment will become less complex after that.tomatotree - Wednesday, March 24, 2021 - link
I read 'simplified' here as meaning that they replaced many of the complex multi-exposure non-EUV steps with simpler single-exposure EUV steps.JayNor - Wednesday, March 24, 2021 - link
AMDJayNor - Wednesday, March 24, 2021 - link
new keyboard ... amd suggested for a couple of yrs that they were moving to 7N+ for Milan. Reviewers took that to mean the EUV version of 7N.I'd guess tsm shares EUV equipment for 5nm, and will do the same for 3nm, so it isn't clear to me that apple moving to 3nm will free up that equipment for amd.
Bagheera - Thursday, March 25, 2021 - link
we get that you love Intel. please stop talking like you understand silicon fabrication.zodiacfml - Tuesday, March 23, 2021 - link
Meh. It appears that Intel is definitely no match to TSMC, hoping for more availability or cheaper ASML equipment by 2024. I reckon, the AMD vs Intel narrative will remain the same for several years.AnGe85 - Wednesday, March 24, 2021 - link
Nop, quite the contrary 3nm-scanners become much more expensive than the current gen.Teckk - Tuesday, March 23, 2021 - link
Must be a hard pill right now for all those that dissed AMD for going the chiplet way with Zen.Once Intel gets it manufacturing fixed it’ll be good competition.
eva02langley - Wednesday, March 24, 2021 - link
They are 5 years behind AMD when it comes to Infinity Fabric and Chiplet designs. With AMD relentless execution, Intel will never catch up unless AMD screw up big time.JayNor - Wednesday, March 24, 2021 - link
amd current infinity fabric won't handle pcie5 bandwidth.supdawgwtfd - Wednesday, March 24, 2021 - link
Oh of course. IF will stay the same and get behind...You seriously believe that AMD will just leave IF as is?
Qasar - Wednesday, March 24, 2021 - link
supdawgwtfdof course he does, you havent noticed he praises intel, while bashes amd in most of his posts ? to say that AMD's current IF wont handle PCIe 5 bandwith, is like saying Intels DMI wont handle PCIe 7 bandwidth. to say amd OR intel wont update their respective interconnects as PCIe bandwidth also increases, is well, just plain stupid.
Bagheera - Thursday, March 25, 2021 - link
Intel next gen RKL can't even beat Zen 3.You were saying?
SkyBill40 - Wednesday, March 24, 2021 - link
The real question? Where in the hell do they plan to build it in Chandler? Space is already at a premium and there's next to no land left at the Ocotillo campus, which is 10 minutes from my front doorstep. Unless they're planning on building on what little space remains and what may not be enough for another fab the size of F42, their options are truly limited.I suppose we'll see... in five years or so.
dsplover - Wednesday, March 24, 2021 - link
Nice to know they’re back in the game. 2022 will be a great year, this year is just more cores, more Watts, more 14nm.Thanks for smacking Intel around AMD.
arashi - Wednesday, March 24, 2021 - link
Your house is about to be "accident"-ed.Condolences.
SkyBill40 - Thursday, March 25, 2021 - link
I don't live "that" close... but close enough.ilt24 - Wednesday, March 24, 2021 - link
Below is a google map link to the Ocotillo site, looking at it I think they still have room on the existing land, especially if the build a parking garage.https://www.google.com/maps/search/intel,+ocotillo...
tomatotree - Wednesday, March 24, 2021 - link
Yep, you can see a big dirt lot behind the current fab in one of the pictures Intel released as well: https://newsroom.intel.com/news-releases/idm-manuf...SkyBill40 - Wednesday, March 24, 2021 - link
The dirt to the west (left) of the Fab 42 location is Native American land. They're not giving that up. They'd have to build south of Chaparral Way (where the green land is) if they even own it or to the spot currently occupied by some existing structure immediately south of F42.waterdog - Thursday, March 25, 2021 - link
Intel owns the land south of their current facilities down to Chandler Heights Road, including the irrigated fields.konbala - Wednesday, March 24, 2021 - link
Hope it won't end up like a Meteor.Silver5urfer - Wednesday, March 24, 2021 - link
All smoke and mirrors. They killed real product Optane 3D X point. They sold mobile 5G to Apple. Got booted off by Apple.Lost all the Fab lead to TSMC long back. No HEDT successor since 2 years. Mainstream RKL is DOA except for Gen 4 platform with DMI upgrade at high cost of Power consumption and lack of Core scaling upgrades. Lost to EPYC Rome and now Milan 7003 CPUs.
