This one caught us off guard, but Apple just announced the A6 SoC powering the new iPhone 5 features 2x faster CPU and GPU performance compared to the iPhone 4S. Apple reserves major Ax SoC number iterations for architecture changes, combine that with the performance claims as well as some other stuff we've heard offline and there's one conclusion: the iPhone 5 uses ARM Cortex A15 cores inside. Update: It uses a custom Apple core!

Our guess is two cores. No word on the GPU yet.

The A6 is 22% smaller than the A5, although it's not clear if that's a package or die size claim yet. There's a good chance this is built on Samsung's 32nm LP HK+MG process.

Comments Locked

118 Comments

View All Comments

  • Lucian Armasu - Wednesday, September 12, 2012 - link

    Of course it's smaller. A5 was made at 45nm. This one would be made at 32nm
  • Exophase - Wednesday, September 12, 2012 - link

    But Cortex-A15s take up more space than Cortex-A9s - I don't know how much, but with all of the changes it has to be substantial. Would not be that surprised if it takes up twice as much space sans L2 cache.
  • name99 - Wednesday, September 12, 2012 - link

    Why do they need Exynos? Why can't Apple can get the core from ARM, add on their various other blocks, and just submit the whole thing to a fab?

    People seem to believe Apple can't have become a "real" fabless SOC house; but we know they have bought a bunch of companies in this space; we know it is something they have long wanted to do; we know they have been leading up to this for the past five years, designing and customizing more and more of what goes into their SOCs.
  • iwod - Thursday, September 13, 2012 - link

    While it is true you could still have 2x CPU performance from a Quad Core A9, but anand commented on the performance increase example aren't multi threaded in nature, which means it has to be either double the frequency or A15 with slightly higher frequency then current A5 SoC.
  • BuddyRich - Wednesday, September 12, 2012 - link

    but the Arm Cortex A15 core was not confirmed, just a speculation on Anand's part? Wonder when we'll get confirmation one way or another.
  • tipoo - Wednesday, September 12, 2012 - link

    Just speculation
  • zorxd - Wednesday, September 12, 2012 - link

    It wouldn't be the first time speculations were wrong about the iPhone. Everybody assumed the 4S had 1GB RAM, since every other iteration doubled the RAM. It turned out it had only 512MB. Also, everybody assumed that the iPhone 4 had a 1GHz processor like the iPad, while it was really only about 800 MHz.
    Apple only said that the CPU was twice as fast. So it could be a dual A9 clocked at 1.6GHz, a dual A15 at unknown frequency (probably arround 1.2 GHz), or a quad A9 clocked as low as 800 MHz. We will see.
  • name99 - Wednesday, September 12, 2012 - link

    I think it's worth looking at the tasks they showed as sped up. They're all IO related. I suspect PART of what is going on here is flash that is 2x as fast. (And, last time I checked, Apple's current flash is about 2x as fast as the flash used in competitor high-end phones.) So that's part of it, and that's the kind of thing you can use to show 2x speedups even if your per-thread performance isn't 2x.

    But they do specifically say "2x faster" CPU, whatever that means...
    An alternative possibility to A15 which I raise simply for completeness is that they may have dusted off their old macro-scalar patent and applied that to the A9. My interpretation of macro-scalar was that it was a form of ISA-visible hyper threading, and you could argue that well-done plus say a 50% clock boost gets you to 2x speedup (at least for certain types of well-parallelized code) in the available area.

    Against that argument, we have the problem that SMT works best as part of the full superscalar OoO setup; trying to graft it on to the Pentium-style paired dispatch of the A9 is going to give some boost (mispredicted branches, cache misses) but not that much. On the third hand, you have to start somewhere, and maybe Apple fits it into the A6 with an A9 core as a trial run for when they REALLY care about using it, in the A15, when they can give 4 virtual cores in 2 core area and power?
  • tipoo - Wednesday, September 12, 2012 - link

    Could they not have just crammed double the old A9 cores in there, and still had it smaller since it's on a smaller fab process?
  • tipoo - Wednesday, September 12, 2012 - link

    Or even upped the clock speed plus other enhancements. No way of saying right now that it's A15 for sure.

Log in

Don't have an account? Sign up now