Intel is dying and we can’t save it
sithsiri.medium.com
sithsiri.medium.com
But dying??
Nup, I'm not seeing this at all.
I mean AMD for years survived in a situation much worse then Intel currently is.
Intel's productions nodes might be behind TSMC's but they are still quite valuable, not all chips need 5nm processes. Heck we could say most chips don't need it.
Intel is still innovating, just because they currently lack behind doesn't mean that will stay that way.
Even when lacking behind wrt. the production node Intel still manged to produce Chips which are often not much worse then the competition.
Intel is not just about CPU's.
Etc.
So no Intel is not dying at all.
It's taking damage, and lost dominance but it still has a fairly good shot at survival and might even take back dominance (in a 2-3 years). Or it might not but still stay competitive.
And if we look at the world situation TSMC might not be a think anymore in a view years. It would make me supper angry but thinks are heading in that direction (China Taiwan conflict).
I was bullish on Intel until quite recently. I was able to convince myself that they were taking the L on 10nm to focus completely on a full, true EUV transition, in order to be at the top of the game just everyone else was hitting that wall in order to dominate again. That turned out to be overly optimistic.
For instance, Steam's hardware report indicates that 1/3 of their install base already has 6 cores or more [1].
https://data.firefox.com/dashboard/hardware#goto-os-and-arch...
That's one of the few great things about capitalism - competition drives innovation.
Just so long as the aforementioned monopolies don't choke out competition. Sadly, we seem to end up with a lot of pseudo-monopolies in the tech industry.
I wonder how much more competition we could have with strict monopolies regulation on acquisitions and mergers?
I still have my Core2 Duo in my HTPC... it's starting to feel doggy though.
Though that may have been a bit easier for them to solve, because the main thing they had to do was scrap NetBurst (Pentium 4) and return to a derivative their old architecture (which they were still using for the notebook market).
I'd be interested in hearing more about this. Are new graduates actually not applying?
In sports it's fun to say that 2nd place is just the first loser, but that's all in jest and friendly banter. When making things like this overly dramatic, it serves nobody except those trying to short the stock.
Earning more than $60k/year puts you in the 1% already. [1]
I would be fairly surprised if $60K is top 1%, even if you have the entire world population in the denominator. But when I checked your link, it's not using the entire population for the denominator, (which is correct, IMO) so I don't think it's even close.
In the US, there are about 110M fulltime wage earners, (out of 330M population) with median being ~$50K and top quartile ~$80K. My very rough estimate based on that might be 45M people >$60K. That is a lot more than 110/330 * 1% * 7.8B = 26M before adding in Europe (lower wages but twice the population) or anywhere else.
Median in the US is not clear to me - I saw a figure of 60K and 35K within 2 minutes of searching. I think wage earning gets divided across family size, so those 45M people in your calculation, after dividing by family size, would have a much smaller income.
I think the difference between this convoluted statement and normal measures of inequality could be approximately translated to "most people making > $60K have dependents". It seems like a bad way to talk about inequality, because what goal would it imply to reduce inequality? Prevent people from having dependents, or prevent them from not having dependents?
Having an income of $60k while having a dependent is very different than when not having a dependent. I seems correct to say that without a dependent, you're in the top 1% of the world's wage earners (with respect to your financial freedom); while if you have a dependent, your income is no longer fully yours - and you are no longer in the top 1% of the world's wage earners.
Understatement much? You could also say Elon has more money than 50% of humans on the planet.
Well, they won't even have that EUV capacity in 2-3 years. ASML is almost fully booked at this point.
The way the industry is moving to custom chips won't wait for Intel. And realistically, the trend is already there, next stage of cloud evolution would be application specific chips for cloud native services, like storage/DB/ML, etc. There is zero chance it would scale back.
And I am not sure why everyone here is taking 'dying' so literally. Losing influence is another form of dying. It will take a long time for a company to cease to exist after it becomes irrelevant.
- "How did Intel die?"
- "Well, first slowly. Then quickly."
There's not much that a company so heavily invested in x86 can do in the face of Apple Silicon, aside from executing well on 7nm and bringing it to market ahead of schedule.
I don't see much of a similarity between Intel CPUs and Blackberry phones, considering that many flagship notebooks (even Apple's) have Intel inside. I for one am heralding the new era of voracious semiconductor demand, and am rooting for Intel to get its game face on and turn things around before the opportunity passes.
In reality, there is nothing that Apple or AMD are doing that Intel could not also do, isn't already doing, or isn't in a hurry to also do.
Consider that at 14nm+++++, Intel still manages to outpace AMD with its 7nm CPUs and latest architecture. AMD competes mainly on price, which is not a great position to be in. Intel's profit margins are massively better.
It's possible that x86 will become less relevant in the future, but in that case Intel will still be one of the be best chip designers in the industry. They will once again produce ARM chips, perhaps borrow one or two insights from Apple's architecture. They will continue to outsource manufacturing to other fabs, perhaps even become fabless.
On the contrary, Intel has refused to do any architectural innovation for nearly 20 years. They are capable, but overly conservative. They should have been in a hurry 10 years ago. It's well documented that they weren't and were resented for it. AMD came up with a winning strategy (multi-die packaging) and Intel still doesn't seem in a hurry to catch up. You could swap "AMD" in that sentence with "Apple" and maintain the meaning.
When superscalar or simultaneous multithreading came out you didn't see competitors mill about waiting for the customers to forget about the advantages.
Just because AMD may be winning in Cinebench doesn't mean it's winning in the business side.