TSMC and Samsung now leading. Okay maybe compete with GloFo for the Semi industry for multiple contracts and all but their biggest news of IP licensing is shady and not clear at all. Why will Intel sell their top x86 IP to some corporate for SoC ? Microsoft is already with AMD 7N custom SQ1 chip. Who want to get Intel lost lead uArch and their failed 10nm ? On top non comparable to ARM since Intel does their own business with selling x86 parts while ARM doesn't. That's a major thing. This is direct competition. How does it even work ? EMIB failed with Kaby G. Foveros ? Its great but AMD is not dumb neither is Apple nor is Ngreedia. Nor is Samsung lmao everyone is having their own direction.
FFS get the damn 10nmSF properly beat Zen 3 and upcoming Zen 4 (but with ADL having 8 only x86 big cores and small pathetic Non HT Atom cores) how is it possible. Ice Lake Xeon already looking outdated vs 3PYC Milan.
Why don't they STFU and prove it with a strong product ? Damn. All I wish is x86 be healthy and competitve thanks to AMD. For that since this uArch puts people in power rather than corpos in power (ARM, highly custom and need more OEM help - Qcomm CAF, Blobs and etc for Android CTS / Updates plus performance improvements vs x86 stability back compat AND DIY scene with PC market.
JfromImaginstuff - Wednesday, March 24, 2021 - link
Intel was in a pretty similar position in 2005, I think. But yea, hopefully they do pull themselves out and compete.TheinsanegamerN - Wednesday, March 24, 2021 - link
Intel in 2005 had an advantage though: While their american engineers were farting around with netburst, their israeli team had been refining the pentium III core into the pentium M, which was doing quite well. That eventually became core 2.Today, in the mobile space intel does have sunny cove CPUs, but they are restricted by the 10nm node, just like their desktop parts. After 2 years we still only have quad core parts, with promises of an 8 core part.
It all coems down to fabs, unlike in 2005 where intel just had to dump a bad design, intel now has to overcome a manufacturing challenge that they have spent 7 years on with little apparent success. I imagine it will take at least another 3-4 years before either their 10nm or 7nm pan out any significant improvements.
So this is closer to intel in 2002, hurting from AMD smacking them around and about to get another handful with athlon64, I mean, zen 4.
GeoffreyA - Wednesday, March 24, 2021 - link
"their israeli team had been refining the pentium III core into the pentium M"I think credit needs to be given to Intel Israel more often. Usually, we think of Intel as one abstract entity, and those engineers are forgotten. Besides Pentium MMX and M, they designed Sandy Bridge, Skylake, and Sunny Cove. Perhaps Core/Conroe too. Anyhow, Conroe was based strongly on the Pentium M. Quite likely, the successor to Sunny Cove is on their drawing boards as we speak, if it hasn't been taped out already.
