This is 3rd hand, but I heard that the 14nm was rushed with a crazy schedule, and as a result the greybeards who actually knew everything all left. Intel is having to rebuild all that institutional knowledge from scratch again.
They'd be stupid to accept it. Don't work at a place with a bad culture for the money. Unless it's life changing money for a very short duration, it's unlikely to be worth it. Value your quality of life.
The account I just read (linked from the other thread) suggested it was more just that they happened to set really ambitious targets and make a lot of big bets in 10nm, some of which simply did not pan out. Sounds like they may have ended up in a situation where diagnostic and investigation tooling also wasn't working properly due to the use of cobalt, so they'd have ended up with fabrication failures that were then difficult to investigate, as the debugging tools also weren't working. Painful.
20% YoY Revenue increase, 43% YoY Growth in Datacenter Segment. 76% Growth in Non-Volatile Memory solution, another 7% YoY growth in Client Side computing.
This is truly amazing results.
I get stabbed over at Anandtech every time I made an argument for Intel as an AMD investor. The Internet is full of Intel bashing comments but reality is Intel has been making record revenue quarter after quarter.
For the not so good part.
>The company's 7nm-based CPU product timing is shifting approximately six months relative to prior expectations. The primary driver is the yield of Intel's 7nm process, which based on recent data, is now trending approximately twelve months behind the company's internal target.
The most important bit is that last part, 12 months behind Internal target. To put this into perspective.
Intel original 7nm, ( on the assumption it is still the same 7nm because Intel has already changed their 10nm spec) was suppose to be slightly better than TSMC 5nm+, but not as good as TSMC' 3nm. ( In terms of transistor density, by no means it is the only metric it should be judged on ). It was scheduled for 2019, pushed to 2020 then 2H 2021, and are now looking at 1H 2022, which means in terms of consumer getting their hands on it, it would be more like 2H 2022 assuming no further delay.
Given TSMC has never missed a beat in their execution and has iPhone SoC production. It is highly likely by the end of Calendar year 2022, TSMC would have shipped more 3nm Wafer than Intel's 7nm for the whole year.
Why hasn't Intel given up on their internal process nodes? Their advantage is clearly not in owning their process nodes at this point. Dump them and adopt TSMC.
They may be betting that a switchover to TSMC would take as long or longer as doing it themselves. As well there's hubris, sunken costs, and Uncle Sam taking a vested interest in semiconductor manufacturing to take into account.
One could ague that vertical integration has been their strength up until recently. I don't think it suddenly becomes a bad idea just because they dropped the ball recently.
Who would they order from? Keep in mind they are slightly behind Samsung in ic volume. Samsung is #1 and Intel #2. Almost double the volume of TSMC. It takes a few years to build a new plant.
Literally impossible without a very long term plan... and when it’s over they lose control of their destiny. They can beg TSMC for more chips if they want to meet their sales targets, but maybe they’ll sell to Apple or Nvidia instead.
Raw wafer counts aren't particularly relevant here, because you can't really re-tool a NAND or DRAM fab into a competitive logic fab. Based on those numbers, about a third of Samsung's wafer count is logic, putting them slightly ahead of Intel (logic and a bit of NAND). TSMC's wafer count (all logic, AFAIK) is higher than Samsung+Intel, but not all of that is on leading-edge nodes. TSMC 28nm and 16nmFF nodes are going to be sticking around for many years to come.
The numbers are even more tilted in Intel‘s favor if you consider that the foundries do a huge part of their volume business on outdated nodes. Intel doesn‘t do that. They have many times the capacity of TSMC in their market segment.
They're clearly not reaping any reward from their process, TSMC has them beat hands down and they continue to show lack of forward progress towards smaller nodes. They have a much bigger IP portfolio than AMD, and frankly, it appears their foundries are a cost center/boat anchor rather than a profit driver for next-gen silicon.
They don't have to give up their existing 14nm capacity, but next-generation processors could much more easily (apparently) be fab'd externally.
I'm not sure I agree. From what I remember manufacturing is a big part of Intel's profits. Sure, they might not have the best process, but they still have _a_ process, and customers are willing to pay for it.
I agree, there's no reason for intel to shut down their fab, but I don't know that they need to hold progress on their CPUs for like, 5 years now? They can continue to work on their process while fabbing next gen CPUs at TSMC IMO.
Intel is behind, but not actually that behind. Spinning off their fab business at this point in time is throwing money in to the sea. TSMC doesn't have the capacity to handle the volume of Intel orders anyway.
Once they figure out UV, there are lots of in house innovations in the pipeline that could prolong Moore's law for the next 10-15 years. This is according to Jim Keller, who gave talks about how there could be up to 50x density increase. Most of that technology is proprietary and together with TSMC they are heading toward a quasi monopoly.
This obviously puts more wind in AMD's sails. That said, Intel's Datacenter growth is very impressive.
