Intel Corp (INTC)

[Unkwn]

As far as I can see Qualcomm should get the same sort of performance from a dual core A57 at 20nm as the current Apple A7 dual core gives. So yes the A8 may leave them in the dust again but I think the A57's will give Qualcomm a good bit of breathing room.

Why do you think you can compare Apple A7 with ARM A57? Just because it is 64 bit capable? I remember something like maybe 30% higher performance for A57 compared to A15. Considering the lower clock of A7 (ok, we don't know the turbo), I don't think it can be compared with A57. In addition, it is practically known that A7 is much wider, which increases performance but also requires lower clocks due to power consumption.

I wouldn't expect much from an A57 design regarding compute performance, especially taking into account its quite high power consumption with the planar 20nm process. I don't believe it'll be much better regarding power characteristics than 28nm - there is good reason why Intel went in all that hassle to introduce Finfets with their 22nm node. That wasn't just for fun. I expect those A57 to be clocked quite low, maybe ~2 GHz for turbo. We'll see.

The additional A53 cores will help in benchmarks though, so benchmark figures will be good and it has the octacore checkbox Intel most likely won't have. We'll see if the market is really that blind regarding real performance - I doubt it since the iPhone is a good seller (though it's a bit out of competition due to iOS).

I think it's just a stopgap measure to allow them to offer 64 bit capable chips in '15. They'll most likely be back to custom designs by '16.

It sure is. But 2016 is late. That one will go against Broxton. We'll see what Intel can do with its All-In mobile design at 14nm.

The idea, at least from this hardware designer's POV, is to save power. But there is little power cost in just scheduling the lightweight threads on big cores. Clock and power-gating ensures that the big cores don't run longer or harder than they need to. They are flexible enough to handle both light loads efficiently and heavy loads effectively.

IMO, the advantages of heterogeneous cores is on the order of milliwatts. That's actually significant in the mobile world, to be sure, but the costs and difficulty in programming for heterogeneous cores isn't worth it.

I'd even go that far and say there is no imminent benefit in doing so. Carefully designed bigger cores will show the same low power characteristics as those hard to handle (especially in software) tandems. I believe ARM just pushes it to increase their royalty streams and they don't stress the big core design effort for power saving as far as they could - may be dangerous for them in the future. A15 and A57 are not exactly great performance/watt cores.

Intel already showed what you can do with creative approaches. Their Core-M feature that puts the processor in staccato mode at lower load (i.e. turning it on and off for short periods to allow for power gating in between) is simple but ingenious. That's one possible way to allow very low power consumption at low load with big cores - I suspect they have a nice patent pool around this.