drothgery - Wednesday, March 24, 2021 - link
Since Golden Cove-based Adler Lake is supposed to ship this year (Willow Cove-based Tiger Lake already replaced Sunny Cove-based Ice Lake for U-series chips, but Sunny Cove will still be in Ice Lake SP server CPUs; Sapphire Rapids is Willow Cove), I think it's safe to say it's taped out already. Ocean Cove (the successor to Golden Cove) is supposed to tape out later this year and launch in Meteor Lake CPUs Q1 2023.JayNor - Wednesday, March 24, 2021 - link
sapphire rapids is golden cove.GeoffreyA - Wednesday, March 24, 2021 - link
You're right, but when I wrote that, I meant Sunny as the latest microarchitecture, perhaps wrongly, because all Willow did was alter the caches, as far as I know. Concerning Golden Cove, is it just a microwaving of Sunny/Willow or is it a proper revision of the architecture? Can't seem to find much data on it.Otritus - Thursday, March 25, 2021 - link
Golden Cove is a proper revision of Sunny CoveJayNor - Wednesday, March 24, 2021 - link
intel statement is that Optane plans are not changed by Micron's exit. Micron will complete their supply obligations for gen2. Intel and Micron were already split on gen3 and beyond.eastcoast_pete - Wednesday, March 24, 2021 - link
I guess this is "The Empire Strikes Back", or, at least, announces that it intends to do so (see fine print for disclaimers and details).On a more serious note, took them long enough. Intel's strength for the last two decades came also from being better at fabbing chips than the competition, and without that, their competitive edge has dulled a lot.
EthiaW - Wednesday, March 24, 2021 - link
How can intel persuade any clientele to use its fab if itself is outsourcing the high-end products to TSMC?tomatotree - Wednesday, March 24, 2021 - link
Probably pretty easily if TSMC doesn't have any spare capacity, as has frequently been the case recently. Especially if it's also cheaper (which remains to be seen).JayNor - Wednesday, March 24, 2021 - link
about half of tsm volume is 16nm and older.EthiaW - Thursday, March 25, 2021 - link
TSMC has being holding the mature nodes market firmly before 2010, when it was far less competitive and sophiticated than Samsung or Intel as for advanced ones. Mature nodes are the foundation of the TSMC empire we know today, I don't think intel can have an edge over them or even GF in this market.Holliday75 - Thursday, March 25, 2021 - link
Intel's edge. "Hey guys we have capacity!"That is about all you need right now.
Bagheera - Thursday, March 25, 2021 - link
You think Intel only pumps out leading nodes?six_tymes - Wednesday, March 24, 2021 - link
this CEO gets it. He is forward thinking. We need competition, and it seems as if Intel is getting back on track to bring it.Hifihedgehog - Wednesday, March 24, 2021 - link
Sure, he gets it. The problem remains if they can get their foundry problems in order enough to close the gap of the moving target and then maintain cadence with their arch-rival who is led by an equally brilliant CEO.eva02langley - Wednesday, March 24, 2021 - link
He can't compare to what Lisa did. She is literally... THE SAVIOR... of AMD. She didn't just turned around a company on the verge of bankruptcy, she literally turn it around to the industry standard.eva02langley - Wednesday, March 24, 2021 - link
He get it, but will not be able to change ANYTHING... TSMC and AMD will continue their relentless execution which will force Intel to difficult choices...Keeping their current fab strategy is a huge mistake. Basically, they are going to be the dog for now on.
Hifihedgehog - Wednesday, March 24, 2021 - link
Exactly. Intel might have a world-class CEO now at the helm, but this may be a case of too little, too late. AMD was able to get ahead not of Intel only because they had an amazing CEO, but because Intel had lousy management who meanwhile squandered their time away. Think of this in terms of a timeline of getting two world-class triathletes. AMD got theirs when Intel had a sluggish sloth in the race and they darted into the lead and kept running faster and faster miles into the distance. Intel finally has gotten their star player but AMD still has theirs, and she has gotten her team a good 10 to 20-mile lead and is not letting up. I am not counting out anything here because the unexpected has happened so many times throughout the history of the tech industry. However, all Gelsinger may end up doing is preventing Intel from sinking anymore than actually ever restoring market leadership, but that is only one possibility of many. Heck, Sega used to have the world's best gaming graphics a whole five years ahead of the PC when they released Daytona USA to arcades in 1993. But by the 2000s, Sega's console business was deader than a doornail and they were using embedded x86 systems that were a bit behind the best in PC graphics. We just never know what might happen with how volatile the tech industry is.JayNor - Wednesday, March 24, 2021 - link
yet intel persevered and now has 10nm fabs, while AMD gave up.supdawgwtfd - Wednesday, March 24, 2021 - link
Relevance?You could also say that Intel is giving up as well as they are outsourcing some of their chips.
