"TSMC's 10 nm node, which first showed up in Apple's A10X chip in the iPad Pro, followed by the A11, has been fraught with
issues"
Just because the idiot press runs with this story doesn't mean MacRumors has to.
The truth is that, far from there being any sort of problem here, what Apple and TSMC have done is freaking incredible.
Every year Apple delivers a new processor design, almost every year TSMC delivers a new process. And the two work together incredibly. The A11 CPU is around 3 to 4x faster than the A7 CPU in single-threaded performance, half from process (2x speedup from 1.3 to 2.6GHz), half from constantly improving Apple micro-architecture. And that's not including the substantially higher increase in GPU performance, or the the media encoders, or the NPU or ...
This pace
A7 28nm
A8 20nm
A9 16nm
A10 16nm++
A11 10nm
is way beyond Intel, who've now been stuck at the same process node for a tock and two optimizations...
And we're supposed to be upset about this? TSMC is "troubled" because they can reinvent the world every 18 months?
When Apple DOES have a chance to optimize against a process (like the A9 to A10 transition) they can make good use of the opportunity. (Frequency boost of 27% in that case.) But even when they can't optimize for the process, just using the new process in the least daring, most conventional way still gives you a nice win (10% frequency boost for the A11, along with lots more transistors for the small cores, the custom GPU, the NPU).
The A11X will be on a new process (7nm) so probably, like the A10X, will not be very daring in ramping up frequency, just using the density for extra cores and larger GPU/NPU.
The A12 likewise will be on that 7nm process and perhaps again not too daring in frequency (maybe another 10% taking it to 2.85GHz) so expect a similar story like this next year --- sure, sure, Apple's new chip now beats not only Intel's m-Class mobile CPUs but now also beats the H-class 45W pro laptop CPUS --- which all goes to show how Apple and TSMC are in "deep trouble with a problematic 7nm process".
Finally enough with this "short-lived" vs "long-lived" process nonsense. Once again a stupid attempt to create drama where there is none. Remember that supposed failure, the TSMC 20nm process? Well, TSMC is still offering it. People are still using it. It's a perfectly valid design choice for certain tasks, just like 28nm or 65nm, or 16nm are all also valid design choices.
[doublepost=1505507511][/doublepost]
That's likely pessimistic. A small core is worth about 1/4 of a big core. (Varies depending on the exact benchmark, up to 1/3 for some code, down to ~1/10th for dense FP code). So that would naively give A11X around 20K for multicore.
But that assumes a single 4-small-core cluster, and we don't know what Apple's plans in this regard are.
One possibility is that small cores are to be used as is publicly assumed, to run "lightweight" threads. I don't buy this; I don't believe there is enough of that sort of work on an iOS device to require four cores. My PERSONAL suspicion is that these cores are to be used in conjunction with the GPU to switch the graphics model from todays standard model to a ray-tracing model. (Computation on the GPU, control and sequencing on the small cores.) If I'm correct, expect to see a ray-tracing API at next WWDC, and in a few years a UI that incorporates ray tracing.
Anyway assume I'm correct. Then the number of small cores needs to scale with the size of the screen, meaning that we might get a six-element cluster (which seems tight for a single L2), or two 4-element clusters, both of which boost the multicore score even higher.
(BTW I'm referring to the integer score, not the aggregate score, because integer is the one that matters most for normal purposes, and throwing in the crypto, FP and mem scores just confuses the issue.)