Multi-chip packages have downsides that become obvious in certain enterprise workloads, so the question becomes: Is it worth giving up on the monolithic approach, or adding multi-chip products, in order to completely dominate the market, even when that is not more profitable?
Clearly, the answer for Intel must be "no", otherwise they would being doing it.
What are the downsides, exactly? Thread-to-thread latency is the key one that comes to mind. It is far cheaper to make a multi-die package because yields are so much higher for any given node. The business advantages are great. The bleak picture for Intel that is being painted is warranted.
Have you been living under a rock for this past year? This is not true. AMD is beating Intel in performance in every market segment right now.
If you look at the mobile chips that did ship on 10nm, Intel is ahead as well.
Also, never compare chips that are actually on the market to unreleased ones. Zen3 might well get a small frequency boost when Intel finally releases Rocket Lake.
> Also, never compare chips that are actually on the market to unreleased ones.
Watch me do it:
https://www.notebookcheck.net/Rocket-Lake-S-vs-Zen-3-Geekben...
Am I going to gamble my house on this being true? Of course not. We're having a nerdy discussion on IPC. It doesn't really matter either way.
Perhaps you're right and these chips won't "improve evenly" across all benchmarks, but then again nobody has "even" workloads. Rocket-Lake-S has some machine learning acceleration that's going to be useless for many people, valuable for others. It has a GPU onboard, which can come in handy. I'm personally interested in AVX-512. No, I don't need to know why you think AVX-512 sucks.
This is an entirely valid position, and I agree that it bodes well for Intel that they are able to do this. However, it is a very different statement from your original:
> Consider that at 14nm+++++, Intel still manages to outpace AMD with its 7nm CPUs and latest architecture. AMD competes mainly on price, which is not a great position to be in. Intel's profit margins are massively better.
AMD is in no way "competing mainly on price". In fact, if you look at prices on the market instead of MSRP, they are generally the more expensive option right now, and are still sold out on most models almost everywhere.
Otherwise, I do not see any issue with what I said. I am looking at Intel as a company, and there I see massively better profit margins. Street prices may not reflect this during periods of shortages, but it is the dealers who are pocketing that difference, not AMD.
Which means Intel is more or less on equal design footing with AMD, they're just having trouble getting it built.
And Intel has multiple huge revenue streams that are not in question at all. All their networking stuff, for example (everyone, even on AMD, still goes for motherboards with Intel LAN instead of anyone else's, and same with wifi). And things like Optane. And all their tooling & software suites. Intel's CPU division is struggling to get wins, but Intel overall as a company isn't.
Yeah, a lot of people seem to ignore the fact that AMD was not killing it in the CPU market for a while a few years back. I wouldn't be surprised if Intel makes a comeback eventually. It's not a huge deal to me who's in the lead at any given time, I'm more interested in competitive CPUs.
I'm most interested in the future of x86 as a whole. It's definitely got a while left, but I wonder how much market share will start to be taken by ARM (and hopefully RISCV).
Intel is ABSOLUTELY just about CPUs. I can’t think of a single thing that they produce that I would ever pay them money for other than CPUs, especially if they aren’t relevant in the CPU world anymore. Intel has tried to get into so many other markets over the years, but the company’s culture and the nature of their high margin, high R&D business made it impossible to compete in those other markets.
Intel's productions nodes might be behind TSMC's but they are still quite valuable, not all chips need 5nm processes. Heck we could say most chips don't need it.
That’s a very different market with very different margins. If they shrank so much that they became like Maxim or something, then that’s really not much different than going out of business altogether.
Having a US based company with fabrication facilities on US soil is vital to national security and will be for the entirety of my lifetime and likely my grandchildren's lifetime (if I ever have any). For all the bluster of the investors wanting to chop Intel up: they will find a swift and full rejection from the government unless their plan includes selling the assets to another US firm.
TSMC building a fab in the US won't be "good enough" - the US military will not allow the security of our country to be in the hands of a foreign corporation.
Now whether Intel remains a dominant player? Who knows, but they aren't going anywhere. I wouldn't be even a little bit surprised if for instance the 3-letter agencies cloud contracts suddenly included a clause that all of their infrastructure must run on Intel CPUs.
I'm not saying it's right, I'm not saying it's fair, but I don't see any other outcome.
It's easy to say "that's dumb" - but what's your alternative? If it includes "just buy CPUs from insert foreign entity" - that's a nonstarter for myriad reasons.
There is no way the USG will let Intel fail.
Internal information on stuff we don't know about, especially apropos of ring -1 (or even lower if that's even possible), will never leave US soil.
I think that's an important objective. I'm not sure what other way there is to fulfil it, other than spend even more taxpayer money funding greenfield US chip industry
The bigger problem is that Intel is a company that both designs CPUs and manufactures them. The government isn't actually interested in Intel's CPUs specifically, only the domestic manufacturing capability. Intel has to be split up before the government can get what it wants.
>Why can’t people learn from all the instances where a government tried to micromanage and horribly failed in the mid/long run?
Counter example: Central banks are making lots of money available but the governments are not using it to employ people (either directly or indirectly) so it just fuels a long term asset bubble without achieving the desired economic growth.
[1]: https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...
If the US loses influence over Taiwan and subsequently TSMC, they've lost. In this phase of global geopolitics, whoever controls the chips has the influence. You can't do anything without compute resources.
I'm not sure that I've personally bought in to the AI revolution, but it sounds like it's generally agreed to be the future.
At the same time, there is so much room for basic improvement in the defense space. I have a hard time believing 5nm vs 10nm is a serious problem when half the vetronics system is analog.
But yes, in a big picture strategic sense, AI capabilities seems like the area most likely to pose a serious risk by not leading the world in chip technology.
