One thing that may save Intel’s bacon big time is the increase of defense budgets worldwide as many countries go into a rearmament drive. All forms of weaponry today require chips, lots and lots of them.
Intel doesn't make many defense chips nor do they plan on as the defense industry is low volume and heavily bureocratic and doesn't generate the kind of margins Intel is interested in.
That's why there are very few fabs in the US certified for defense contracts. Because it's not a profitable business to be in for a fab if you've got better options.
For Intel to want to get in on this, uncle Sam would have to open his wallet far far wider than he does right now.
You can spin up everything in tech in no time by just spending money unless it's a 3nm chip fab.
However, fabricating chips is a business that everyone wants to outsource. It's work with nasty hazardous chemicals. Development and research are tricky and expensive and require a lot of physical grunt work. A lot of very smart engineers have to interface with technicians performing tedious and dangerous work that they have to not screw up. It's also one of the most important businesses in the world.
Intel has to somehow make the culture and engineering work so that they can perform the physics, chemistry, manufacturing and material science to make a new chip fab. This is what they've screwed up so far, and what they'll have to make work if they want to get back in the game. Can they pull it off?
Well, Raptor Lake processors released recently are quite good, and it seems that Intel's EUV manufacturing processes will be ready this year. I think Intel has a chance to catch up with, or even surpass TSMC in the next couple years.
Not entirely sure what is new here. And dont agree Intel ever left. The likely hood of current trend projection and inferring from publicly available information, Intel will regain their lead by 2025. Assuming Pat execute on its plan, which they seems to be delivering it on time if not ahead of schedule. It will however very unlikely to be great margin competing against TSMC even with subsidies. Most ( cough Peter M... cough ) analyst seems to be thinking either TSMC is doomed because they dont know how to play the political game of subsidy ( Not True ) or Intel is doomed because Intel will never catch up ( also not true ).
Intel does not appear to have any serious problem in their capabilities of logic design (even if e.g. the launch of Sapphire Rapids had multiple delays due to design bugs and insufficient validation before the tape-outs), but until now there has been no evidence of Intel recovering their great CMOS manufacturing process disadvantage vs. TSMC.
The Intel Sapphire Rapids CPUs would have been superior to the AMD Genoa, except that the CMOS process used by Intel to make them provides such functional dependencies between the clock frequency and the power-supply voltage and between the power consumption and the power-supply voltage, that whenever the CPU in in a state where the clock frequency is limited by the power consumption, i.e. whenever it runs a multi-threaded application, the clock frequency of a Sapphire Rapids CPU is ridiculously low in comparison with that of an AMD CPU made by TSMC.
This leads to low performance in multi-threaded applications. Moreover, the Intel CMOS process has low yields, which forces them to make only Sapphire Rapids CPUs with fewer cores, smaller cache memories and higher prices than AMD Genoa.
As long as Intel remains handicapped by their inferior manufacturing processes, they will not be able to make competitive products.
At the end of this year, Intel is expected to launch Meteor Lake the first product made (partially, the integrated GPU will be made by TSMC) using the next Intel manufacturing process.
This is their chance to regain some of the lost ground. It remains to be seen if they will succeed.
Until now, the next Intel manufacturing process (Intel 4) does not seem to be in very good shape, because Intel had to cancel the desktop variant of Meteor Lake and replace in with a "Raptor Lake Refresh", presumably because "Intel 4" does not provide yet high enough clock frequencies, exactly like Intel had to replace for desktops Ice Lake with Comet Lake and Tiger Lake with Rocket Lake.
Because Meteor Lake is expected to have a greater IPC than Raptor Lake, reducing the gap towards the Apple CPUs, any regression in performance can be caused only by a clock frequency during a power-limited scenario that is lower than estimated when the Meteor Lake design has started, which is the same what has happened with the Intel "10 nm" process during the Cannon Lake/Ice Lake/Tiger Lake years, until Intel eventually succeeded to tweak their manufacturing process and design rules enough to be able to obtain a better performance than with their old process.
