The A6: What's Next?

Apple has somehow managed to get a lot of the mainstream press to believe it doesn't care about specs and that it competes entirely based on user experience. Simply looking at the facts tell us a different story entirely:

Apple SoCs
  2007 2008 2009 2010 2011 2012
Process 90nm 90nm 65nm 45nm 45nm 28/32nm
µArch ARM11 ARM11 Cortex A8 Cortex A8 Cortex A9 ?
CPU Clock 412MHz 412MHz 600MHz 800MHz 800MHz ?

Apple has been at the forefront of the mobile hardware race, particularly if we look at the iOS platform as a whole (iPad + iPhone). Apple was among the first to move from ARM11 to the Cortex A8, and once again with the move to the A9. On the GPU side Apple has been even more aggressive.

Apple hasn't stayed on the same process node for more than two generations, echoing a philosophy maintained by even the high-end PC GPU vendors. It also hasn't shipped the same microprocessor architecture for more than two generations in a row.

Furthermore Apple even seems to be ok with combining a process shrink with a new architecture as we saw with the iPhone 3GS. It's generally thought of as a risky practice to migrate to both a new process technology and a new architecture in the same generation, although if you can pull it off the benefits are wonderful.

The truth of the matter is Apple is very focused on user experience, but it enables that experience by using the fastest hardware available on the market. With that in mind, what comes in 2012 with Apple's sixth-generation SoC?

It's fairly obvious that we'll see a process node shrink. Apple has been on 45nm for two generations now and the entire market will be moving to 28/32nm next year. If Apple sticks with Samsung, it'll be on their 32nm LP process.

The CPU architecture is a bit of a question at this point. We already know that Qualcomm will be shipping its next-generation Krait architecture in devices in the first half of 2012. TI, on the other hand, will deliver an ARM Cortex A15 based competitor by the end of next year. The aggressive move would be for Apple to once again migrate to a new process and architecture and debut a Cortex A15 design at 32nm next year.

Looking purely at historical evidence it would seem likely that we'd get a 32nm dual-Cortex A9 design at higher clocks first. If Apple wants to release an iPad update early next year, that's likely what we'll see. That still doesn't preclude a late 2012 release of a dual-Cortex A15 solution, perhaps for use in the next iPhone.

Note that we haven't talked much about potential GPU options for Apple's next silicon. Given the huge upgrade we saw going into the A5 and likely resolution targets for next-generation tablets, it's likely that we'll see pretty big gains there as well.

GPU Performance Using Unreal Engine 3 Siri
POST A COMMENT

201 Comments

View All Comments

  • tipoo - Monday, October 31, 2011 - link

    Anyone know if there is a reason this hasn't made it into any Andriod phone yet? Does Google specify compatible GPU's, or is it cost, or development time, etc? Looks like it slaughters even the Mali 400 which is probably the next fastest. Reply
  • zorxd - Monday, October 31, 2011 - link

    The only reason is that no one used it yet. The TI OMAP 4470 will use the 544 which is probably a little faster.
    The SGS2 is using the slower Mali 400, however it was released 6 months ago. Yet it's not that bad, even beating the 4S in Glbenchmark pro.
    Reply
  • zorxd - Monday, October 31, 2011 - link

    I meant no SoC vendor is using it. Reply
  • djboxbaba - Monday, October 31, 2011 - link

    The numbers were incorrect and have been updated, the 4S is ~2x faster than the GS2 on the GLBenchmark Pro. Reply
  • freezer - Thursday, November 03, 2011 - link

    But not when running at phone's native resolution. Thats what people will use while running games on their phone.

    iPhone 4S has much more pixels for GPU to draw while having much smaller screen. Not very optimal for gaming right?

    http://glbenchmark.com/result.jsp?benchmark=glpro2...
    Reply
  • djboxbaba - Thursday, November 03, 2011 - link

    Correct, but we're comparing the GPU's by standardizing the resolution. Of course in the native resolution this will change. Reply
  • thunng8 - Monday, October 31, 2011 - link

    I don't see any GL benchmark that the Mail 400 beats the 4S??? Reply
  • freezer - Thursday, November 03, 2011 - link

    That's because Anandtech review shows only the 720p offscreen results.

    This gives very different numbers compared to running GL Benchmark Pro in phone's native resolution.

    iPhone 4S has about 60% more pixels than Galaxy S2, and so its GPU has to draw much more pixels in every frame.

    Go to glbenchmark.com and dig database yourself.
    Reply
  • Ryan Smith - Monday, October 31, 2011 - link

    The 544 should be identical to the 543 at the same clock and core configuration. It's effectively a 543 variant with full D3D feature level 9_3 support. The primary purpose of the 544 will be to build Windows devices, whereas for non-Windows devices the 543 would suffice. We don't have access to PowerVR's pricing, but it likely costs more due to the need to license additional technologies (e.g. DXTC) to achieve full 9_3 support. Reply
  • Penti - Tuesday, November 01, 2011 - link

    Who will use it to support Windows Phone though? Qualcomm uses their own AMD/ATi based Adreno GPU. I guess it will be TI's attempt off getting Microsoft to support Windows Phone on their SoC in order to supply say partners of theirs like Nokia. Or might just be a later purchase/contract date for the other SoC vendors. Getting the IP-blocks later, but many did opt for the Mali-400 so why wouldn't they opt for the successor too? It seems to have worked out good. Samsung is just one of the vendors that usually did use PowerVR. I guess ST-E will use it in order to support Windows Phone on Nova A9540 SoC too. While Android vendors might opt for the older A9500 still.

    Interesting to see how Nvidia do lag in this field though.
    Reply

Log in

Don't have an account? Sign up now