Intel makes the front page of the nightly business news for losing somewhere between 4% and 14% of their market share. Next day, Intel abruptly fires their current mainframe head dog and churns their entire internal organization structure yet again for the 4th time since they fired their old errant bean picker in chief.
Intel dudes, based on your Board of Director's statements these periodic purges will continue until you guys start doing something right .......One questions if this endless game of musical chairs is really helping Intel do any long term good stuff ?????? At some point in time lining them all up on the edge of the aqueduct and pushing every 10th one off with a spear butt becomes sorta counter productive.
The guy with the spear butt keeps saying the same thing while sending his solders over the edge down the long drop .......
"Get off 14nm. Get 7nm working right."===================================================
As Huawei doubles down on making their own chipset building equipment (Huawei has no choice, Trumph's entity listing says no state of the art ASML scanners nor any other western tech for Chinese Military linked businesses). At the same time Intel acts to slowly decimate its own business groups just to get them to move off the dime technologically, all the while Huawei is also busy floundering about seeking a future pathway.
So perhaps it is time for us to stir some rumors about the future of electronics past 2mn.IBM and Samsung have built some normal style 2nm tech samples and some 2nm Gate All Around samples.
Both of these sample runs ran into signal loss through the tiny connecting wires going to both styles of processors, making the 2nm results a lot less reliable than 3nm & 4nm results.
Soon "smaller" will not be better any more.===================================================
At this same time AMD has announced they will be building distributed AI and distributing upper level graphics into the guts of every chiplet.
In essence AMD will be distributing their massive GPU and AI and the localized chunks of systems memory at the localized processor chiplet level
keeping it actually located inside their CPU structure. At 5nm and below, the new extra available flat real estate on the socket makes this expansion possible.
Boosting the L-3 and L-4 memory up vertically on top of the top of the die also helps make more room for this functional AMD expansion. By keeping most of the action close by
inside the chiplet processor itself AMD is avoiding this growing issue of large signal loss due to very small fine connecting wires.
And this will actually work out to be both faster and better for AMD, as these new chipsets do not need to communicate long distances across a motherboard or through complex buss structures or through secondary communications chipset subsystems all of which stuff always act to slow down the overall data flow. Aside from the exterior inputs and post computational outputs, everything remains inside the CPU itself.
Ampere and others have recognized these same small wire leakage issues in their own testing and they have developed their "full wafer sized chipset systems" to put 100% all of the computer into the resulting huge wafer sized chipset itself. AMD is doing the basically the same thing, just using their superior chiplet design in aggregate as a distributed full computer system with very close by local memory and GPU and AI distributed clusters. The AMD system is much better in making less testing scrap, as a defect in the Ampere system can scrap an entire wafer, while several flaws in the proposed AMD system result in very minimal chiplet scrap.
Base upon this generic issue of 2nm signal loss due to fine wires, AMD clearly sees that they only have 2-3 future lithography shrinks left for them to utilize ....... and they also see Intel in the background kinda sorta matching AMD's current high end performance levels using much much larger and much cruder lithography running at relatively huge power draws, all the while using lots and lots of sneaky cute little AI tricks just to get the job done "fast enough".
AMD can do the same sorts of AI tricks in their much faster processors, but they will approach these tricks more openly if they go there as AMD/Xylinx owns all the AI tech base patents that all these tricks depend upon.
Intel does not own this tech at all, AMD owns it ---- thus the Intel sneaky "don't ever talk about it" approach Intel is using.
What is also clear is the consumer need for general processors has doubled during the 2 Covid years and it will double again inside of 3 years as we recover from Covid.
Intel and Global Foundries are still betting their older tech base is still actually good to go for as long as it physically lasts as the marketplace is telling them that OLDER LITHOGRAPHY IS BETTER THAN NO LITHOGRAPHY and that they can still sell all they can make at this point in time. AMD currently gets to charge a sizable premium for their best in class chipsets, and all the others must discount their prices to move their inferior designs if demand eases up any at all.
Huawei is also taking this side of the bet, as they have their existing old AMD CPU licenses to produce 7nm and larger chipsets, and this is the tech level where Huawei will be building their new Chinese foundries using IBM/GF style old homemade lithography systems.