IBM's Fab was on the DoD trusted foundry list [2] Slide 13, Intel's is not.
There's an article that already rang this alarm around national-security when IBM divested itself of its Fab. [3]
1. https://www.computerworld.com/article/2837426/ibms-chip-busi...
2. http://jteg.ncms.org/wp-content/files/documents/DoD%20Truste...
3. https://semiengineering.com/a-crisis-in-dods-trusted-foundry...
1. https://www.reuters.com/article/us-intel-manufacturing/intel...
2. https://www.wsj.com/articles/trump-and-chip-makers-including...
Now, the US may lag behind top of the line fabrication tech (It already does). But it will never lose it's ability to fabricate chips.
For military applications, unless they start getting really into artificial intelligence (maybe they already are?) they don't need super advanced chips for most of their equipment. Simple micro-controllers are enough for most military applications.
I don't think that Huawei is doing is doing just fine.
You have to think about scenarios where countries are opposed to the US for one reason or another.
The Patriot Act being renewed for 6 months and thus putting Cuba in the embargoed countries even more than they already were should be a good second example.
Taiwan isn't under Chinese control, but if there was an escalating situation, you would assume that it would fall under Chinese control VERY quickly.
Up here in Canada, there is no need for fabs as long as the US has them. Canada has no real choice but friendly relations with the US (geography forces this), the upside is that many of the strategic resources belonging to the US are available to Canada.
https://www.eetimes.eu/eu-signs-e145bn-declaration-to-develo...
> As such, China began an initiative to build its own processors back in the early 80's that eventually evolved into the "Made in China 2025" campaign that has the goal of producing 70% of the chips used in China by 2025.
(ignoring that TSMC is Taiwanese and not currently under PRC rule but is instead de-facto independent and mostly internationally recognized as being independent...)
I know rare-earth metals aren't that rare and we could ultimately supply them locally but the lack of any open mines or refining setups is just as big a strategic hole as losing domestic Fab isn't it?
I am happy to be corrected.
At that time, The ratio of excellent to useless people I experienced there was 1:40. That is way out of whack. I am certain it only got worse because all the 1/40’s I knew left.
They are going to have to go and get people who have modern skillsets and mindsets and those people don’t want to live in Hillsboro. And it’s going to be expensive.
And those people don’t tolerate dead weight.
My take is that Bob Swan was only ever a placeholder. The board needed to be put in the worst possible negotiating position: Literally on their knees begging because they are getting their asses kicked in every single competitor category.
A smart guy like Pat wouldn’t take the CEO job until they (a) are crawling on all fours and (b) give him the latitude to clean house as vigorously as he wants and (c) a preposterous paycheck.
No one intelligent would want that job without all three. That’s why Bob got to babysit that role for two years.
This has to be hyperbole but the rest of your comment makes me think you're being serious.
I have only worked at small companies <50 people, so I don't have first hand experience with how waste can hide in large companies but 1:40 excellent to useless people? I can't believe it'd even be half that.
My guess is you thought those people were useless because you didn't understand the role they played in the company.
However, I'll admin again that I don't have first hand experience. I'd genuinely love to hear from others who have experience with large companies.
At one point I posited that Sony USA were merely a political pawn for Sony Japan. Hire a few tens of thousands of Americans, gain political capital -- and care not what they actually do all day. In our case, it was death by scrum and sticky notes.
aw come on, I'm sure you'll meet somebody, find happiness and forget all about him
First, you have Price's Law [0] which states that half the work is done by the square root of the number of people. This in itself already implies a lot of useless people, especially at a company the size of Intel.
You might also have whole orgs/departments where pretty much everyone is phoning it in, so that would drive it up as well.
The way I see it is the more layers there are and the more removed people at the top are from the work, the greater chance the of useless people being able to go under the radar.
Most of the low hanging fruits can be picked by one individual. Instead of trying to pick more apples from the same tree people should make sure that there are not too many people assigned to the same tree.
The article confirms my idea:
>Look at your current profession. Are you in a position to create substantial value? If the answer is no, move on to a different place where you CAN.
I've worked for 100-person, 1000-person, and 100000-person companies, and the biggest companies have so many resources that they barely know what to do with them all. Most startups would kill for a fraction of a percent of the "let's screw around" R&D money that the big corps have.
Calling them useless may be a bit of a stretch. If you said "mediocre", I'd believe it. Also, in my experience there, the ratio is not uniform - you could go to a department with lots of top notch folks, and then to another department with everyone being mediocre.
Some data points:
Asked a person in a SW interview to write a function to calculate the factorial. The candidate was repeatedly told he could use whatever language he wants. He insisted on using a language he had little experience in (C++), and not the one he was experienced in (Python). He was given latitude to use any method he wanted: Recursive, iterative, etc.
He didn't even come close to solving it.
He was still hired.
He is not an outlier when it comes to SW folks at Intel. Yes, Intel definitely has some really good SW engineers, but the 1:40 ratio easily applies in that sphere.
Up till 2014, my department insisted on using cvs for version control. They said they saw no benefit to git (or even SVN).
In another team I was in, we were stuck with SVN. The senior manager wouldn't allow us to use git because he was sure it would be over the heads of many of the employees (sadly, he may have been right). Then in 2017, when IT announced that SVN was being EOL, to avoid git they switched to MS TFS (even though TFS themselves recommend git!) I still hear from folks there about the resistance to switch to git. And the following comment I've heard from multiple folks:
"I don't want to use git because it comes from Microsoft!" (conflating github with git, and upon further discussion realizing they have no idea MS bought Github - they think Git and Github both originated from MS).
