Can Intel Come Back?
ft.com
ft.com
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?
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.
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.
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.