Can anyone explain what evidence there is that Intel will be able to deliver all this?
Not taking a view just interested in how to assess the credibility of the roadmap.
Not taking a view just interested in how to assess the credibility of the roadmap.
It's still a tall order, but my reading is that they seem to be doing all the right things. And they certainly aren't short of cash to invest with.
They certainly are trying to get the band back together. I see little reason this means they can turn around a huge beast.
Why do you see little reason? We have repeated examples in recent history of a strong leader turning around what appears to be an immovable object. Jobs at Apple. Nadella at Microsoft. Su at AMD.
I see little reason to believe Intel can’t do the same. They have the technology and history to be successful, they just appear to have had management that wasn’t prepared to deal with a changing market.
Same with marketing team. Get back the 90's team that made many memorable adverts. Blueman group, bunny suit ads, the works. Pentium, the word alone brings fond memories. Absolutely legendary. The current Marketing/PR team has created a disaster after disaster (see LTT rants about all kinds of PR issues).
10nm to 7nm. 7nm to 4nm 7nm+ to 3nm 5nm to 2nm
And all of them make senes and has been known for quite some time. I remember writing something on SemiWiki in 2018. So really they are pushing their 7nm to next year, their first node with full usage of EUV. An iteration of 7nm in 2023. I say iteration when in reality that was always the original target of Intel's 7nm. And 5nm in 2024.
10nm is already shipping, and judging by their product lineup seems to be yielding a lot better. Their 7nm next year are all going to be for their Aurora project anyway. So my guess is that it is more like a place holder with little actual capacity for anyone else including Intel's DC and Consumer business. Shipping their own product with 7nm+ / 3nm in 2023. And the real test will be 2nm in 2024. Is it aggressive? Yes, plausible? I am quite optimistic.
>evidence there is that Intel will be able to deliver all this?
Yes. Intel's biggest problem wasn't 10nm, or their chip performance. It was their mentality of keeping Fat profit margin from end product sales. And business School to this day continue to teach the same crap over and over again. That is why having MBA as CEO in tech hardly ever works. The evidence here is quite clear. Intel is really betting on being a Foundry. I do not believe Qualcomm and Amazon will sign up Intel if the price wasn't right. It means Intel is finally doing their second Andrew Grove Moment pivoted Intel away from DRAM to microprocessors, except this time to Custom Fab. As he once said. "Our Strategy is Simple, we build Fabs and we fill them."
Intel should split into two. One a fabless company focus on chip design. The other a foundry business so there is no conflict of interest like Samsung. Apple pull out of Samsung partially because Samsung compete with Apple on smart phone.
if Intel is serious about IDM 2.0. Intel needs to earn their customers trust. TSMC only focus on production and look where they are today.
On the other hand, Global Foundries took this route and look where they are today.
https://www.tweaktown.com/news/50551/intel-confirms-10nm-pro...
The capital, agreements and plan to catch up is there. But it's always healthy to be skeptical on execution.
My understanding is Intel's rivals also took a more incremental development process, which basically allowed them to capitalize on Intel's missteps.
7,4,3,20A in 4 years.
I think you're saying that the incremental half nodes steps are a more conservative approach?
TSMC pioneered the half node step, and it was very successful. So yes, I think copying successful strategy is a conservative approach.
That's some serious spin there. The chatter has been that Intel has had poor focus, poor working conditions, poor investments, poor communication within the company and a plethora of hubris. They need a culture overhaul and that seems to be what's happening. The jury is out to where they take this. I would bet on their competitors at this point.
that's a completely uninformed take on the topic. This is like betting on Airbus because Boeing has problems with the 737 - when the reality is there is no circumstance whatsoever where the federal government would allow Boeing to go under.
Intel is already being propped up with some sweet gigs doing supercomputer stuff for the National Labs (that their actual hardware probably does not justify on the merits - much like the national labs throw AMD a bone with CDNA in the GPU compute market) and a bunch of R&D tax cuts. The US also negotiated their way into a TSMC gigafab as a fallback strategy but they very much would prefer the American fab company makes it through as well, and they'll spend the cash to make it happen.
It's far cheaper and faster to get the existing competitor back up to speed, than to build a new foundry company from scratch, or even to just build a TSMC gigafab.
Their competitors also aren't american companies. GloFo is controlled by UAE investors. TSMC is controlled by Taiwan. Samsung is controlled by a Korean chaebol. AMD and Apple are fabless companies. Intel is the only US company still designing chips and producing them at a US foundry on any kind of a competitive level in the performance computing market.
The Israel parts of Intel had essential roles for decades, both in CPU design and in manufacturing, so it is very unlikely that there exists any Intel IP or know-how to which Israel does not have complete access.
The main difference is that Israel is considered as an ally.
edit: Intel seems to play a bit of marketing games here. (source Anand tech).
2021 H2, Intel 7: Previously known as 10nm Enhanced Super Fin or 10ESF.
2022 H2, Intel 4: Previously known as Intel 7nm.
None of these numbers have aligned with physical features on the chip for a while, so they are just an arbitrary convention now. It's better for all of the manufacturers to use the same convention.
A much more meaningful way to compare processes across different manufacturers is directly by transistor density, but the "nm" process generation naming tradition is so entrenched that every one is still using it but fudging the number in some sort of attempt to position/align their own processes against those of competitors.
Of course it's not just transistor density that matters, but also power usage and speed, and no one number can capture it all