BTW, I'm not a git fan boy - I much prefer mercurial.
If you're good at SW and somewhat up to speed with current technology, Intel is a very frustrating place to work.
But then again, their compensation for SW is simply not competitive:
https://www.levels.fyi/company/Intel/salaries/Software-Engin...
Intel doesn't have as many of those great programmers as I would like, but I think their bigger problems are organization rather than individual. There just isn't enough high-level coordination and sharing of information. Every group just kind of does their own thing, and if you want to know how something works you have to know who to ask. Too much tribal knowledge (which often has a short expiration date) and too little writing stuff down in one place where it's easy to find.
It's funny you mention git, because I joined Intel from a startup that used SVN and that was the one bright spot about Intel's technical culture that they used git pretty extensively. I don't think I've ever used a non-git source repo at Intel. I assume that there was some major internal struggle to get to that point, but it was before my time.
While people would likely disagree on the exact ratio in a given company, it's far worse than you would ever imagine. The inevitable eventually happens: they cease being able to make (or make very well) the very product(s) they're known for. Occasionally companies will get shocked back to life as they see it all slipping away. More commonly they just wither and die as they are unable or unwilling to change the culture that got them to that point.
Do I believe 1:40? If anything, I think that might be a conservative number at this point. Intel has become tragically/comically incompetent.
I can't think of another company like that, maybe Microsoft years ago or GE. Is this something I'm imagining or is that a frequent topic of conversation at all levels of Intel?
Intel will shift to a TSMC model. They have the best fabs on the planet, and the best fab engineers. I believe it is something like a 3 millions dollars lost per minute if they are idle. They have already started doing that a few years ago, I suspect this will be their final form.
IMHO: The only thing holding them back from the transition are the hundreds of small boondoggle-groups staffed by old-timers too scared to retire, and too scared to do something daring, yet still somehow hang on to their hidey-holes. They lost a ton of key architects to Apple a few years ago, which I also suspect was the reason why the M1 is so badass.
...and if you really want to get sentimental, here's an AMD poster we had in our cubes back in ~1991:
https://linustechtips.com/uploads/monthly_2016_03/Szg2Ppo.jp...
The Sanders-as-Indiana was both funny and infuriating....
(The Farrah Fawcett looking woman was Sander's bombshell wife, compared to Grove at the time, who drove a beat up old car.)
I think fabbing for outsiders would probably be a good idea, but splitting the fabs out from the designers seems like a bad idea.
Even if they axed all process R&D and returned the cash to shareholders, due to the eye watering costs of designing at 10nm and lower I expect there will be a lot of business to keep their fabs turning over for the next decade.
https://www.intel.com/content/www/us/en/foundry/overview.htm...
In late 2018, it was rumored they were exiting the business because of low uptake and because of constrained supply of their leading edge process, but it doesn't look like that happened.
The biggest thing Intel didn't do with Itanium was release affordable AT/ATX form factor boards. They priced it way, way too high, chasing early margins in "enterprise" without realizing that market share is everything in CPUs. This is the same mistake that Sun, DEC, and HP made with their server/workstation CPUs in the previous era. With a new architecture you've got to push hard for market share, wide support, and scale.
If I'd been in charge I would have priced the first iterations of Itanium only a little above fab cost and invested a lot more in compiler support and software.
Edit: also whatever happened to the Mill? The idea sounded tenable but I have to admit that I am not a CPU engineer so my armchair take is dubious.
Anyway the ship has sailed. Later research in out of order execution has yielded at least similar performance gains and post-X86 the momentum is behind ARM and RISC-V.
One good side of x86 style instruction sets is that there is a lot you can do in the cpu to optimize existing programs more. While some really advanced compiler optimizations may make some use of the internal details of the implementation to choose what sequence to output, these details are not part of the ISA, and thus you can change them without breaking backwards compatibility. Changing them could slow down certain code optimized with those details in mind, but the code will still function. And I'm not even talking just about things like out of order execution. Some ISA's leak enough details that just moving from multi-cycle in-order execution to pipelined execution was awkward.
This ability of the implementation to abstract away from the ISA is very handy. And some RISC processors that exposed implementation details like branch-delay slots ended up learning this lesson the hard way. Now the Itanium ISA does largely avoid leaking the implementation details like number of scalar execution units or similar, but it's design does make certain kinds of potential chip-side optimizations more complicated.
In the Itanium ISA the compiler can specify groups of instructions that can run in parallel, specify speculative and advanced loads, and set up loop pipelining. But this is still more limited than what x86 cores can do behind the scenes. For an Itanium style design, adding new types of optimizations generally requires new instructions and teaching the compilers how to use them, since many potential optimizations could only be added to the chip if you add back the very circuitry that you were trying to remove by placing the burden on the compiler.
Even some of the types of optimizations Itanium compilers can do that mimic optimizations x86 processors do behind the scenes can result in needing to write additional code, reducing the effectiveness of the instruction cache. This is not surprising. The benefits of static scheduling are that you pre-compute things that are possible to pre-compute like which instructions can run in parallel, and where you can speculate etc. And thus you don't need to compute that stuff on-die, and don't need to compute it each and every time you run a code fragment. But obviously that information still needs to make it to the CPU, so you are trading that runtime computation for additional instruction storage cost. (I won't deny that the result could still end up more I-cache efficient than x86 is, because x86 is not by any means the most efficient instruction encoding, especially since some rarely used anymore opcodes hog some prime encoding real-estate.)
