According to geekbench the A10 have higher ipc compared to skylake, i wonder why they do not make a desktop version since it would be faster and use it for they own mac family instead of using intel cpu's..
There's a lot of arguments beside performance (IPC in particular) like software compatibility, emulation, etc. but Apple did that not long ago so you are right in questioning why they aren't doing what you proposed. You could just use that as a counter argument to check if Geekbench is reliable after all.
I did a cross generation comparison and it resulted in Kabylake having 50% and 135% higher IPC than Penryn in integer and float subtests (basically the same over Conroe), yet A10 still beats it by 38% and 21%. Kabylake managed to outperform A10 only in two subtests, GEMM and FFT (I'm completely ignoring memory and AES results), where it also happen to use AVX and FMA.
Considered how great of an increase AVX and AVX2 are when in use for Kaby to win by only 25% and 5% smells bad, I mean unless Intel engineers are idiots there should be a catch for why A10 performs so good in any other test and still doesn't fare too bad in easily vectorizable scripts were core can stretch it's vector units.
Think about it: 38% better integer IPC! We got a measly 50% in 10years over Conroe, if I was making CPUs and believed this bench I'd dissect A10 for anything valuable... and it's not like all that performance comes with higher power alone because the platform is always a phone so it must be efficient too.
I'm still skeptic of GB4, on the other hand if 40%IPC over Skylake is indeed possible it would mean good things can still come... one can hope we are still far from a single thread wall and IPC can scale well beyond current designs.
A10 does appear to have higher IPC than Skylake, but remember that Skylake is a design that scales to well past 4GHz. Can the A10 CPU do the same?
I'm not so doubtful anymore given the 30% increase in clocks they got on the same node, what next on 10nm: 3GHz? That and higher than Skylake IPC (if true) would make an hell of a laptop cpu, then double/triple the cores and it reaches easily the same overall performance even without SMT.