This means that Intel 4 is still not moving forward well enough, at least not as well as they had predicted at the start of the Meteor Lake project.
Hopefully for a healthy competition, this time they will succeed much faster to tune their manufacturing process.
Intel has had process problems. I don't think those continue to be their primary issues.
It wasn't process problems that kept sapphire rapids from being released till now. 10nm/Intel 7 has been ramped up and fabbing products for over a year.
They are very late on critical products that are clearly not the fault of the foundry side. This is squarely on the design side.
Running good engineering orgs is hard, Intel's problems run deep and on many levels.
The original Sapphire had to be "redesigned" ( or reworked ) for Intel 7. You cant just take an high yield Intel 7 Golden Cove Core and "Edit" it with some extra features for server and then expect it yield.
Second being Intel 7 isn't exactly high yield either. At least not 14nm level. It is good enough for small die size chips, but it is something different for large die size ( Aka Sapphire Rapid ). Which is why the comparison between Intel and AMD on server market doesn't make sense. They have a very different economy of scale.
Third being Intel doesn't put resources in Intel 7 any more. It is no longer a priority. They are losing one way or another that is why they put everything into their next node Intel 4.
Intel is expecting Intel 3 in 2023 and products out in 2024, this isn't too far behind TSMC schedule.
> The original Sapphire had to be "redesigned" ( or reworked ) for Intel 7. You cant just take an high yield Intel 7 Golden Cove Core and "Edit" it with some extra features for server and then expect it yield.
All the other products that shipped on Intel 7 had to also be designed for the process. Intel was also late on arc, which wasn't even fabbed on Intel process nodes. Ponte Vecchio... Late late late.
One could've asked the same question 10 years ago. If you did, how's the last decade look, for Intel? Not bright.
Better question: "Should Intel come back?" I think the world going to be better off once the monolithic block of Intel is no longer standing in the way of the flood of innovations they've successfully suppressed.
They did some really cool stuff once, but that company is gone and what replaced it is a rent seeking monster.
Against whom? Classic desktop builds don't matter anymore, PC sales never fell so low. Not even the performance is what's needed. It's all perf/watt. x86 - which is all they have, beside some side hustles and cemented deals - is fading fast.
This is what I struggle with. The most important things to me are performance and compatibility. Performance/watt means almost nothing to me if I get 80% of the performance and can only run a limited catalog.
Yes, this will change over time. Unless there's a compatibility layer that's perfect though, I don't see this whole "x86 will be dead in 5 years" argument. That's what they said almost 15 years ago.
Intel arent generating enough cash for a massive burst of investment.They're losing ground to AMD on the desktop and are losing marketshare to AMD and ARM architectures in the server space.
People are genuinely excited when AMD introduces a new processor but there's crickets and tumbleweed when its Intel.
Where they were once the leading foundry, they cant even get past 10mm whilst TSMC and heading towards 3mm.
23 comments
[ 340 ms ] story [ 2115 ms ] threadThat's why there are very few fabs in the US certified for defense contracts. Because it's not a profitable business to be in for a fab if you've got better options.
For Intel to want to get in on this, uncle Sam would have to open his wallet far far wider than he does right now.
However, fabricating chips is a business that everyone wants to outsource. It's work with nasty hazardous chemicals. Development and research are tricky and expensive and require a lot of physical grunt work. A lot of very smart engineers have to interface with technicians performing tedious and dangerous work that they have to not screw up. It's also one of the most important businesses in the world.
Intel has to somehow make the culture and engineering work so that they can perform the physics, chemistry, manufacturing and material science to make a new chip fab. This is what they've screwed up so far, and what they'll have to make work if they want to get back in the game. Can they pull it off?