I have been an AMD investor since pre-Zen launch and I think it is fair to say that EPYC sales feel like they haven't reached the levels some have hoped for quite yet. If AMD continue to widen the gap between EPYC and Xeon, I suspect we may look back on this inertia as a bit of a 'roadrunner effect' and in a year or two's time get a shocking quarter where AMD suddenly starts taking surprisingly large bites out of Intel's datacenter market share.
Also on the GPU side of things, it is rumored that Nvidia has struggled to get capacity at TSMC and is having to use Samsung's, supposedly inferior, 8nm process for its forthcoming Ampere cards, at least until their 7nm capacity comes online at TSMC. Not sure how much truth there is to this, but highlights the capacity issues that may currently be in play.
>I have been an AMD investor since pre-Zen launch and I think it is fair to say that EPYC sales feel like they haven't reached the levels some have hoped for quite yet.
I've been looking just these last few weeks at buying a new Epyc Rome system, and one thing I was a bit surprised (though I shouldn't have been, I just haven't thought about it in a long time) about was that while Rome is incredible as a CPU the whole ecosystem around it has lagged by quite a bit. Specifically, the motherboard selection is still well behind what you can find for Xeon. That's not AMD's fault and there is a bit of a standard egg/chicken bootstrap issue, the market has to be deep enough to really start to fill out the niches. Epyc being so good is what has allowed them to start clawing their way up at all, and the MB scene is already massively better then even the start of this year. Still though it's made it harder for me at least to unthinkingly pull the trigger, every option for my intended use-case entails compromises. There is no perfect fit.
I suspect over the next 6-12 months even that will be less and less the case. And hopefully that will mean accelerating momentum for AMD and bring some more long term competitiveness to that market. But it's been interesting being on the purchase hunt and seeing the development in real time, it already feels like the whole game has changed for CPUs but of course everything supporting that lags a cycle or two.
INTC investors are quite unhappy tonight. Down 11.25% at 54.20. Anyone expecting the Robbin Hooders to understand comparative lithography, well... there's a bridge still available in NY.
Maybe INTC will still make a lot of money but there's a "nanometer gap", damnit!
If there really is not further delay, then Intel will be fine. But their 10nm was initially only delayed by 12 months too. And then another 12 months. And another. Intel have big problems if they have those same delays again.
Revenue growth is great but enterprise tech companies can grow revenue long after they are as dead as a doornail. Lotus Notes (!) posted revenue growth in 2008 (!). Intel in some serious shit and the revenue numbers don’t change that.
>Lotus Notes (!) posted revenue growth in 2008 (!).
i worked with Lotus Notes in 1997, and even that 1997 version would be miles ahead of many enterprise software in 2008. A super-hyper-important project had finally [and quietly] failed at Sun around 2008 after several years of trying to build a functionality which was just a one item in the large list of 1997 Lotus Notes capabilities :)
Wrt. the original discussion - Intel - anekdotally i don't see any moves on our huge mostly C++ platform (with our customers running machines with hundreds cores and terabytes of memory) toward AMD, unfortunately. Giving the cost structure of software vs. hardware, our customers will run only what is certified by us, "minor" theoretical savings on AMD hardware be damned. I think that only when Intel will get really slower than AMD, then the enterprise glacier will start moving. If i were AMD, i'd not even bother with enterprise on-premise datacenter, and would focus instead on those few places where the future is - AWS and the likes. By the time Intel, if it happens at all, brings the price/performance and energy/performance (the metrics which are really important to cloud and not that to enterprise datacenter) back to competitive level, the cloud may be owned by AMD.
The 7nm may well likely snowball into a much bigger issue and put the company in jeopardy. E.g. allow competitors to surpass and erode the brand value/perception.
This year is already proving to be a watershed one for Intel, largely due to AMD's continued resurgence and taking the performance crown in mobile (Renoir) and expected in desktop (Zen3). Apple announced their move away from Intel silicon and even AWS making strides in their own ARM server chips (Graviton).
By itself and without further delays, 7nm will already be a hail mary of sorts due to the TSMC's pace and roadmap. But remember that 10nm yields have not been good so will it be able to handle producing server and desktop chips.
If there's another delay (e.g. 7nm in 2023), would Intel need to sell of their fabs?
The naming convention Intel has chosen for the last 10 years (Core iX Generation X) has basically been a long running joke. The average person can’t figure out if a used machine from 2015 has the same CPU as a 2020 machine.
Not that they’ve made the chips noticeably faster for web browsing in that timeline anyway.
Basically, this news is the blackberry storm launch announcement.
As other commenters I'm reminded of the classic RIM roadrunner effect - the decapitated chicken that keeps running.
Another thought though: is it completely unthinkable for Intel to pull an AMD themselves and spin out their fabs and rely on whoever has the best fabs out there? I mean - if they are years late with something that is already outdated there are not too many downsides to it.. Next-gen process nodes are now so delayed that they might as well use the time until then to retarget their designs for TSMC/Samsung.
They can’t. Intel has at least five times the production capacity of TSMC when it comes to the bleeding edge nodes, and companies are already fighting for a share of TSMC capacity, which means contracts will be expensive.
