I don't understand why people keep referencing Geekbench scores. It is one of the worst benchmark out there.
For instance the DGEMM part (multicore) scores about 20-25GFlops on an i7-4770. A real DGEMM implementation scores 150-200GFlops on this CPU. Thats 6 to 10 times slower than it should be.
Don't use Geekbench unless you want to show the world your technical incompetence.
Another bad example is GFX-Bench. According to GFX-Bench Manhattan 1080p offscreen the 980 Ti is about 70% faster than the 390X. In reality across multiple games it is only 25% faster. In the GPU space this is a huge discrepancy. Normally biased games only give a bonus up to 10-20% compared to the average framerate across multiple games.
So GFX-Bench is heavenly biased and should not be used.
Web benchmarks are problematic too, because the introduce a second variable - the browser. When using web benchmarks one should run the benchmark in several browsers and publish only the best score. We want to know how fast a SoC is, not how slow some browsers are.
I personally think people overestimate the SoCs from Apple. Mainly because of Geekbench and GFXBench.
Many benchmarks show that is actually not the case, e.g. Kraken, 3DMark, Sunspider, WebXPRT, etc.
In this Benchmarks the Apple SoCs are placed like expected - At mucher lower positions than the Core M.
And everyone seems to forget one important thing - power consumption.
The ipad air used 10W sustained. There is no data for the ipad air 2 but you can expect it is about the same.
A Core M uses 6W sustained in the best configuration. That is 4W to power the screen and everything else. More than enough.
If the max power consumption increases with the iPad Pro, which is not that unlikely, because it is a much bigger device, we are way beyond of the power consumption of a maxed Core M.
It is very likely that the A9X contains a fourth core. Core M is only dual core. If the have about same MT performance, the much more important ST performance of the A9X will only be the halve of Core M. But actually I expect the Core M to have a better MT performance.
Lastly we should not forget that the A9X has a really big die and many more transistors than Core M, which means that the defect rate is also much higher. This leads to the assumption that the A9X is much more expensive to produce in comparison to the Core M.
All in all I don't think the A9X is any threat to Intel at all. It is likely that it has a higher power consumption, is more expensive to produce and has a much lower ST performance aswell as graphic performance if you take FP32 into account.
I also think Apple will now hit a wall with the A10X. The can't really add more cores because more than 4 are not used in 99% of time. The can't increase the frequency that much without increasing the power consumption drastically. The can't increase the memory bandwith because it is not the bottleneck now. The can't widen the execution engine, since that is already very wide. The can't add new instructions as they are tied to ARMv8.
With other words, the got allmost all low hanging fruits the could get. The fact that the increased the core count is a sign of that. The performance graph they showed, also displays that, if you interpolate.