Not entirely sure what is new here. And dont agree Intel ever left. The likely hood of current trend projection and inferring from publicly available information, Intel will regain their lead by 2025. Assuming Pat execute on its plan, which they seems to be delivering it on time if not ahead of schedule. It will however very unlikely to be great margin competing against TSMC even with subsidies. Most ( cough Peter M... cough ) analyst seems to be thinking either TSMC is doomed because they dont know how to play the political game of subsidy ( Not True ) or Intel is doomed because Intel will never catch up ( also not true ).
The Intel Sapphire Rapids CPUs would have been superior to the AMD Genoa, except that the CMOS process used by Intel to make them provides such functional dependencies between the clock frequency and the power-supply voltage and between the power consumption and the power-supply voltage, that whenever the CPU in in a state where the clock frequency is limited by the power consumption, i.e. whenever it runs a multi-threaded application, the clock frequency of a Sapphire Rapids CPU is ridiculously low in comparison with that of an AMD CPU made by TSMC.
This leads to low performance in multi-threaded applications. Moreover, the Intel CMOS process has low yields, which forces them to make only Sapphire Rapids CPUs with fewer cores, smaller cache memories and higher prices than AMD Genoa.
As long as Intel remains handicapped by their inferior manufacturing processes, they will not be able to make competitive products.
At the end of this year, Intel is expected to launch Meteor Lake the first product made (partially, the integrated GPU will be made by TSMC) using the next Intel manufacturing process.
This is their chance to regain some of the lost ground. It remains to be seen if they will succeed.
Until now, the next Intel manufacturing process (Intel 4) does not seem to be in very good shape, because Intel had to cancel the desktop variant of Meteor Lake and replace in with a "Raptor Lake Refresh", presumably because "Intel 4" does not provide yet high enough clock frequencies, exactly like Intel had to replace for desktops Ice Lake with Comet Lake and Tiger Lake with Rocket Lake.
Meteor lake is a big step up in efficiency but will be a regression in performance. Another reason to cancel the desktop variant.
This means that Intel 4 is still not moving forward well enough, at least not as well as they had predicted at the start of the Meteor Lake project.
Hopefully for a healthy competition, this time they will succeed much faster to tune their manufacturing process.
It wasn't process problems that kept sapphire rapids from being released till now. 10nm/Intel 7 has been ramped up and fabbing products for over a year.
They are very late on critical products that are clearly not the fault of the foundry side. This is squarely on the design side.
Running good engineering orgs is hard, Intel's problems run deep and on many levels.
The original Sapphire had to be "redesigned" ( or reworked ) for Intel 7. You cant just take an high yield Intel 7 Golden Cove Core and "Edit" it with some extra features for server and then expect it yield.
Second being Intel 7 isn't exactly high yield either. At least not 14nm level. It is good enough for small die size chips, but it is something different for large die size ( Aka Sapphire Rapid ). Which is why the comparison between Intel and AMD on server market doesn't make sense. They have a very different economy of scale.
Third being Intel doesn't put resources in Intel 7 any more. It is no longer a priority. They are losing one way or another that is why they put everything into their next node Intel 4.
Intel is expecting Intel 3 in 2023 and products out in 2024, this isn't too far behind TSMC schedule.
All the other products that shipped on Intel 7 had to also be designed for the process. Intel was also late on arc, which wasn't even fabbed on Intel process nodes. Ponte Vecchio... Late late late.
Better question: "Should Intel come back?" I think the world going to be better off once the monolithic block of Intel is no longer standing in the way of the flood of innovations they've successfully suppressed.
They did some really cool stuff once, but that company is gone and what replaced it is a rent seeking monster.
Yes, this will change over time. Unless there's a compatibility layer that's perfect though, I don't see this whole "x86 will be dead in 5 years" argument. That's what they said almost 15 years ago.
People are genuinely excited when AMD introduces a new processor but there's crickets and tumbleweed when its Intel.
Where they were once the leading foundry, they cant even get past 10mm whilst TSMC and heading towards 3mm.
Intel look like a company in terminal decline.