I feel the same about Samsung's Exynos mobile SoCs. While they are definitely worse than Qualcomm's, if they went away Qualcomm would be the only chip maker for Android flagships (excluding Huawei, but without Google play services they don't really count).
Seems like even if they went with TSMC, they are a little bit late to the party. AMD has been optimizing against TSMC's nodes for years.
AMD is also an organization that has been operating with the 3rd party foundry model for 10+ years. How long would it take for Intel to undergo an org-chart/culture transition into a similar model? I have to believe this would be a fairly disruptive time for such a large organization.
Not saying Intel can't pull it off, but it seems like their strategy will have to exist on a >5 year timeline at this point.
For a long time intel had the best manufacturing capabilities. That was a competitive advantage. Just because you are trailing doesn't mean you will trail forever. They might not catch up, but they could catch up as well.
This has got to be interesting for folks like AWS, Google, MS. Is this the opening for AMD? Or will we see things like AWS ARM chips take off? They should both be close to entering production on 5nm chips once TSMC gets enough Apple chips knocked out. Are we going to see more specialization in the form of custom chips as process nodes are stalled out?
I wonder to what extent a six month delay could influence hyperscaler decisions for compute purchasing. Every major cloud now offers AMD Epyc instances at a discount relative to Intel's 14nm(\+)* offerings. It seems unlikely Intel will be competitive until 2022. It seems unlikely any of the top three cloud providers would go all AMD for new instances but I could see the ratio shifting heavily in AMD's favor.
Nobody can flip their whole fleet suddenly -- the TCO-positive lifespan of a datacenter computer is way longer than most people assume -- but at least some of the hyperscalers operate a "machine of the day" process, where 100% of new machines are identical. In that sense, whoever gets the design win gets all of the orders.
But that just means it's on a delay. If AMD gets the design win in 2020 or 2021 then it means more trouble for Intel even if they're competitive again in 2022, because then they're the ones waiting on the next redesign.
Regular people didn't but PCs in 2000 that's just not true. I think most consumers buy laptops nowadays, but that shift started happening around 2005ish, not 2000.
You said regular people haven't bought PCs for a couple of decades. A couple of decades ago PCs were still too expensive for regular people to buy, they were mostly owned by businesses and the upper middle class.
Sales volume peaked somewhere in the mid 2000s. Since then the PC ownership rate hasn't really declined (it's still ~75%), but people haven't bought as many PCs because the rate of performance improvement has slowed so people haven't had as much of a need to replace their existing PCs.
I think the phone makers are trying to avoid that fate now by discontinuing software updates for older phones or making it expensive to replace failed batteries, but eventually one of them is going to defect and offer a phone with in-kernel drivers and a replaceable battery and then win half the market when customers get wise to not having to lay out hundreds of dollars every three years instead of every five or ten.
75% of people own a PC?! Sorry I can’t agree with that. I can’t remember the last time I saw a PC in someone’s house. Maybe ten years ago or more. People have laptops and macs.
And conversely I don’t know anyone who didn’t own a PC in 2000. 2000 was the very height of the dotcom boom. The dotcom boom wasn’t fuelled by a tiny upper middle class - it was about everyone suddenly being online.
>A couple of decades ago PCs were still too expensive for regular people to buy, they were mostly owned by businesses and the upper middle class.
This sounded wrong to me, but in the interests of not just making a kneejerk response, I googled and found total global sales from 1997-2000 estimated at ~400 million PCs[1] and also a claim[2] that at that time about 60 percent of US households had at least one personal computer.
Also, since the 80s, affordable computers were available, and sold in the millions,[3] so I don't think it's valid to say the reason they were mostly owned by businesses and the upper middle class at that point was because they were too expensive. It was more that people in the upper middle class used computers at work and could relate to why they were the future, and that future included them/their children, so they were likely to see value in a new computer rather than something else.
There are some now by AMD, of course, with the new 4900HS chips. Linus was fawning all over the Asus ROG Zephyrus G14 with the 4900HS. Then, we'll see Apple's ARM entries. I don't know that these are mortal wounds but they are certainly the thin edge.
4900HS is 25% faster than the i9-9880HK (Intel's fastest contender) and uses much less power.
This is a 2000s view of Intel that is not accurate in 2020. Look at the numbers.
- 7% growth for consumer (the “Intel PCs” you mention)
- 43% growth for data centers
- 70%+ growth in nonvolatile storage (SSDs)... much of which goes into high end servers in data centers
Will there be more data centers And cloud offerings in 3 years, or less? Many More!
Will there be more or less PC sales in 3 years. For Intel it’s less or the same (Apple is 8-10% of the market that is disappearing and PC growth is flat to slightly up. So 3 years out it will be roughly the same)
Also keep in mind the margin on Xeons vs The laptop chips
That go in a $599 plastic Dell.
