The folks over at iFixit are hard at work dissecting the newly announced/soon-to-be-available iPhone 4S. We've already gone over performance expectations as well as provided a high level hardware analysis, but with a tour inside the smartphone we're able to confirm a few suspicions. First is the Qualcomm MDM6610 baseband, a slightly different part from the MDM6600 that we had theorized earlier. There isn't a whole lot of documentation out on the 6610 but we're digging.

The second confirmation iFixit's teardown gives us is the size of the A5's on-package memory: 512MB. A quick look at the image above yields the Samsung part number: K3PE4E400B-XGC1. Each highlighted E4 refers to a separate 2Gb LPDDR2 die. The A5 features a dual-channel LPDDR2 memory interface, thus requiring two 32-bit die to fully populate both channels. The final two characters in the part number (C1) refer to the DRAM's clock period, in this case 2.5ns which indicates a 400MHz clock frequency (F=1/T). My assumption here is Samsung's part number is actually referring to clock frequency and not data rate, implying there are a pair of LPDDR2-800 die in the PoP stack. It's not entirely uncommon to run memory at speeds lower than they are rated for, a practice we've seen in graphics memory in particular for as long as I can remember, so I wouldn't take this as proof that Apple is running at full LPDDR2-800 speeds.

At a high level there aren't any surprises here, the A5 on the iPhone 4S is virtually identical to what was used in the iPad 2 - although running at a lower clock speed and likely a lower voltage as well. Many are surprised by the inclusion of only 512MB of RAM on the A5's PoP stack, however since most apps not in use are kept out of memory to begin with having more memory doesn't actually buy you a whole lot of performance. There are also potential battery life concerns with larger DRAMs: more room for apps to remain resident in memory leaves more DRAM cells to refresh, which impacts power consumption (although eventually Apple will have to cross this bridge, likely with the next process node transistion). I suspect the biggest issue created by not outfitting the iPhone 4S with more DRAM is limiting game developers to smaller levels, fewer unique textures, etc...

My real question is what comes next. Will we see a 28/32nm A5 used in the iPad 3 in 1H 2012 or will Apple continue to push the envelope and embrace a 28/32nm A6 with a pair of Cortex A15s in the second half? The latter is what I'd like to see, although it would effectively move all of the flagship iOS product launches to a fall ordeal.

Source: iFixit



View All Comments

  • Anand Lal Shimpi - Thursday, October 13, 2011 - link

    That is also true, Brian and I were having the same discussion there - without conservatively managing apps in memory, more memory would simply allow more data to be resident in memory, which results in more frequent cell refreshes and could impact battery life. I suspect the move to a smaller process node will help give Apple the breathing room it needs to accommodate more data in memory with a negligible impact on battery life.

    Take care,
  • dagamer34 - Thursday, October 13, 2011 - link

    I find the argument of having more RAM sucking up power to be silly, otherwise, it wouldn't have made sense for the iPhone 4 to have 512MB of RAM when all other A4 class devices have 256MB.

    And what's even weirder is that everyone seems to be ignoring that "less RAM = having to refetch data over 3G radio". Or the fact that kicking apps out of memory means they have to be re-initialized again, burning CPU cycles. I don't know about you, but I think that's a much more costly process in terms of battery life.

    None of this really applies to desktop operating systems because they have virtual memory, nor are they really aware of battery life constraints. If the system is under memory pressure, it'll just dump data to the disk and call it back up when it needs to. All mobile OS avoid this costly procedure because it's extremely costly to write just a few bits of data to Flash memory as entire pages have to be erased first.

    Regardless, the most likely real reason Apple didn't include 1GB of RAM is to a) save costs and b) they don't feel like they need more. The OS clearly runs very well compared to the competition due to not having to worry about garbage collection and every OS developer being very aware of the amount of RAM used (use too much and your app gets killed).
  • Penti - Thursday, October 13, 2011 - link

    iPhone 4 did need the extra memory and was the first of their phone-devices that had memory in PoP configuration to begin with. Radio is in it's own basebandchip with it's own RAM so I'm not sure what your trying to get at. You don't need a 512MB cache for the radio. Having less RAM does mean about half the power usage in this case. With regard to just the ram. Not that the iPad 2 had 1GB. It's really just the same chip as in iPad 2 with production sorted out under the iPad 2-run and binned for the phones. Otherwise probably no alterations.

    Background apps does mean more then DRAM-usage they are usually terrible when it comes to using the network and cpu just see the Skype app on Android etc. You definitively needs to put apps to sleep or suspend to flash when it comes to power otherwise the phone would quickly be busy doing things it doesn't need to do. But remember the iPhone 4 not only competed against an run more larger apps but also introduced third party multitasking/background apps. Upping the ram was only sensible then. Neither was power the same concern as now, when dualcore and new baseband meant it wasn't getting better in terms of batterylife.
  • quiksilvr - Thursday, October 13, 2011 - link

    If it is two 512MB chips and not just one, 1 GB chip, then yes. Reply
  • zorxd - Thursday, October 13, 2011 - link

    Android devices with only 512MB RAM aren't known for their superior battery life over 768MB and 1GB counterparts.
    I believe they did it mainly to save costs, knowing that the iPhone 4S would sell well anyway.
    Just like the iPod Touch and the iPad1 which both only had 256MB.
  • MySchizoBuddy - Thursday, October 13, 2011 - link

    Nvidia coming up with quad core this year. Can Apple afford to stay in the dual core world for iPad 3? Reply
  • ltcommanderdata - Thursday, October 13, 2011 - link

    What are quad cores going to be used for exactly in a tablet or smartphone? Even computer applications are slow to really optimized from dual core to quad cores. Many things on a tablet or smartphone are already hardware accelerated with dedicated chips such as audio and video encoding and decoding. There are dedicated image signal processors for the cameras. Other tasks could potentially run faster on the GPU once Apple implements OpenCL.

    And if Apple does switch to dual core ARM Cortex A15, which I think is realistically too early for the iPad 3, performance relative to a quad core Cortex A9 won't be a huge concern.
  • michael2k - Thursday, October 13, 2011 - link

    The A5 is anywhere from 2 to 6 times faster (GPU, some CPU) than a Tegra 2.

    Can Nvidia afford to NOT go quad core this year?

    Apple's A5 is likely to be nearly identical to a quad core Tegra 3, from a performance standpoint, with a die shrink+clock boost.
  • zorxd - Friday, October 14, 2011 - link

    the Tegra 2 CPU is slightly slower clock per clock. But never 2x. Also, Tegra 2 is clocked 25% faster than A5 in iPhone 4S. Reply
  • vision33r - Friday, October 14, 2011 - link

    The key here is "clocked" not measured performance as the case we saw in Anand's benchmarks the Tegra 2 was embarrassed even single core A8 SOC had best it in some areas which should never have happened.

    Of course Android fans like to buy big numbers instead of actually comparing actual benchmark results to gauge true performance.

    The SSGS2 being higher clocked than IP4S but fails to beat it across the board is a shining example of this.

Log in

Don't have an account? Sign up now