iPad Widely differing NAND flash speeds for iPad Air 2

Discussion in 'iPad' started by userjohn, Jan 18, 2015.

  1. userjohn macrumors member

    Joined:
    Mar 24, 2011
    #1
    Can anyone suggest good apps or methods to test NAND flash speeds in the iPad Air 2 (or iPhone 6 Plus)?

    Using the Passmark PerformanceTest app, I've tested all iPad Air 2 configurations (ignoring color) except 128GB Wifi. I didn't see major differences between Wifi and Cellular for 16GB or 64GB, so I only list cellular results below, and skipped 128GB Wifi for now.

    Is there any reason to expect differences between Wifi and Cellular? I've seen speculation the Wifi might be more likely to use MLC (because more space is available), but no such evidence.

    Average results, 10 trials per device:
    Size: Write, Read (MB/s)
    16GB: 31, 725
    64GB: 163, 939
    128GB: 181, 947

    Are the much higher write speeds for 64GB and 128GB because they use TLC rather than MLC? (I haven't checked the type of NAND in the models.) However, isn't TLC supposed to be much slower than MLC? I recall seeing benchmarks showing TLC was much faster writing zeros (not very useful), but much slower writing random data (much more relevant). Perhaps the benchmark writes zeros? Does anyone know how it works (e.g., type/size of data, all sequential or some random access)? The read speeds are also substantially higher for the 64 and 128 models, but is that another benchmark artifact?

    I also noticed MUCH higher variability across the 10 trials for the 16GB (vs. the 64GB and 128GB), especially for reading. Should MLC vs. TLC differ in consistency of speed, or is there another explanation?

    128GB showed modestly higher write speeds than 64GB, but they presumably both use TLC. Should speeds normally differ in this way based on size (even if both MLC or both TLC), or is that difference likely for other reasons (e.g., variation among individual units or vendors)?

    It would be nice to see other people post benchmark results (using the app above or a better benchmark), and identify their model, and type of flash (MLC/TLC) if known.
     
  2. Fzang macrumors 65816

    Fzang

    Joined:
    Jun 15, 2013
    #2
    So the average write speed for 16 GB iPads is actually lower than the maximum USB 2 throughput? Yikes.
     
  3. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #3
    Yes, and also much slower than the wireless speeds available over 802.11ac.

    Also, as mentioned above, I wonder whether even that "low" 16GB write benchmark may overstate the write speeds for "real-world" (random) data on the 16GB, and whether the 64GB and 128GB are even worse for such data (despite the benchmarks showing higher values).
     
  4. canesalato macrumors 6502a

    canesalato

    Joined:
    Jan 31, 2010
  5. Ledgem macrumors 65816

    Ledgem

    Joined:
    Jan 18, 2008
    Location:
    Hawaii, USA
    #5
    If you look at SSD benchmarks and vendor technical specifications you'll find that devices with larger storage are faster than those with small storage, even if they're of the same product line. This is normal with flash memory.
     
  6. Charliebird macrumors 6502a

    Joined:
    Mar 10, 2010
    #6
    Why does this matter? iOS doesn't really cache things to storage to be quickly retrieved like a full blow OS. Any real world performance differences is going to be negligible.
     
  7. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #7
    I've seen specs sometimes vary by size, but also often not. For example, the Samsung 850 Pro SSD only has a slightly lower Sequential Write spec for the 128GB size, and the other 3 sizes (256GB, 512GB, 1TB) are the same, and all other speed specs are the same for all 4 sizes.

    On the 2013/2014 Retina MacBook Pro, the 1TB flash is much faster (both reading and writing) than the 512GB (perhaps because the former uses 4 PCIe channels rather than 2). However, the 1TB and 512GB sizes seem to be the same speed on the Retina iMac (I think both are slower because 4 channels aren't available).
     
  8. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #8
    Even if iOS doesn't currently page memory to flash, I want the highest possible speeds for all NAND flash operations, and have little patience for waiting.