Intel as a “PC” company is dying. Intel as a data center provider (SSDs and heavy processors) is ascendant
A single socket AMD threadripper 3995WX (the "ECC" version just released) is likely going to destroy a dual socket Xeon 8275+ (the ones used for the Amazon C5 metal instance) in most benchmarks like the 3990X did, while running at roughly the same power and a fraction of the price. Once AMD convinces cloud providers that its ECC works just as well as Intel's, there's really nothing holding AMD back from taking a huge chunk of the server market.
Growth is flat. For example, vs their 2016 Q3 results they've had 8 down quarters in the Client Computing Group aka consumer.
They'll have the data centres for awhile but when you only have a few customers buying huge volume they will eventually use their buying power to shrink your margins down to nothing. Silicon is due to become a commodity anyway within a decade because of the end of Moore's law.
These two segments account for ~75% of $INTCs revenue today.
But the datacenter is hugely vulnerable. AMD is cheaper and Amazons ARM may be cheaper as well and more importantly validates that cheaper matters. To end customers these offerings are a click away. And while the volumes will take time due to inertia and verification work the pricing happens at the margin. And it seems that margins of processors are already being impacted. Investors are understandably worried as there is nothing on the roadmap stopping further erosion in the one market making the bulk of profits,
If all Apple cared about was faster chips, they could easily call up AMD and move their platform there. It'd be a lot less work for them.
Moving to their own ARM chips lets them gain tighter control over the platform, and not be beholden to Intel or AMDs roadmap or desire to implement specific features Apple wants.
Everybody _wants_ faster chips, but for most people, the chip is in idle state 99% of the time. Outside of the data center and serious gamers, chip speed has been fairly irrelevant for years.
> chip speed has been fairly irrelevant for years.
When you are emulating another architecture, speed often becomes a much bigger factor.
Particularly since Apple caters to a lot of markets where performance is more important. Developers, video & audio pros, animators, photographers, and other pro users are a big part of what makes Apple's higher end systems move. Even if your CPU is idle 90% of the time, when you run a build you want it to go fast.
I'd argue that while Apple serves those markets, it hasn't catered to them for a long time, and it would be fair to say that in many cases it's downright neglected them.
There's a whole lot of people who are in the 'pro user' market who are incredibly frustrated with Apple - slow on hardware refreshes, generally way behind on chip generations, and a focus on design over reliability or performance.
The thing that stops a lot of people from migrating away is a lack of better options.
Some are tied to the Apple Ecosystem, and so there's no legit way of running OSX on non-Apple hardware.
Others don't need to use Apple, but Microsoft in the Windows 10 era is arguably worse, and they still haven't figured out that they need to fire all those people who keep pushing user-unfriendly choices like changing defaults on already installed systems, pushing app store apps down people's throats, etc.
For those who could move to Linux - there's a lack of nicely designed hardware that has all the modern bells & whistles, and that doesn't have weird software quirks. Particularly in the mobile space, but also to an extent in the desktop space, too.
Even when you look at a PC builder like System76's systems, their laptops still look like something out of a catalogue from 10 years ago, with none of the clean aesthetic and polish of Apple, or even premium Windows laptops. The desktops are better on that front.
Sure, you could, like I have, run a modern Linux distro on say a Lenovo X1 Carbon, or a HP Spectre X360 if you want something that looks pretty decent and has good performance, but even they have weird hardware quirks. Sound, for instance, is amazingly frustrating, and something I have to mess around with regularly, sometimes multiple times a day. It makes me look like an incompetent idiot when I'm talking away on a conference call to discover that for some bizzare reason, between Firefox/Chrome/Chromium and the System sound settings - what was working just fine not 30 minutes ago has decided that it no longer wants to cooperate, and has switched, yet again to a profile I don't want and don't need.
On the desktop side is no better - the few companies selling desktops, are mostly selling overpriced garbage aimed at Windows users, so you need to go through a Custom PC Builder using enthusiast equipment, which means spending half your time trying to find out if the 50W of RGB LEDs that the damn device comes with can be switched off (Answer: Probably not, unless you run Windows, or are willing to pull out a soldering iron. Which makes for a great discussion if you need warranty support on something).
> I'd argue that while Apple serves those markets, it hasn't catered to them for a long time, and it would be fair to say that in many cases it's downright neglected them.
This comment is dated. They had a stretch where they'd lost their way, but Apple's Mac lineup has been in pretty good shape for the past ~2 years. With them fixing the MacBook keyboard, they have a lineup which is more or less solid across the board.
A lot of their processor issues revolved around using specific CPUs which Intel was slow to update... an issue they won't have when they move to their own silicon.
Absent of AMD processors for desktop is recently become a problem for performance. Absent of NVIDIA GPU has been also a problem for performance and CUDA availability.
> I am waiting for some friggin call-home/update-check/irrelevant network request.
Is that right? Any examples? I feel most of the times I'm waiting is for the network because everything is online now. Even my work is all done on a remote VM in a data center. I connect using vscode over ssh. And the app is snappy af except when it needs to do a network opeation.
For some basic apps:
Notepad++ annoys me most. Modern, updated regularly.