Hifihedgehog - Wednesday, March 24, 2021 - link
And TSMC will have 3nm by 2023, and Intel will be utilizing their nodes. That's a slap in the face to their manufacturing leadership. It proves that they are no longer the world leader in silicon manufacturing. That crown goes to TSMC.Bagheera - Thursday, March 25, 2021 - link
more like AMD persevered despite being on the verge of bankruptcy.Intel was always flush with cash yet they couldn't bring 10nm to market in 2016 out their original roadmap. let's not forget their 14nm was delayed twice before that.
Not even gonna talk about the 7nm.
eva02langley - Friday, March 26, 2021 - link
Intel fabs are what dragging them behind, not the contrary. The only reason why they stick to their fabs is because they can't get that kind of capacity on a better process. If they were going for the better product, they would lose half of their sales and cut their margins since they need to pay for taping out their chips instead of doing it in-house.The only reason why Intel CEO is now going into the fab business is so they can buy waffer at TSMC and maintain their fab business... it is a pure admission of failure.
adelio - Wednesday, March 24, 2021 - link
After the debable of texas having to cut power to their chip plants do these companies have to invest more in standby power. Can Arizona handle the peak loads that may come with all these new plants.America is renowed for failing to plan ahead. Rather pay dividends than invest in infrastructure.
Leeea - Wednesday, March 24, 2021 - link
That is Texas, which is a pathetic joke of a state. Texas choose to not be part of the federal electrical grid because it did not want to comply with "onerous" regulations associated with cold weather reliability.Arizona is part of the federal electrical grid and did not and will not suffer the problems Texas had.
TheinsanegamerN - Wednesday, March 24, 2021 - link
Because freak weather effects never shut down power ins tates like california, right? Teh rest of the US never has freak weather and power outages, no siree, its all perfectly fine, ONLY texas has weather outages.ilt24 - Wednesday, March 24, 2021 - link
The Texas power outage was a replay of what happened in 2011. After the 2011 outage it was identified that the cause of much of the issues was lack of winterization. The Texas power industry decided it wasn't worth the cost of winterizing the system and since they were unregulated they didn't have to.Some 70 people died and millions were without some combination of heat, water and power for...but per the likes of Rick Perry..“Texans would be without electricity for longer than three days to keep the federal government out of their business,” which he was saying at the same time Governor Abbott complaining Binden hadn't approved enough federal disaster relief.
RBFL - Wednesday, March 24, 2021 - link
Each day of the outages they were billing their domestic customers at the same rate as they get in a year, ~$10B. ERCOT kept the price too high for too long which resulted in an overbilling of $16B, but whatever we can't fix that after the event.Many wind providers will go bust, as they had to buy power at the high rates to cover the intermittency clauses in their supply contracts.
Yet the carbon based producers and the fuel suppliers that profited hugely as a result of their inability to operate have no sanctions, only giant C suite bonuses.
Ain't lack of regulation great, if you are a member of the good old boys club.
Bagheera - Thursday, March 25, 2021 - link
the relative lack of idiots here compared to sites like WCCFTech is so refreshing.JayNor - Wednesday, March 24, 2021 - link
One other bit of news from the presentation...Alder Lake desktop chips will be launched before the mobile chips.
I suppose the CES demo in a desktop hinted this, but now Intel has announced it.
Qasar - Wednesday, March 24, 2021 - link
il believe anything intel says, when it actually happens, or ships in volume.Bagheera - Thursday, March 25, 2021 - link
Intel announces a lot of things. Doesn't change the fact that:- 14nm delayed by 1 year.
- 10nm delayed 2 years (was due 2017), though more like 4 years since 10nm still not mature enough for desktop.
