|
|
|
|
|
|
by KronisLV
556 days ago
|
|
|
I don't get why Intel had to scrap 20A and try to go for 18A, making things that much harder for themselves in the process while they're already in a tricky position, especially with speculation and mixed messaging and stuff like this: https://overclock3d.net/news/misc/former-intel-ceo-speaks-ou... It feels like every delay is pretty bad for their image and the public opinion, though of course I can't quantify it. It would probably be a bit different if they were shipping products: Core Ultra was a bit of a sidegrade in some respects but with better pricing it would actually be a good option. Even their Intel Arc cards had a rough launch but despite that most of the issues have been largely addressed and things are looking up, especially for the pricing that they can pull off and the market segment that they're targetting (like AMD also pulling out from the high end, because they can't compete with Nvidia). My current CPU is a Ryzen 5 4500, which uses a "7nm" process and which PassMark tells me is smack dab in the middle of everything that's available out there. Something like the Core Ultra 245K would be a very significant improvement, even a lower TDP part would be good, as long as it's priced correctly. Yet, Intel is trying for a moonshot with 18A instead of getting something "good enough" out there and competing on price, while doing that was more or less what let AMD survive for years (e.g. before Ryzen), doing which would probably put them in a more stable position and would let them iterate. |
|
|
That CMOS fabrication process was intended for a single Intel product, the laptop CPUs Arrow Lake H, which will now be made by TSMC.
The costs for bringing that process good enough for mass production would have never been recovered.
It is much better to channel all resources to improving the fabrication yields of the superior 18A fabrication process, which is intended to be used for many important Intel products, including server CPUs, and also for any external customers they will be able to lure.
While Intel has used a slogan like "5 nodes in 4 years", two of those were actually half nodes, i.e. "Intel 3" is an improved variant of "Intel 4" and "Intel 18A" is an improved variant of "Intel 20A".
It would have been much better for Intel if they would have also skipped "Intel 4", but they were in a hurry to launch Meteor Lake before the end of 2023 and "Intel 3" would not have been ready for mass production by that time, and they were still not convinced that they must completely outsource the fabrication of their CPU chips until they have a competitive fabrication process.
With some difficulties, Intel has succeeded until now to meet all their milestones toward their stated goal of having a competitive fabrication process, i.e. 18A, and competitive CPUs, by the end of 2025.
The main thing that can be criticized about their roadmap is that it had included too many intermediate steps, with not yet competitive fabrication processes and CPUs, which have required nonetheless a lot of development efforts, slowing the progress towards the real goal. Nevertheless, it is likely that most of the intermediate steps have been unavoidable, because without new products for a couple of years their sales would have been even lower and their losses greater.
Only if Pat Gelsinger would have remained CEO until that time (one year from now) it would have been possible to assess whether he has been a bad CEO or a good CEO.