Apple’s A4 actually a PA Semi SoC that’s currently virtualizing ARM until future iPhone OS update?

“Last week, I had all but conceded that the A4 was Cortex-A8 based, but developments this week demonstrate that whatever the A4 is, it is not merely a re-hash of the Samsung iPhone processor, as so many have concluded,” Mark W. Hibben blogs for Technomicon.

“A more consistent theory is the one I have offered in the past: the A4 is derived from the PA Semi PA6T, which is a dual core Power Architecture processor,” Hibben writes. “The reason why the performance is only so-so is that the A4 is still functioning as an ARM processor emulator. A future update to iPhone OS will undoubtedly take care of this problem.”

“Aside from Apple’s natural secretiveness, probably the main motivation was to avoid scaring off the developer community. Developers have been understandably leery of major platform changes in the past. When Apple switched to PowerPC from Motorola 68K, many developers simply dropped out rather than upgrade their development tools. With Apple providing the development tools, this really isn’t a problem any more, but I still hear developers grumble at the mere mention of a processor architecture change. They shouldn’t worry,” Hibben writes. “The change in instruction set should be fairly transparent to the developer in Xcode. Most of the work to adapt to iPad that developers have to do involves the larger screen size, an adaptation they would have to make in any case.”

Hibben writes, “The perception that the A4 was Cortex-A8 based had me very worried about the future of the iPad and Apple in general. It appeared that iPad was about to be waylaid in the market place from two sides. On the one side will be Cortex-A9 based tablets and smart phones running Google OS and Flash 10.1 for Mobile. On the other side will be Windows Netbooks running Windows 7 and the new 32 nm process Intel Atom processors that are right around the corner. This would have been very stiff competition for the humble ARM Cortex-A8. Fortunately, Apple does not appear to have burdened itself with such a mill stone.”

Full article, the 11th week in a series of A4 posts, here.

[Thanks to MacDailyNews Reader “KenC for the heads up.]

24 Comments

  1. This actually fairly typical of the entire design approach for Apple…maintain a measure of backward compatibility but build devices that the rest of the world will catch up to. This not only makes these systems more future proof, they actually IMPROVE over time (to a point) not fall into obsolescence in a year.

  2. First post where he posits the A4 might be a PA Semi in week 2, after
    Steve’s keynote:

    http://www.technomicon.com/iPad_Saga_-_Week_2.html

    Explaining in week 3 the choices Apple management faced in choosing a SOC
    for the iPad, and how the timeline forced them to an ARM8 or PA Semi:

    http://www.technomicon.com/iPad_Saga_-_Week_3.html

    In week 6 he explains how Apple could use the PA Semi A4 SOC-family in the
    iPhone and Macbooks:

    http://www.technomicon.com/iPad_Saga_-_Week_6.html

    In week 10, he points out the memory controller on ARM8s are 32-bits wide, while the A4 is 64-bits wide.

    http://www.technomicon.com/iPad_Saga_-_Week_10.html

    Here is the post, Week 11, where he states the A4 is not an ARM as the process width and die size don’t match up with an ARM8:

    http://www.technomicon.com/iPad_Saga_-_Week_11.html

    I don’t know if anyone recalls, the story from last week, when the new Macbook Pros came out that someone wondered if Apple could use an A4 in a Macbook, well, in Week 6 above, the writer shows that it might be possible with a 4-core A4 at 1.5Mhz, or something like that.

  3. @bart

    Wow, virtual processor platform… hmmm, that would be — SLOW

    There is no why that they are going to be able to emulate the ARM processor with reasonable speed on another mobile processor. I doubt, you could get reasonable speed on a desktop processor.

    It would be “so so” performance in comparison to the iPhone. It would be dog slow.

    Emulation is extremely inefficient. The fastest parts of a processor, such as setting registers become a huge bottleneck in emulated environments.

  4. @edster,

    Not necessarily. It depends on where the emulation is taking place. If it is at the user-code level (similar to what happened during the switch from MC680x0 to PowerPC), then you are correct, it would fairly inefficient. But it could be done (all or partially) in the A4 firmware. Just like with nearly every major processor since the 1990’s, the ISA is virtual, with the actual hardware running the firmware to give the appearance of the ISA. This is really not new, either. When IBM came out with the 360 series in the 60’s & 70’s, they had a range of actual hardware (with appropriate throughput and price), but they all “looked the same” as far as programming them due to the firmware. I’ve heard of a case where the IBM360 firmware was re-written to emulate an IBM 7040.

    In any case, the end result is that a firmware upgrade in the future could result in a more streamlined throughput for anticipated changes.

  5. @ John–

    No. The acquisition of P.A. Semi was for their technology.

    Several of their engineers went on to start a new company, which has also since been acquired by another company, which people have speculated could be Apple.

    Nowhere in that scenario is the word “flop.”

  6. I always thought that PA Semi doing an ARM design didn’t sound quite right. It’s in their DNA to do whatever they do with a Power chip. That’s what they’ve always done.

Reader Feedback

This site uses Akismet to reduce spam. Learn how your comment data is processed.