Basically I'm not sold on static scheduling for high performance but general purpose CPUs, and am especially not sold on the sort of peudo-static scheduling used by Itanium where you are scheduling for instructions with unknown latency, that can differ from model to model. The complete static scheduling where you must target the exact CPU you will run on, and thus know all the timings (like the Mill promised) feels better to me. (But I'm not entirely sure about install type specialization like they mention.)
But I'm also no expert on CPU design, a hobbyist at best.
I think that's overestimating, although you are right it is damn expensive.
Order of magnitude I would say it's more like: Fab has lifetime of 2-3(?) years, and costs $10B to build and amortize. So every minute of idled factory capital = $6,000 in pure cost of the facility.
(although, if you think of it in potential lost revenue terms, then you may be more correct.)
(Actually we can also do that arithmetic, according to https://www.forbes.com/sites/willyshih/2020/05/15/tsmcs-anno...
240,000 wafers per year * 500 chips per wafer * $100? = $12B per year revenue. Also is something like $22k per minute)
Many Intel fabs are from the early 2000s. Some are from the 90s. What do you mean 2-3 years?
https://en.m.wikipedia.org/wiki/List_of_Intel_manufacturing_...
>> Many Intel fabs are from the early 2000s. Some are from the 90s. What do you mean 2-3 years?
> I think they're referring to each line as a new fab, rather than the complex of buildings they're in, which is pretty fair.
Even so, wouldn't the 2-3 year number only apply to CPUs? Couldn't they keep the lines running making things that don't require top of the line processes, like USB controllers, etc.
Of course the lines stay open after that for cases you've laid out, but there's a lot of financial reasons why that needs to be basically gravy train money at that point with the lines being fully amortized and having paid for themselves many times over.
I am no expert but I didn't think the site would be just accumulating its n-1, n-2 generation lines to produce lower grade stuff in the same fab walls. They take up space that they want to continue producing top of line latest output. I thought.
Adding on to that, in a lot of ways the fab is the building, and at a bare minimum you'd be resetting a lot of the yield issues you fixed by moving it to a new space. So then you'd be left with an old node and yield issues, and what's the point of that?
There's also a ton of trade secret style IP still in the old nodes that could help a competitor, so why let it leave your property? Intel for instance is so into preventing trade market theft on the specifics of their nodes that they (at least used to) manually drill out every mobile device camera allowed in their fabs, in addition to the normal "no outside electronics past this point" security stations.
That's one of the biggest reasons to choose them. If Intel's "x86" stops being x86, they'll lose one of their biggest competitive advantages.
See the first comment here, for example: http://www.os2museum.com/wp/vme-broken-on-amd-ryzen/
Microsoft even had a version of NT3.5 for The Itanic. It seemed we were just about to achieve critical mass in the server world to switch to a new architecture with huge address space, ECC up the wazoo and massive integer performance.
Then the PC revolution took off with Win95, and the second war with AMD happened (and NexGen sorta). This couldn't be solved with legal battles. This put all hands on deck because there was SO much money to be made with x86 compatibility. The presence of x86 "up and down" AMD & Intel's roadmap took over the server market as well: it was x86 all over the place.
And that, chil'ren, is why x86 was reborn in the 90's just as it was close to being wiped out.
Now Apple has proven you can you seamless sneak in a brand-new architecture, get hyooj gainz, and we are none the wiser. This is fantastic news. I think we are truly on the cusp of x86 losing its dominance in the consumer space after almost 35 years of dominance.
Lmao, never heard that one before
The fact that they're almost equivalent to those two at 5nm when Intel is at 14nm+++++++++ shows how killer the RTL and tape out folk are at Intel.
The real fight has always been at the fab layer.
Only when it comes to single core. And even then Intel has to spend 2x silicon - Zen 3 CCD at 7nm with a die size of 80.7 mm² while Intel's 10th gen 8C die is 170mm2. Of course it isn't silicon itself which is costly, it is the processing of each wafer what is the cost here, and getting 2x less cores per wafer is bound to affect the margin/profit/etc.
And of course Intel has no chances of putting out 16C chip into normal desktop like AMD does with 3950x and 5950x (those chips successfully go into the territory of 2x Xeon workstations costing arm and leg)
Do keep in mind that a fair amount of a desktop chip is pad space, IO hasn't shrunk since 22nm nodes, DUV steppers are dirt cheap, Intel's foundries are fully amortized, and yields at 14nm are going to be approaching 100%.
Intel has really stumbled on fabrication the last 2-3 years, but the idea that they won't correct that ignores decades of precedent.
I'm an AMD fanboy from the K6 days, and I love what Lisa Su has been able to deliver for them, but if it wasn't for Intel's stumbles, they would still be single digits at best in most markets.
Look at their progress from Pentium 4 to Core 2 Duo (makes me feel real old to see that was 15 years ago). We probably won't see quite as dramatic an improvement over the next 3 years, but I'll be stunned if AMD still has the performance lead at that point.
Apple's a wildcard, but they aren't going to sell their chips, and they represent a fairly small part of the PC market. No other non-x86 chips have made even a tiny dent in the PC market, and even in the datacenter, it's still mostly x86.
The absolute worst I can imagine happening to Intel over the next decade is that they no longer define the CPU market. But that's a long, long, way from dying, and will probably be good for the market and good for Intel in the long run.
You guys remember the blue man group? Good days.
We're nearing the end of the road on node shrinks, Intel can best hope to be marginally ahead on process.
What Apple's M1 shows is that a fundamental redesign is needed to get gains. Intel's problems appear to be in design, software, and business execution.
But doing a new ISA or muscling into ARM or other radical things precisely require design/software/business execution or it will utterly fail.