- 7nm delayed 2 years (was due 2021, projected 2023)
When it delivers I'll believe it.
GeoffreyA - Wednesday, March 24, 2021 - link
A bit off topic, but this being an article on foundries, if I may, I'd like to ask how exactly do these processes even work at <= 5 nm, where I believe quantum tunnelling starts to get out of hand. Have they mitigated it, using special techniques? Or are they just working round that? I suppose physics is showing that Intel, AMD, and the rest will have to switch to something else soon.Bagheera - Thursday, March 25, 2021 - link
this is a good readhttps://semiwiki.com/semiconductor-manufacturers/i...
GeoffreyA - Thursday, March 25, 2021 - link
Thanks! I will read it.Oxford Guy - Wednesday, March 24, 2021 - link
I guess the pretense of journalistic neutrality has gone out the window.‘to work on exciting new projects”
“raring to go”
Superheroes and supermonopolies (ASML) make me rather queasy.
GeoffreyA - Thursday, March 25, 2021 - link
"raring to go"As a Baldur's Gate fan, this reminds me of Minsc's phrase, something something and raring to go. Minsc and Boo certainly had the best lines.
eva02langley - Wednesday, March 24, 2021 - link
Intel will lose against TSMC. They will never beat their nodes. TSMC will be on 3nm in 2023... and let me remind you that 7nm is garbage. Bob Swan said the following...https://www.tomshardware.com/news/intel-announces-...
For perspective, rival foundry TSMC plans to be on the 3nm node in the same time frame as Intel's new schedule for 7nm. Intel clearly isn't pleased with its execution on the 7nm node, as an embattled Swan remarked that "And we feel pretty good about where we are, though we’re not happy. I’m not pleased with our 7nm process performance" at the end of the call after a bruising question and answer session with analysts. Swan also said "we have root-caused the [7nm] issue and believe there are no fundamental roadblocks," and that the company would provide further updates at an upcoming Architecture Day.
JayNor - Wednesday, March 24, 2021 - link
that quote was from July 2020. They did six months of experiments and Bob Swan declared the problem resolved in the q4 earnings call.eva02langley - Friday, March 26, 2021 - link
Like 10nm is resolved? No but seriously, are you for real? You are doing Intel's Damage Control? I hope you are paid for that. /sarcasmWaltC - Wednesday, March 24, 2021 - link
Same old Intel...what the company needs very, very badly are new architecture x86 CPUs, but of course Intel is throwing tons of money out to see what happens--to see what the proverbial, distant, 7nm cat drags in. Intel's idea of progress is doubling, tripling its immediate overhead with seemingly no recognition of the fact that by the time Intel brings its FABs to parity with where TSMC is right now, 7nm, TSMC will be heavily into 5nm if not 3nm, and Intel will still be far behind. Typical Intel--does not know how to do much more than throw good money after bad (Gelsinger remembers Itanium well, I'm sure) and to make stupid assumptions about its invulnerability and longevity. AMD would simply not exist today if it was crippled by the same fiscal imperatives that drive Intel. AMD cannot afford to make the mistakes Intel has made and continues to make--which is probably exactly why Intel has dropped behind AMD to the massive extent that it has today. Bringing old dudes out of retirement to relive Intel's halcyon years is about as desperate as it gets, imo.WaltC - Wednesday, March 24, 2021 - link
Well, I do have something positive to say about Gelsinger, unlike Biden, I don't think he's senile...;)Hifihedgehog - Wednesday, March 24, 2021 - link
He's also a very honorable person unlike 99% of the scummy folks I have seen running these companies. He is the last person that would be having an extramarital affair. He also is a donor to many humanitarian agencies, not a corrupt money-laundering, self-serving foundation like you see Bill Gates and others in Silicon Valley using (e.g. Gates donates and invests tax-free money to the Gates Foundation, who, in turn, purchases Microsoft products and services to use to provide "free" charitable services to impoverished nations).GeoffreyA - Thursday, March 25, 2021 - link
I take off my hat to Mr. Gelsinger.Hifihedgehog - Thursday, March 25, 2021 - link
Check out these articles. He’s a gem.https://www.crn.com/news/virtualization/vmware-s-p...
