3rd generation iPad (March 2012): 1 GHz A5x (dual core), 1 GB RAM, PowerVR SGX543MP4 (quad core)
I suspect 6th generation iPhone will use 800 MHz A5X (dual core) with 1 GB RAM and Power VR SGX543MP4 (quad core).
I think I've got it. Apple can vary process, cores, clock, and architecture, to get user experience. But the translation is non-linear:
- shrinking the process by half (from 45nm to 32nm on an edge, about halves the area) reduces power consumption, but by less than half (by about 30%) - effect is dulled
- doubling CPU cores gives less than double performance (about 1.5) - it's dulled
- doubling *GPU* cores does give double performance - it's the same
- decreasing clock by x% decreases power consumption by more than x% - it's exaggerated
- extra CPU performance isn't strongly needed for user experience
- extra *GPU* performance *is* strongly needed by user experience
I think GPU improvement makes a huge difference to the user experience, while CPU improvement isn't needed at all. One measure is whether *present* apps are pegging the CPU/GPU - another trickier consideration is that if more power is available, then maybe apps will be enabled that use it (note: this is not necessarily a given: e.g. it was true for decades in PCs, but today, for most people, it's not true).
The strong impression I get is that CPU is fast enough, but GPU is not. Firstly, I've read that CPUs in iPhone and iPad are *not* getting pegged by present software - with the sole exception of photo-editing software (photoshop-like). This is interesting, because most such image manipulation runs better on GPUs than CPUs (since the task can be divided into independent subtasks, that therefore don't need to communicate, aka "embarrassingly parallelizable").
Secondly, when playing with iPads, they are much smoother than equivalent Androids, but there is still a little bit of latency and glitchiness if you really look for it. Maybe this doesn't matter for "user experience", if you have to "look for it", but I feel that it will feel better unconsciously, the smoother and more responsive it is - a kind of physical thrill. And, of course, games can take advantage of better graphics - and people appreciate better graphics, certainly up to the level of 360/ps3. Even if not 100% apparent on the device, you'll see it with HDMI out.
I think they will go: process shrink; dual-core CPU; quad-core GPU and then adjust the *GPU* clock rate to give the desired battery life.
That is, they take the power savings from the process shrink of *both* CPU and GPU, and use it *all* on the GPU by x2 the GPU cores. This gives x2 the graphics performance, but also uses x2 the power. If this is greater than the power savings from the shrink of both, they downclock to close the gap. E.g. if they downclock it to 90%, they're still getting x1.8 of the graphics power of the previous iPhone 4S. Basically, doubling GPU cores and halving clock is always a win, because cores scale linearly, but downclocking scales battery life *better* than linear. So add as many GPU cores as you can fit - and a process shrink enables you to fit more in the same space.
tl;dr iPhone 5: A5X, process shrink, with graphics slightly under-clocked.
PS: architecture change to cortex A15 CPU, and rogue G6200 GPU won't be ready until 2013 - for the iPad 4. But it will cross the magic threshold of performance parity with present-gen 360/ps3 consoles.