M1 is going to open the floodgates for ARM on PCs. That's what will kill Intel.
AMD can probably pivot, I have faith in their designers.
Now, Chromebooks are something of an argument for ARM on the desktop, as they at least seem to provide identical behavior regardless of ISA, but they aren't putting up the necessary numbers quite yet. x86 still owns PC gaming, and has a pretty solid hold on console gaming. I'm not going to be worried at all for Intel until that shifts too.
Intel is dying because they should have entered the foundry business in the early 90's. Process was always their forte, they were never good at the CPU architecture design anyway as iAPX/i860/Itanium amply demonstrated.
Instead they blew $100B getting into antivirus, network security, mobileye, infineon, and a dozen other failed businesses, $50B in illegal kickbacks keeping AMD out of the unprofitable low end laptop market, and another $50B of "contra revenue" subsidizing their inferior me-too products against mobile SOCs they should have been fabbing in the first place.
Otellini on the iPhone:
"We ended up not winning it or passing on it, depending on how you want to view it."
"At the end of the day, there was a chip that they were interested in that they wanted to pay a certain price for and not a nickel more and that price was below our forecasted cost. I couldn't see it. It wasn't one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100x what anyone thought."
And in hindsight, he was a moron, unqualified to lead an engineering company.
When people point to MBAs ruining Intel. Point to that statement.
You could invest in emerging markets.
Or rearrange financial statements deck chairs to hit your stock options and roll your eyes at the nerds.
Traditional wisdom used to be that if you wanted cutting edge performance, you really needed your own cutting edge in-house fab. The vertical integration advantages of tuning your ASIC design to your process were unassailable. If you couldn't afford your own foundry, tough luck, you suffered a severe penalty in performance and/or cost.
But the merchant foundries, first and foremost TSMC, have really turned that traditional wisdom around. Both in terms of matching, and then surpassing Intel on deploying the latest process nodes, but also in terms of making it possible for 3rd party ASIC developers to use that process as efficiently as an in-house design team would be able to.
Intel is a large company with a massive market cap. The bigger they are, the longer they fall. They just have so much runway and market gravity.
For comparison, Apple shot itself in the foot repeatedly for at least 10 years before Steve Jobs rejoined, and it took 10 more years after that before the iPhone came out, which no layman could have predicted.
The fact that Intel's management have chosen to spend two decades flushing that amazing free cashflow down the M&A toilet instead of fixing their workplace culture and getting their design house in order is what Pat Gelsinger needs to fix.
Intel is definitely in a hole (that they watched being dug below them and did nothing...), but they can get out. They have plenty of time to do so. The only real problems to overcome are their own internal culture and the sheer difficulty of single-digit-nanometer manufacturing. And the second one is easier.
Don't count them out yet. (But do laugh at them. Because they definitely deserve it, and it just might motivate them.)
Look at IBM; while Big Blue was once seemingly omnipotent and they've since fallen from that position, it's still a double-digit billion company with over 300k employees. It's not as big (relatively speaking) as it once was, but it sure didn't die either.
Other companies have died, such as Digital Equipment, but that was the result of 1) a massive paradigm shift in computing and computing equipment (introduction of "personal computing"), and 2) some pretty bad business decisions (not just failing to get a hold of the PC market). Thus far, I'm not really seeing 1) happen, and 2) alone probably won't be enough to completely kill intel, just cripple it temporarily.
Personally I think strategic M&A is a far more intelligent use of cash and credit than share buybacks and dividends.
Intel has arguably made at least a few solid acquisitions. AMD is buying Xilinx a few years after Intel bought Altera, which tells us that the Altera acquisition couldn't have been totally misplaced -- AMD has the benefit of hindsight with which to evaluate its own acquisition. Mobileye seems to be well positioned as conventional automakers are beginning to offer mainstream EVs.
Perhaps the problem is that there are certain higher-order functions (M&A, stock buybacks) that should generally be conducted only when the lower-order functions (core business) are firing on all cylinders. Companies have finite resources, and executives have limited time and energy, that shouldn't be starved away from the core business unless absolutely necessary.
Mobile? No, thanks, those are peanuts. Cheap/home PCs? Nah, either. PC builders very sensitive to prices? The don't want them. And slowly they are being sweetly cornered in the top market of the very expensive servers with very juicy margins, but feeling how the small dogs keep bitting up. Suddenly ARM wants to build for cheap servers, and AMD has a chip that also fights for that top market.
Obviously process nodes are much harder than pretty much anything in the cloud, but this is a serious test of (pure and simple) whether their leadership has the (gender-neutral) balls to not back down.
I'm not entirely sure that is true. TSMC plays fast and loose with their definition of node. Any fab folks want to speak up?
What I mean by that:
https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-n...
"And also goes some way to explaining why, despite TSMC offering a nominally 7nm process, the general consensus has been that Intel’s 10nm design is pretty much analogous. But what’s 3nm between fabs? At that level, probably quite a lot. But if the 7nm node is more of a branding exercise than genuinely denoting the physical properties of that production process then you can understand why there’s supposedly not a lot in it."
Look it up yourself. Intel being on 14nm vs (say) 5nm is actually quite flattering to Intel.
EDIT: All of this stuff is usually stated at ISSCC every time a new process is announced, so it isn't NDA. I haven't followed this in years which is why I was asking for a process person to step in.
3nm, sure, good luck.
Tsmc 5nm is 27/54/36 with a 6T and 7.T option. More open design rules. HD SRAM cell is about 0.025um2.
This is almost exactly what you expect for 1 node of litho scaling.
The more difficult part is that the designs and related process optimizations for intel are different.