One of the countless things he recently did was climb Kilimanjaro to the summit while donating to build a high school in Kenya:
https://www.crn.com/news/channel-programs/30010584...
GeoffreyA - Thursday, March 25, 2021 - link
Thank you. I just read them now, and am beginning to like Pat even more. What a person. I just hope that some of his values can filter down through the rest of Intel.DigitalFreak - Wednesday, March 24, 2021 - link
'with Intel extending cooperation with the State of Arizona, as well as the current administration’s target of improving semiconductor manufacturing inside the country.'Unless these fabs are going to be used to manufacture ARM CPUs or other non-x86 semiconductors, I don't see it making much of an impact on the need to rely on China / Taiwan.
tomatotree - Thursday, March 25, 2021 - link
They will be. Part of the announcement is that they're now open to building ARM, RISC-V, and other non-x86 processor designs in their foundries.Machinus - Wednesday, March 24, 2021 - link
One day Intel will innovate to 13nmBagheera - Thursday, March 25, 2021 - link
I'll take that. progress is progress lolsix_tymes - Thursday, March 25, 2021 - link
a ceo that's actually doing work, and knows what he's doing. congrats to Intel.Ranari - Thursday, March 25, 2021 - link
$20B into new fabs, with Samsung considering as well. Translation: We expect China to take over Taiwan (i.e.,TSMC), and we're going to take the business.Oxford Guy - Saturday, March 27, 2021 - link
For that gambit to work it would have to have an answer to the ASML monopoly, eh?Do you think China has managed to get its hands on all the ASML special sauce or will be able to?
Eliadbu - Sunday, March 28, 2021 - link
EUV scanners are very complex machines with advanced technologies and parts that are difficult to manufacture do they can deal with the unfavorable physics of EUV. China has very low chance to develop this kind of machine on their own, the best they can do is try to steal the technology and even then making working machine will be difficult task.Oxford Guy - Wednesday, March 31, 2021 - link
So, the answer is: Yes... China will become equivalent to ASML at some point.What we don't know is how long that will be.
This means the US and the EU need to change the situation by investing more to keep the edge. In fact, the US both have already lost the foundry edge and the US doesn't have an ASML.
TheJian - Friday, March 26, 2021 - link
LOL. Deleting my posts now...comic. NO swear words, just two links posted. I guess AMD portal site still...You guys hate Patrick Moorhead or something? Posting an article from competition blocked or what? Patrick worked at AMD for a decade...NO respect? pfft..Qasar - Friday, March 26, 2021 - link
awe.. did i miss an anti amd rant by the jian ? ahh will, probably just fluff, bs, and fud anywayeva02langley - Friday, March 26, 2021 - link
Leo nailed it... sorry intel $hills, but nothing is solved and things won't get better.https://www.youtube.com/watch?v=Lqs0oCKMt2s
IUU - Sunday, April 4, 2021 - link
2023 ,maybe too late, for Intel's 7nm. But better late than never. As Intel has been practicing lately. 7nm at Intel's fabs will probably be equivalent to TSMC's 5nm. Only difference is that 5nm is on production now, not in 2023. If TSMC doesn't slow down, Intel will still have to compensate for lagging a node behind by offering a superior architecture.In order for Intel to be like 2006, it needs to produce at the same nm as TSMC, because Intel's nanometers tend to be a "node" better.
IUU - Sunday, April 4, 2021 - link
Taking into consideration Apple always takes advantage of TSMC's lead Intel's chips may never catch up, and the risk of obsolescence is real.healthbestfood - Sunday, April 18, 2021 - link
healthbestfood href="https://healthbestfood.com/pudina-rice-pudina-rice... rice | pudina rice recipe | mint rice | pudina recipes | pudina pulao,</a>