But wants to update plugins before even appearing it seems?!?
iTunes(thanks to iWife) annoys me with it's updates. It checks for updates after you've started the program, pops up to get in your way and doesn't give you an "Install on exit" option.
Paint.Net does it right. Opens up, lets you get things done, checks for an update, gives you the option to install on exit. More should be done this way.
Playing games, Steam and all the many games with their own servers etc.
Example of bad, It takes me roughly 2 mins to get into GhostRecon:Breakpoint.(not that good a game really but that's another day).
Example of good, Rocket League. Starts up, gets you going. (Disappointingly not a League of actual Rockets but good fun anyway. :-))
> Moving to their own ARM chips lets them gain tighter control over the platform, and not be beholden to Intel or AMDs roadmap or desire to implement specific features Apple wants.
You are talking strategy, I am talking tactics.
Long term, Apple will benefit hugely by having their own silicon for exactly the reasons you suggest.
Short term, being competitive on performance is massively important. Arguably, the most important thing to make the transition successful. So much so that I doubt this transition would even be happening if they weren't confident about out-performing Intel in the near term.
Aren’t these node sizes mostly marketing BS that can’t be compared across manufacturers anyway? If Intel’s products can compete on performance or price or power consumption then who cares what they call it?
Some of it is marketing BS, but it's hard to argue that AMD isn't benefitting hugely from TSMC's 7nm process. Likewise, Apple knows that they can get 5nm parts from TSMC this year for their phones and laptops and they know the kind of performance advantage they're going to be getting out of them.
The objective metrics are things like: transistors per square millimetre, power per billion transistors, frequency, SRAM bit-cell size, etc...
Marketing calls it "7nm+++super", engineers get graphs of power-vs-frequency.
E.g.: comparing transistor densities:
Intel 14nm 38 MTr/mm^2
TSMC 7nm 97 MTr/mm^2
Intel 10nm 100 MTr/mm^2
Samsung 5nm 127 MTr/mm^2
TSMC 5nm 173 MTr/mm^2
Intel 7nm "higher than TSMC 5nm" (planned for 2022)
TSMC 3nm 290 MTr/mm^2 (risk production in 2021)
This should make it pretty clear why Intel is going to struggle to be competitive: The majority of their products are stuck on 14nm, which has 22% of the transistor density of TSMC's bleeding edge product. They can't compete with the likes of AMD and Apple with such outdated tech.
It should also make it clear that Intel's 10nm node is basically the same as TSMC's 7nm.
However, from what I heard (I'm not an expert!), Intel's 10nm node is a "pain in the ass" to manufacture. The design phase is complex, the yields are low, etc...
For comparison, TSMC's 7nm is supposed to be a relatively "straightforward" process, and 5nm is actually an improvement in manufacturability. That is, it's easier to design and make a 5nm TSMC chip than a 7nm TSMC chip. That's actually a pretty big advantage, something Intel doesn't talk about much...
You are spot on on 10nm fab, it's why Apple shipped an update with split 14 and 10nm chips on different models, it wasn't large enough or cheap enough to update all models.
A notable thing about TSMC's 5nm is it is shipping in three months. Thats a massive gap and there is little proof that Intel 7nm is 'much' more than 5nm.
And thats before you consider other upsides seen in the TSMC chips such as cost, or the incredible power management tactics seen in the Ax series chips that are easily generations ahead of Intel.
Intel's 10% stock drop is all about losing investor confidence. They've been given a pass for delaying 10nm for years in part because there was no competition and assurances that it was a one-off thing and they have learned their lesson. This delay threatens that narrative and instead suggest that Intel hasn't learned from their 10nm debacle.
Is there any reputable reporting, or even just wild gossip, about why Intel is behind?
The cynical answer is “they were focused on profit margins and didn’t invest in process”, but is it possible they bet hard on some technology that didn’t pan out? Ultraviolet lithography? Some exotic indium gallium arsenide magic that didn’t work in mass production?
There seems to be consensus about their 10nm failures: they were trying every trick in the book except switching to EUV lithography, betting that they could get ahead and stay ahead while EUV power levels creeped upward toward what's needed for volume manufacturing. But Intel bit off more than they could chew and ended up with a process that simply didn't work (cf. the disabled iGPU on Cannonlake). So Intel scaled back their ambitions and spent more time working out the kinks, and they're apparently still a year away from having a 10nm desktop CPU.
I'm not sure if we have any good clues as to what's going wrong with Intel's 7nm efforts, but it seems reasonable to guess that they weren't putting as much effort as their competitors into getting ready for EUV.
And given how long Intel's had these issues at this point, it seems pretty clear that they have some broader management issues.
With regards to the specific technologies Intel was trying to do a lot of new things with 10nm. Initial rumors said that replacing copper with cobalt in some of the metal layers was causing problems. But later rumors were that Contact Over Active Gate (COAG) used mostly in the GPU section of a chip was the main culprit and why none of the early 10nm parts had working GPUs.
103 comments
[ 2.9 ms ] story [ 173 ms ] threadThis is truly amazing results.
