I will say that Intel's roadmap is confusing and a mess. I think they never recovered from their tick-tock release cadence, and their inability to reach 10nm has jumbled things up.
The lack of working 10nm is the source of all of their problems.
For those who don't follow this sort of thing...
- A working 10nm process would enable intel to make roughly 2-3x as many CPUs of the same feature set/transistor count (they went for 2.7x scaling) on a silicon wafer.
- A working 10nm process would also reduce power draw at the same feature set and clock speed
- Once mature, a 10nm process, due to the above scaling, would enable intel to make 2-3x the number of parts at the same/similar cost
Unfortunately several things have happened
- AMD have had success with their Ryzen based design which has forced intel to raise core count (which means more transitors, which means less dies per wafer which raises cost).
- Due to the larger than intended core count on 14nm, these high core count parts are drawing a lot more power than originally intended (witness the early days of X299 motherboards frying VRMs due to ridiculous power draw on the high end X299 CPUs - more than 10 cores - that were originally never intended to be released using 14nm).
- AMD has been able to get 7nm parts in volume from TSMC. 7nm TSMC ~ 10nm intel process, but the big advantage is that TSMC are actually able to produce in volume
- Almost all of AMD's CPUs they are currently building use the same small chiplet building blocks. Rather than trying to juggle logistics of making many different dies, AMD can focus on stamping out as many Ryzen chiplets as possible and them combine them in different ways to build different products. i.e., they can focus on making many small parts as efficiently as possible. Intel have several different core sizes they need to juggle manufacturing of based on what they think is required, in advance.
- intel had a run of security vulnerabilities that significantly impacted their datacentre workloads. this resulted in enterprise needing higher core counts to do the same work
End result:
- AMD can put out higher core count parts than intel can for the same price
- AMD's design after a couple of revisions is MORE efficient than intel's current designs, per core, in performance per clock and performance per watt
- in order to satisfy demand for higher core counts, intel is having to make higher core count CPUs than 14nm process was originally intended for. As each die is larger there is a higher percentage of it containing a defect. Not only is intels cost going up due to less dies per wafer, but the defect rate is rising as well which means they're having to throw more partly broken dies out
- due to getting less working dies out of a wafter, intel is running into supply constraints
So this is a bit of a perfect storm for intel. Their design is behind. Their process is behind. AMD are kicking goals in a big way. In order to try and compete on performance and core count, intel have sacrificed volume which has pushed up cost per part, but due to both TSMC's 7nm success and AMD's chiplet strategy AMD have been getting amazing manufacturing efficiency with a design that at this point is outright superior to anything intel have this side of 2022.
The next year will be interesting. Intel have
nothing competitive either now or prior to 2022 at least (assuming they can hold firm on their current roadmap).
To compete on price they are going to bleed money, and even if they were to win sales based on cost, they are stretched trying to manufacture enough parts right now anyway due to the 10nm process delays.
[automerge]1578875109[/automerge]
So, if you have to buy a new laptop before 2022, one depends on AMD adding Thunderbolt and AVX-512 compatibility.
There is nothing to stop apple adding thunderbolt to an AMD based machine - as Apple are a co-developer/licensee of the protocol.
Some already existing AMD boards for desktop even have thunderbolt that can be made to work.
This is by no means an unsolvable problem.
Also, for a large segment of the market, USB-C/3.x is "plenty fast enough".
www.intel.com
