the wrote:Typically maximum number of devices on a bus scales by powers of 2 and Apple already has six cores which would put the initial number at a mere 8. Presumably they have some coprocessors and/or the GPU on the same coherent but which would easy break that 8 limit. Next value up would simply be 16. I assumed that you could have picked that up from the subtext of my statements and your own knowledge but I mistakenly assumed that you'd actually read it.
I picked up that you were hand-waving about some meaningless nonsense, yes. "Presumably" it ain't 8, so, ^2, SIXTEEN! Such nuance! Much subtext! Meanwhile you're doing all this other hand-waving math about how they're going to have like SIXTY, that is SIX, ZERO cores later on.
I would think you'd pick up on that utter incongruence yourself, seeing as you're the one saying it. But hey, that would mistakenly assuming that you understand what you actually write.
the wrote:Because the SIMD performance of a AVX 512 enabled Xeon is more than twice that of the peak of what Apple's A11 can achieve? Not difficult to fathom this one. The A11 has SIMD deficient compared to Sky Lake-SP. Making this up by adding additional cores would narrow this gap.
Oh, so you know the comparative peak SIMD performance of the A11 now?
Interesting. Would you mind citing that for me?
the wrote:And? I openly admit as such as raw data on how much energy a memory controller consumes or just interconnect consumes are not readily available. This also ignores the various corprocessors that'd be removed from the A11 as they could be spun off into another chip that'd be easy to manufacture. The point being made is that Apple's cores are low power and if that isn't changing then scaling in that dimension is linear. There will be an increases in power consumption due to amount of IO work that needs to be done both on-die and off. However, these figures are not too far off from other ARM based server chips on the market with similarly proposed IO.
No, your point is that the weakness of the individual cores can made up by just adding more. Fin.
That is completely wrong. It is fundamentally untrue. There are some workloads that are inherently single-threaded, and even in the generic case such performance is irreducibly important to a certain degree in everything:
https://en.wikipedia.org/wiki/Amdahl%27s_lawthe wrote:If you got better numbers that don't require Jedi mind tricks by all means present them.
I'm still baffled at how you're an expert on the comparative SIMD performance of the A11.
Like, sure, I don't have "better" numbers:
that's because I don't just make them up.