I get stabbed over at Anandtech every time I made an argument for Intel as an AMD investor. The Internet is full of Intel bashing comments but reality is Intel has been making record revenue quarter after quarter.
For the not so good part.
>The company's 7nm-based CPU product timing is shifting approximately six months relative to prior expectations. The primary driver is the yield of Intel's 7nm process, which based on recent data, is now trending approximately twelve months behind the company's internal target.
The most important bit is that last part, 12 months behind Internal target. To put this into perspective.
Intel original 7nm, ( on the assumption it is still the same 7nm because Intel has already changed their 10nm spec) was suppose to be slightly better than TSMC 5nm+, but not as good as TSMC' 3nm. ( In terms of transistor density, by no means it is the only metric it should be judged on ). It was scheduled for 2019, pushed to 2020 then 2H 2021, and are now looking at 1H 2022, which means in terms of consumer getting their hands on it, it would be more like 2H 2022 assuming no further delay.
Given TSMC has never missed a beat in their execution and has iPhone SoC production. It is highly likely by the end of Calendar year 2022, TSMC would have shipped more 3nm Wafer than Intel's 7nm for the whole year.
I am waiting to see AMD's results.
Literally impossible without a very long term plan... and when it’s over they lose control of their destiny. They can beg TSMC for more chips if they want to meet their sales targets, but maybe they’ll sell to Apple or Nvidia instead.
They don't have to give up their existing 14nm capacity, but next-generation processors could much more easily (apparently) be fab'd externally.
I have been an AMD investor since pre-Zen launch and I think it is fair to say that EPYC sales feel like they haven't reached the levels some have hoped for quite yet. If AMD continue to widen the gap between EPYC and Xeon, I suspect we may look back on this inertia as a bit of a 'roadrunner effect' and in a year or two's time get a shocking quarter where AMD suddenly starts taking surprisingly large bites out of Intel's datacenter market share.
Also on the GPU side of things, it is rumored that Nvidia has struggled to get capacity at TSMC and is having to use Samsung's, supposedly inferior, 8nm process for its forthcoming Ampere cards, at least until their 7nm capacity comes online at TSMC. Not sure how much truth there is to this, but highlights the capacity issues that may currently be in play.
I've been looking just these last few weeks at buying a new Epyc Rome system, and one thing I was a bit surprised (though I shouldn't have been, I just haven't thought about it in a long time) about was that while Rome is incredible as a CPU the whole ecosystem around it has lagged by quite a bit. Specifically, the motherboard selection is still well behind what you can find for Xeon. That's not AMD's fault and there is a bit of a standard egg/chicken bootstrap issue, the market has to be deep enough to really start to fill out the niches. Epyc being so good is what has allowed them to start clawing their way up at all, and the MB scene is already massively better then even the start of this year. Still though it's made it harder for me at least to unthinkingly pull the trigger, every option for my intended use-case entails compromises. There is no perfect fit.
I suspect over the next 6-12 months even that will be less and less the case. And hopefully that will mean accelerating momentum for AMD and bring some more long term competitiveness to that market. But it's been interesting being on the purchase hunt and seeing the development in real time, it already feels like the whole game has changed for CPUs but of course everything supporting that lags a cycle or two.
Maybe INTC will still make a lot of money but there's a "nanometer gap", damnit!
i worked with Lotus Notes in 1997, and even that 1997 version would be miles ahead of many enterprise software in 2008. A super-hyper-important project had finally [and quietly] failed at Sun around 2008 after several years of trying to build a functionality which was just a one item in the large list of 1997 Lotus Notes capabilities :)
Wrt. the original discussion - Intel - anekdotally i don't see any moves on our huge mostly C++ platform (with our customers running machines with hundreds cores and terabytes of memory) toward AMD, unfortunately. Giving the cost structure of software vs. hardware, our customers will run only what is certified by us, "minor" theoretical savings on AMD hardware be damned. I think that only when Intel will get really slower than AMD, then the enterprise glacier will start moving. If i were AMD, i'd not even bother with enterprise on-premise datacenter, and would focus instead on those few places where the future is - AWS and the likes. By the time Intel, if it happens at all, brings the price/performance and energy/performance (the metrics which are really important to cloud and not that to enterprise datacenter) back to competitive level, the cloud may be owned by AMD.
This year is already proving to be a watershed one for Intel, largely due to AMD's continued resurgence and taking the performance crown in mobile (Renoir) and expected in desktop (Zen3). Apple announced their move away from Intel silicon and even AWS making strides in their own ARM server chips (Graviton).
By itself and without further delays, 7nm will already be a hail mary of sorts due to the TSMC's pace and roadmap. But remember that 10nm yields have not been good so will it be able to handle producing server and desktop chips.
If there's another delay (e.g. 7nm in 2023), would Intel need to sell of their fabs?
Unclear if Intel is following that path.
The naming convention Intel has chosen for the last 10 years (Core iX Generation X) has basically been a long running joke. The average person can’t figure out if a used machine from 2015 has the same CPU as a 2020 machine.