More or less TSMC 7nm := Intel 10nm, and TSMC 5nm := Intel 7nm. It's more complex than that, one has denser logic while the other has denser SRAM and what have you, but it's a good baseline.
Since Intel is struggling with 10nm but is shipping, that puts TSMC about a node and a half ahead.
If you want to dig in deeper, wikichip has most of the public specifics on the process nodes (which are heavily shrouded in secrecy). For instance: https://en.wikichip.org/wiki/7_nm_lithography_process
No they aren't. The nanometer names are only kinda comparable within a single company, but they are not comparable beyond that.
It's like saying that Chrome is 6x further ahead than Safari because Chrome is on 87 and Safari is only on 14. Same thing with fabs here, it's just a version number that counts down and has been for a long while.
In this case specifically, Intel's 10nm is more or less the same density as TSMC's 7nm. TSMC's 5nm is then a significant 1.8x the density of their 7nm process, so right now Intel's fabs are ~identical to what AMD is using (7nm), but Apple's use of TSMC's 5nm then puts them a generation ahead. TSMC is leading, but by a single generation not by 3-5x or whatever. Or to put it in year terms TSMC is about 2 years ahead of Intel right now, Intel just now getting "volume" 10nm up & running while TSMC was at a comparable node in volume 2 years ago.
As far as "performance" goes, though, so far TSMC, even their 5nm process, cannot quite match Intel. Remember that as far as the fab goes, performance is just what clock speeds can be hit, and Intel's clocks are the highest - they have the fastest transistors. 14nm+++++ is a meme/joke, but the actual transistor speed of it is still downright impressive. How fast a CPU actually runs depends on the architecture, though, which is influenced by density but in terms of pure single-core performance that's really almost entirely just architecture.
TSMC are shipping chips at something like 170MTr/mm^2, the bulk of Intel's production (even their new chips) are still closer to 40-50MTr/mm^2
And TSMC's 170MTr/mm^2 is still restricted to a single vendor and hasn't ramped up beyond that yet. You can't simultaneously claim Intel is restricted to what their bulk volume lineup is and then use TSMC's non-volume lineup. The bulk of TSMC is still 7nm which is 100MTr/mm^2.
My understanding is that Intel's fast transistors come at a cost of not being particularly dense for the process node they are implemented on. So it's a question of design philosophy as well. Intel has fast, but somewhat sparse transistors, they use a lot of hand layout etc. to really optimize their designs so they can clock high.
Other players use a different philosophy. Slower, but denser transistors and more automated layout may not clock as high, but the improved density gives you more cores (e.g. AMD vs Intel) and/or allows a more "brainiac" core design (wider, deeper ROB, better branch prediction, more cache, etc. like Apple M1), and also gets you a better turnaround time.
Not in the enterprise market where AMD is almost completely absent so far.
A quick search showed that in 2017, AMD server market share was 1%, and in 2020 is expected to have reached nearly 10%[1] in 2019. This is still not the same as 15 years ago, but the market is also not the same, with cloud computing, and this increase is very expressive.
They are bringing IPC improvement with Epyc Milan and is pushing the same innovations towards the server market.
[1] https://www.hardwaretimes.com/amd-server-market-share-grows-...
Investing into increasing their fab capacity is the easiest business decision they'll ever make. And if they need finance they could get investors lining up around the block throwing money their way.
https://www.tomshardware.com/news/amd-chip-shortage-packagin...
Now, let's take a sledgehammer to some other moats and make the rent seekers work a little for their profits.
Things change fast.
VLIW had a lot of hype. DSPs had already proven it, and Itanium, Transmeta, and various embedded processors were using it.
As a mostly Mac-user back then, I liked PowerPC, and POWER3 was leading the way to 64-bit, but I wasn’t expecting PCs to switch to PowerPC. The RISC systems had not all been executed in the face of the not-yet-released threat of Itanium. I was expecting the diversity to continue, for cross-architecture programming to be the rule and not the exception.
It almost seems like take a conspiracy or sabotage for Intel to fumble such a large lead in process. But all it takes it self-sabotage in forced retirements and financial engineering ahead of innovation. America really needs it's CEOs to re-learn the lessons of the global behemoths of manufacturing we used to have- that re-investing in your own future growth is just as or more important than growing your dividends.
That's what they said about Microsoft but it was able to reinvent itself and get back in the game with new leadership.
Intel certainly had its share of problems in the recent years but has plenty of cash in hand, I imagine it can do the same if it really tries.
I really couldn't care less what architecture my chips are, or who fabricates them. All I care about is price and performance.
I write application software, the underlying hardware is pretty irrelevant to me and most of my peers too.
I think that as we lost national sources of pride and community structures at a local level, people transferred their loyalties to brands. LETS RISE ABOVE!
Fully agree. This is a phenomenon that goes beyond tech.
Personally I have always rooted for AMD as the underdog and because Intel was the Goliath that held back CPU performance so it could squeeze more dollars out of consumers.
This feels like (repeat popular opinion) retreading.
Money talks, bullshit walks.
Intel stocks show Intel is doing fine. Which means people who are accountable for their predictions, do not consider Intel dying.
Every CPU should support multiprocessor configurations, every CPU should support ECC memory and VT-x, upgradability should be maximized (so you wouldn't necessarily need a new mainboard for a new CPU), the socket and the instruction set should be licensed to as many competitors as possible. Like in good old 90ies.
Intel contributed too much. The world owes them a lot. Thanks for contributions.
Intel is dying is totally ok, it will open up more competition and innovation.
This is all the result of Intel fab failings. They have the talent to compete, dunno if the leadership wants to take to it.
