you obviously work in this field and are far more knowledgeable than I am, but everything that ive read as sated the there is literally no difference between skylake and kabylake and the only change is a few tweaks that allow for better power efficiency and therefor higher stock speeds, but other than that zero difference. how could a cpu that is identical architecturally be a hardware solution to something that was that is supposedly a result of the same hardware?
Errata is a funny problem.
The architecture itself plays a role, yes, but so does the fabrication methods. Intel hasn't done much to fix errata in the architecture itself since 'core', unless you count instruction migration and memory controllers. One is due to introducing a major standard shift, means you grandfather in a lot of the design flaws, unfortunately. The other reasons are speculation about how intel runs R&D. Now as far as fabrication, gate errata has been an increasing problem since sub 30nm. Most people will notice how this was all controlled using arrays of regulators onboard to control less than perfect circuits in the CPU and other hardware. Like the heavy use of mofsets. Now everything is significantly harder to correct by many orders of magnitude with a motherboard solution. A major problem now, not all nano-ICs can shrink at the same rate. Transistors and cache aren't the only parts of the processor die module itself. So this is where the gate issue throws a monkey wrench into tick-tock or ANY generational shift in production. So Intel can get larger samples of where the cycles are broken and redesign that super tiny circuit on the transistor block. Hoping within a year, the 14nm method x part for specific henries/ohms/farads/etc has been finely tuned to where they don't need to build unnecessarily large or incorrect spec with a workaround.
This is why the architecture itself, while nearly unchanged in function, is actually not all that identical. Kind of like a subdivision of homes, and one architect provided 3 plans to mass produce homes. Well, within each home of the same tier, the outlets, doors, plumping, fixtures, paint, etc are all very different. If you rework the design enough, the layout no longer is identical to the house next door with the same square footing. Odd analogy, but I hope that makes sense.
And yes, I know marketing yada yada, how they like to claim every codename is a brand new architecture, but no, they aren't, just reworked, sometimes slightly, sometimes considerably.
What's actually impressive about the whole topic to me. Given enough resources, a theoretical 100nm CPU die can be optimized to the point of matching the performance, power consumption/TDP of a 1nm CPU die with the current state of technology.