Not that they’ve made the chips noticeably faster for web browsing in that timeline anyway.
Basically, this news is the blackberry storm launch announcement.
Another thought though: is it completely unthinkable for Intel to pull an AMD themselves and spin out their fabs and rely on whoever has the best fabs out there? I mean - if they are years late with something that is already outdated there are not too many downsides to it.. Next-gen process nodes are now so delayed that they might as well use the time until then to retarget their designs for TSMC/Samsung.
I think them moving to outsourced foundries would be a huge move in the wrong direction for fab consolidation.
AMD is also an organization that has been operating with the 3rd party foundry model for 10+ years. How long would it take for Intel to undergo an org-chart/culture transition into a similar model? I have to believe this would be a fairly disruptive time for such a large organization.
Not saying Intel can't pull it off, but it seems like their strategy will have to exist on a >5 year timeline at this point.
Because they are easy to buy. A great non Intel hardware and software laptop+aio by AMD or someone else would be a mortal wound on $INTC.
Regular people don’t buy PCs at all and haven’t for a couple of decades.
Sorry I've got no idea what this means.
Sales volume peaked somewhere in the mid 2000s. Since then the PC ownership rate hasn't really declined (it's still ~75%), but people haven't bought as many PCs because the rate of performance improvement has slowed so people haven't had as much of a need to replace their existing PCs.
I think the phone makers are trying to avoid that fate now by discontinuing software updates for older phones or making it expensive to replace failed batteries, but eventually one of them is going to defect and offer a phone with in-kernel drivers and a replaceable battery and then win half the market when customers get wise to not having to lay out hundreds of dollars every three years instead of every five or ten.
And conversely I don’t know anyone who didn’t own a PC in 2000. 2000 was the very height of the dotcom boom. The dotcom boom wasn’t fuelled by a tiny upper middle class - it was about everyone suddenly being online.
This sounded wrong to me, but in the interests of not just making a kneejerk response, I googled and found total global sales from 1997-2000 estimated at ~400 million PCs[1] and also a claim[2] that at that time about 60 percent of US households had at least one personal computer.
Also, since the 80s, affordable computers were available, and sold in the millions,[3] so I don't think it's valid to say the reason they were mostly owned by businesses and the upper middle class at that point was because they were too expensive. It was more that people in the upper middle class used computers at work and could relate to why they were the future, and that future included them/their children, so they were likely to see value in a new computer rather than something else.
[1]https://en.wikipedia.org/wiki/Market_share_of_personal_compu...
[2]https://hypertextbook.com/facts/2004/DianeEnnefils.shtml
[3]By 1985, the Commodore 64 was selling for <$150, see https://en.wikipedia.org/wiki/Commodore_64
4900HS is 25% faster than the i9-9880HK (Intel's fastest contender) and uses much less power.
https://linustechtips.com/main/topic/1187443-asus-caught-bei...
- 7% growth for consumer (the “Intel PCs” you mention)
- 43% growth for data centers
- 70%+ growth in nonvolatile storage (SSDs)... much of which goes into high end servers in data centers
Will there be more data centers And cloud offerings in 3 years, or less? Many More!
Will there be more or less PC sales in 3 years. For Intel it’s less or the same (Apple is 8-10% of the market that is disappearing and PC growth is flat to slightly up. So 3 years out it will be roughly the same)
Also keep in mind the margin on Xeons vs The laptop chips That go in a $599 plastic Dell.
Intel as a “PC” company is dying. Intel as a data center provider (SSDs and heavy processors) is ascendant
Anyway EPYC is performs far better than Xeon in same price.
Moving to their own ARM chips lets them gain tighter control over the platform, and not be beholden to Intel or AMDs roadmap or desire to implement specific features Apple wants.
Everybody _wants_ faster chips, but for most people, the chip is in idle state 99% of the time. Outside of the data center and serious gamers, chip speed has been fairly irrelevant for years.
When you are emulating another architecture, speed often becomes a much bigger factor.
Particularly since Apple caters to a lot of markets where performance is more important. Developers, video & audio pros, animators, photographers, and other pro users are a big part of what makes Apple's higher end systems move. Even if your CPU is idle 90% of the time, when you run a build you want it to go fast.
There's a whole lot of people who are in the 'pro user' market who are incredibly frustrated with Apple - slow on hardware refreshes, generally way behind on chip generations, and a focus on design over reliability or performance.
The thing that stops a lot of people from migrating away is a lack of better options.
Some are tied to the Apple Ecosystem, and so there's no legit way of running OSX on non-Apple hardware.
Others don't need to use Apple, but Microsoft in the Windows 10 era is arguably worse, and they still haven't figured out that they need to fire all those people who keep pushing user-unfriendly choices like changing defaults on already installed systems, pushing app store apps down people's throats, etc.
For those who could move to Linux - there's a lack of nicely designed hardware that has all the modern bells & whistles, and that doesn't have weird software quirks. Particularly in the mobile space, but also to an extent in the desktop space, too.