The CPU design business is just a tiny part, most of the investment goes into fabs and process.
But for casuals, the PlayStation 4, PlayStation 5, Xbox One series, and Xbox Series X all use AMD APUs.
Also if you were not playing at 4K or competitive e-sports, you probably wanted an 8-core Ryzen chip for multi-threaded tasks even if it was marginally worse at CS:GO with >200 fps.
Could they collaborate with Dell or Lenovo and make RISC-V based SoCs to go into laptops to compete with the macs?
as RISC-V CPUs are rolling out in hundred PCs now
CPUs that are fundamentally insecure (meltdown and all the subsequent microcode level vulnerabilities). Mandatory backdoor called ME. We would not miss a lot.
But where is the competitor that will replace it? Actually healthily competing competitors, not just one would be ideal.
Even Boeing has a stronger competitor.
Not that they’re entirely without merit; Intel faces challenges that, left resolved, will spell the end of the company. Just like any other firm that ignores problems.
Most of these arguments, especially the ones of “what are they going to do about AMD?” seem short-sighted. Intel has been here plenty of times before, facing down strong AMD competition only to come back and crush them a few product cycles later. Jim Keller can only come around so often. No, AMD’s issue is that they can never hold onto that lead through enough product cycles to really make a lasting dent in market share. It’s especially unfortunate for AMD right now, as supply constraints are choking what is its best performing product in years. Can’t really claw back a ton of market share when no one can get your chips.
It’s asked by the tech tubers and other observers why Intel continues to ship old Skylake variants. Process issues with 10nm are a factor here. But I suspect there’s a big one that no one’s really discussing.
And that’s the fallout from Spectre and Meltdown.
Keep in mind that the exploits in these flaws were found in products that were 10 years back at the time they were discovered. Each new chip isn’t a ground up design, but iterations on designs that stretch back years. I suspect, that the reason we haven’t seen truly new designs in over several product cycles is that Intel is having to redo large parts of their architecture that would have otherwise been carried forward in a normal setting. Gamers and enthusiasts may not have seen much of an impact to their apps as a result of these; none of the microcode updates results in big FPS drops in any games that I am aware of. But there were notable enterprise workloads that did. It’s often overlooked that the consumer marketplace is just one part of the larger IT ecosystem, so a redesign there is a must.
The timing of this is horrible of course, as it comes right when their new process is failing. So double-whammy there. Fortunately, or unfortunately, worldwide lack of fab capacity I think is preventing their rivals from exploiting this as much as they could. As for RISC, I think that is an entirely different threat, and to me, far more threatening and plausible. Intel saw this and tried to get into mobile chips, But I feel x86 is intrinsically at a disadvantage there.
I’m interested to see where Intel will be at once they have a new design ready to go. We’ve been here before. Pentium 4 ended up scaling far better than the Athlon XP did, and Core 2 and Nehalem utterly crushed AMD’s offerings when they were launched after a few years of AMD on top. There’s good reason to believe they can do so again, and a big reason why we haven’t seen an attempt is because they’re continuing to undo their mistakes of the past. We’ll have to see if they can.
My two cents.
(Disclaimer: Not a fanboy, I’m writing this on my AMD 3700X. =P )
Well, no one other than Taiwan I guess.
But given how messed up & fragmented the EU military is if the US stops military support for Taiwan china will likely directly declare War and Attack. (To be precise there is no EU military, just that of all members which makes it fragmented and in many cases is their military isn't in the best condition. Also China wouldn't declare War as it doesn't recognizance Taiwan, i.e. they would instead "use military force to attack rebels and terrorists" or something like that).
That's a temporary advantage for Taiwan, and pretty much their only one.
A silicon atom is only around 0.1nm wide, so the end of scaling is coming up like a brick wall. We've got maybe a half dozen nodes left.
After that, industrial espionage, pumping up SMIC or some other new Chinese domestic equivalent at the time, and literally just leaking info publicly so the EU, US, and ROK are viable replacements once chips become truly a commodity will cut the dependence on Taiwan. At that point it's not worth facing China's nukes to tell them that they can't have Crimea^H^H^H^H^H^H Taiwan.
That's still something like a hundred thousand to a million times smaller that what we can actually fabricate, right?
That being said the future never ceases to surprise.
This is the American dream.
But no, China is not that stupid. Sure, they want Taiwan back, but they will not declare war and create global instability right in their backyard. Only stupid people would do this, and the Chinese are not stupid. They are in a marathon. They see the goal in 30, 50, 100 years. They will not give this gift to the American government.
Never underestimate emotional motivations.
P.S., not saying that China will attack Taiwan, just not ruling it out.
Removes the option of a painless transfer of power so that the US is still geopolitically interested in a independent Taiwan.
A better idea is to give a US passport to every TSMC employee NOW (exempt them from having to hold a green card and live in USA for 5 years first). And change the law so that US citizens residing overseas with no US-sourced income are exempted from filing and paying taxes.
Other countries that wish to have leading edge commercial fabs, e.g. Germany or Netherlands, should compete with USA and try to get their passports to TSMC employees first. Tell them your country does not tax citizens on overseas income, and like Taiwan, has universal healthcare, high-speed rail, and gun control. Netherlands can also advertise that TSMC employees likely already have some Dutch friends at ASML.
https://asia.nikkei.com/Business/China-tech/China-hires-over...
I'm sure not all engineers at TSMC are willing to relocate from Taiwan to China but pretty much everything is for sale if the price is high enough.
Setting up a new factory to replace TSMC is not trivial of course, as it's operationally hard, but none of the core tech would be "in China's hands".