Apple A9 SoC

Discussion in 'Mobile Devices and SoCs' started by Entropy, Aug 13, 2015.

  1. Entropy

    Veteran

    Joined:
    Feb 8, 2002
    Messages:
    3,360
    Likes Received:
    1,377
    Which is why tri-core and quad-core makes for a comfortable hypothesis. :smile2: It didn't seem as if Rys was trolling us though, and if he didn't, I tend to trust him.
    I actually don't think they are making too outrageous claims, even though those presentations are marketing events. The numbers are specific enough to refer to actual if selective benchmarking.
     
  2. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    Yes, however we saw what looked like a die X-ray shot during the presentation, and it didn't resemble a triple or quad-core. It might have been the old A8 chip though, I don't have any images handy for comparison...

    What hasn't been discussed enough so far in this discussion is Apple adding SMT. It was mentioned once early on I think, and then people sort of forgot about it. Them doing that could bring another 20%-ish of performance to the table in sufficiently threaded situations; intel chips certainly seem easily capable of ending up inside that envelope.
     
    french toast likes this.
  3. wco81

    Legend

    Joined:
    Mar 20, 2004
    Messages:
    6,920
    Likes Received:
    630
    Location:
    West Coast
    I considered a Nexus 7 instead of the iPad mini 2 but I wanted the LTE model, because it has LTE bands for the US, Europe and Asia. I've used it several times in trips to Europe. Android devices generally have different SKUs for different regions with just LTE band support for those regions.

    I paid for the $629 model, 32 GB with LTE. FWIW, iPad mini 4 for $629 would have 64 GB now.
     
  4. iMacmatician

    Regular

    Joined:
    Jul 24, 2010
    Messages:
    797
    Likes Received:
    223
    I was surprised at the specificity of the numbers, in particular, that the 90% was not just rounded up to 2x.

    It's not any of the older A-series chips at least.
     
    Grall likes this.
  5. idsn6

    Regular

    Joined:
    Apr 14, 2006
    Messages:
    509
    Likes Received:
    163
    They blatantly lied in the same broadcast about the iPad mini 4 having the "power and performance of iPad Air 2".

     
    kalelovil and BRiT like this.
  6. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    Oh yeah, you're right. Good catch there! During the presentation I assumed it was the A8X SoC driving the device, and when the reveal came it's the regular A8, I'd forgotten the presentation (I don't care much about ipad mini).

    So yeah, they lied about that.
     
  7. Voxilla

    Regular

    Joined:
    Jun 23, 2007
    Messages:
    832
    Likes Received:
    505
    There shouldn't be many CPUs around where you can pinpoint the embedded M9 motion processor.
     
  8. Entropy

    Veteran

    Joined:
    Feb 8, 2002
    Messages:
    3,360
    Likes Received:
    1,377
    True enough.
    On the other hand, I can't say I'm able to judge the number of cores from the image of the die shot I have seen. My limitations are not necessarily shared by others, and maybe there is a better image floating around, or the people who were there have a clearer picture of what was shown.
     
  9. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    The image from the most recent presentation showed CPU cores that occupied a roughly proportional area compared to its immediately prior ancestors; A8 and A7. So it's probably not a triple or quad core design.
     
  10. Voxilla

    Regular

    Joined:
    Jun 23, 2007
    Messages:
    832
    Likes Received:
    505
    I think the previous comment of imacmatician is accurate enough:

    • Bottom right: 2 CPU cores + L2 cache (far right)
    • Middle: 2 SRAM blocks
    • Top center to top right: 6 GPU cores, I can see what looks like three copies of the same structure
     
  11. xpenryn

    Joined:
    Sep 17, 2015
    Messages:
    4
    Likes Received:
    0
  12. xpenryn

    Joined:
    Sep 17, 2015
    Messages:
    4
    Likes Received:
    0
    According to other source which leaked those info months ago, the GPU of A9 is running at 770MHZ.
     
  13. xpenryn

    Joined:
    Sep 17, 2015
    Messages:
    4
    Likes Received:
    0
    Codename of A9's CPU is confirmed as Twister.

    Nand interface of A9 is upgraded to PCIe 2.0, from PCIe 1.0 of A8.
     
  14. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    Interesting. 4M L3 is what Apple has used ever since the A7/iPhone 5S IIRC. I'm curious as hell to know what they've done to increase performance so much compared to A8...

    NAND interface is also a surprise. Typically, mobile devices use eMMC, not PCIe. I didn't even know there were flash chips with PCIe interfaces on them.
     
  15. xpenryn

    Joined:
    Sep 17, 2015
    Messages:
    4
    Likes Received:
    0
    Other info includes: Some pecents of 2gb ram are used only by GPU, maybe 512m or 768m. USB 2.0 interface is accelerated from 480Mbps to over 1Gbps.

    Waiting for GB scores...
     
  16. Lazy8s

    Veteran

    Joined:
    Oct 3, 2002
    Messages:
    3,100
    Likes Received:
    19
    A GT7600 @ 770 MHz + dual core Twister @ 1.8 GHz with some improved rebalancing of memory access for the GPU should yield quite a bit more than 90% better GFXBench scores.

    The GPU is probably clocked lower.
     
  17. wco81

    Legend

    Joined:
    Mar 20, 2004
    Messages:
    6,920
    Likes Received:
    630
    Location:
    West Coast
  18. Grall

    Grall Invisible Member
    Legend

    Joined:
    Apr 14, 2002
    Messages:
    10,801
    Likes Received:
    2,176
    Location:
    La-la land
    USB connection doing what exactly, some form of internal I/O I assume. It can't be the lightning interface, since there's just a regular ole USB connector on the other end of that cable. Wifi or cellular radios, perhaps? Anyway, I would have guessed a more efficient interface was used for these kinds of connections; USB is bad at maxing out the full bandwidth, and is CPU intensive too. For random junk like audio codecs, sensor chips and things like that, USB would be unproblematic, but those kinds of devices don't need 1Gbps speeds. So unless they've also messed with the protocol and not just the interface clockrate, I'm kinda wondering what the hell they're smoking... :D
     
  19. french toast

    Veteran

    Joined:
    Jan 5, 2012
    Messages:
    1,667
    Likes Received:
    9
    Location:
    Leicestershire - England
    Yea this what ive been thinking for a long time, would be a great fit for big ARM cores IMO.
     
  20. iMacmatician

    Regular

    Joined:
    Jul 24, 2010
    Messages:
    797
    Likes Received:
    223
    Here's a picture from the 1080p podcast video from Apple.

    [​IMG]

    My guesses as to some of the components are below, besides the M9 whose location was given by Apple. The above picture is at native resolution, this one is upscaled.

    [​IMG]
     
    Entropy and Grall like this.
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...