Even when you look at a PC builder like System76's systems, their laptops still look like something out of a catalogue from 10 years ago, with none of the clean aesthetic and polish of Apple, or even premium Windows laptops. The desktops are better on that front.
Sure, you could, like I have, run a modern Linux distro on say a Lenovo X1 Carbon, or a HP Spectre X360 if you want something that looks pretty decent and has good performance, but even they have weird hardware quirks. Sound, for instance, is amazingly frustrating, and something I have to mess around with regularly, sometimes multiple times a day. It makes me look like an incompetent idiot when I'm talking away on a conference call to discover that for some bizzare reason, between Firefox/Chrome/Chromium and the System sound settings - what was working just fine not 30 minutes ago has decided that it no longer wants to cooperate, and has switched, yet again to a profile I don't want and don't need.
On the desktop side is no better - the few companies selling desktops, are mostly selling overpriced garbage aimed at Windows users, so you need to go through a Custom PC Builder using enthusiast equipment, which means spending half your time trying to find out if the 50W of RGB LEDs that the damn device comes with can be switched off (Answer: Probably not, unless you run Windows, or are willing to pull out a soldering iron. Which makes for a great discussion if you need warranty support on something).
This comment is dated. They had a stretch where they'd lost their way, but Apple's Mac lineup has been in pretty good shape for the past ~2 years. With them fixing the MacBook keyboard, they have a lineup which is more or less solid across the board.
A lot of their processor issues revolved around using specific CPUs which Intel was slow to update... an issue they won't have when they move to their own silicon.
Am I waiting for CPU? No. Storage? No.
I am waiting for some friggin call-home/update-check/irrelevant network request.
Edit:brevity, removal of profane words.
Is that right? Any examples? I feel most of the times I'm waiting is for the network because everything is online now. Even my work is all done on a remote VM in a data center. I connect using vscode over ssh. And the app is snappy af except when it needs to do a network opeation.
iTunes(thanks to iWife) annoys me with it's updates. It checks for updates after you've started the program, pops up to get in your way and doesn't give you an "Install on exit" option.
Paint.Net does it right. Opens up, lets you get things done, checks for an update, gives you the option to install on exit. More should be done this way.
Playing games, Steam and all the many games with their own servers etc. Example of bad, It takes me roughly 2 mins to get into GhostRecon:Breakpoint.(not that good a game really but that's another day).
Example of good, Rocket League. Starts up, gets you going. (Disappointingly not a League of actual Rockets but good fun anyway. :-))
You are talking strategy, I am talking tactics.
Long term, Apple will benefit hugely by having their own silicon for exactly the reasons you suggest.
Short term, being competitive on performance is massively important. Arguably, the most important thing to make the transition successful. So much so that I doubt this transition would even be happening if they weren't confident about out-performing Intel in the near term.
Shifting to ARM for laptops and desktops just sounds like a big burden for them.
Intel has stumbled a lot here and it's showing.
Marketing calls it "7nm+++super", engineers get graphs of power-vs-frequency.
E.g.: comparing transistor densities:
This should make it pretty clear why Intel is going to struggle to be competitive: The majority of their products are stuck on 14nm, which has 22% of the transistor density of TSMC's bleeding edge product. They can't compete with the likes of AMD and Apple with such outdated tech.It should also make it clear that Intel's 10nm node is basically the same as TSMC's 7nm.
However, from what I heard (I'm not an expert!), Intel's 10nm node is a "pain in the ass" to manufacture. The design phase is complex, the yields are low, etc...
For comparison, TSMC's 7nm is supposed to be a relatively "straightforward" process, and 5nm is actually an improvement in manufacturability. That is, it's easier to design and make a 5nm TSMC chip than a 7nm TSMC chip. That's actually a pretty big advantage, something Intel doesn't talk about much...
A notable thing about TSMC's 5nm is it is shipping in three months. Thats a massive gap and there is little proof that Intel 7nm is 'much' more than 5nm.
And thats before you consider other upsides seen in the TSMC chips such as cost, or the incredible power management tactics seen in the Ax series chips that are easily generations ahead of Intel.
The cynical answer is “they were focused on profit margins and didn’t invest in process”, but is it possible they bet hard on some technology that didn’t pan out? Ultraviolet lithography? Some exotic indium gallium arsenide magic that didn’t work in mass production?
I'm not sure if we have any good clues as to what's going wrong with Intel's 7nm efforts, but it seems reasonable to guess that they weren't putting as much effort as their competitors into getting ready for EUV.
And given how long Intel's had these issues at this point, it seems pretty clear that they have some broader management issues.
https://semiaccurate.com/2018/06/29/intels-firing-of-ceo-bri...
With regards to the specific technologies Intel was trying to do a lot of new things with 10nm. Initial rumors said that replacing copper with cobalt in some of the metal layers was causing problems. But later rumors were that Contact Over Active Gate (COAG) used mostly in the GPU section of a chip was the main culprit and why none of the early 10nm parts had working GPUs.
History says otherwise...