    I did a few tests with all 3 sizes reading and writing some video files (about 800MB) using iTunes via Lightning cable, and they all showed fairly similar speeds around 20 - 22 MBps, which is much lower than the benchmark app's results, and far below the capacity of the local Gigabit network, or even the Internet connection (500 Mbps).

    I'm surprised the tests showed neither a big advantage for the 64/128 sizes (as the benchmark app claimed), nor a big disadvantage for them (as I thought might exist for real world files because they use TLC rather than MLC NAND).

    I wonder whether the USB 2.0/Lightning connection and/or iTunes is somehow the real bottleneck causing the speeds I saw, so that I'm not seeing the actual capabilities of the flash.
     
  9. Newtons Apple macrumors P6

    Newtons Apple

    Joined:
    Mar 12, 2014
    Location:
    Jacksonville, Florida
    #9
    So how much are you thinking this is making you wait? You must be a very busy person.
     
  10. Charliebird macrumors 6502a

    Joined:
    Mar 10, 2010
    #10
    Fair enough. I similarly have little patience running synthetic bench marking test on a device when I could be using it for productively and entertainment.
     
  11. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #11
    Yes I am, and the amount of waiting obviously depends on the size of the file. For even the one moderate size file I used for the test, the transfer took about 40 seconds. When an operation is fast enough (a sufficiently small fraction of a second), further speed improvements wouldn't be noticeable, but we aren't even close to that point.

    I'd be willing to pay much more for faster (e.g., CPU, Flash) iOS devices if such options were offered, as they are on Macs, where I can choose faster options, like the 2.8GHz CPU on the rMBP, and the 4.0GHz CPU on the 5K iMac.
     
  12. Newtons Apple macrumors P6

    Newtons Apple

    Joined:
    Mar 12, 2014
    Location:
    Jacksonville, Florida
    #12
    Amen. I also consider myself busy running my business and handle 50-75 calls, a couple hundred texts and close to 80 emails in the typical workday and do not have time to run silly bench tests on my cellular Air2 and 6+.

    I would rather spend my time in my company's business and my employees than worry about those few lost seconds from my NAND!:p

    No sense in worrying about something you can do nothing about as my Apple hardware is not in jeopardy!
     
  13. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #13
    Yes, I also don't have much patience for running benchmarks, so I put it off for months, and still haven't done much testing. I was hoping better information would be publicly available.
     
  14. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #14
    Thanks for the data. Although it probably doesn't matter, just in case: do you have Wifi or Cellular?

    Also, I assume you don't know whether you have MLC or TLC NAND, but if you do, please let me know.
     
  15. iamMacPerson macrumors 68030

    iamMacPerson

    Joined:
    Jun 12, 2011
    Location:
    AZ/10.0.1.1
    #15
    156 write, 1001 read. 128GB Wi-Fi Air 2 Space Gray.

    Honestly this doesn't concern me much. I have never thought of my iPad being slow, and the fact the NAND chips inside write faster then your normal laptop HDD, I'm perfectly happy with it.
     
  16. userjohn, Jan 22, 2015
    Last edited: Jan 22, 2015

    userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #16
    Thanks for the input. Depending on one's usage, the current NAND Flash speeds may be completely sufficient, and it's a huge improvement over previous iPads and I was glad to see it (along with the increase to 2GB RAM).

    I've personally always wanted every iPad I've owned (all 6 generations) to be substantially faster, but that's even true for every Mac I've owned (despite being much faster than iPads). I've never been happier with a hardware upgrade than when I first switched in 2008 to using SSD's in my Macs. Since then, HDD's have felt intolerable to use as a system drive.
     
  17. zorinlynx macrumors 601

    zorinlynx

    Joined:
    May 31, 2007
    Location:
    Florida, USA
    #17
    Larger flash memory chips / more chips means more parallelism in writes, which means more throughput. Reading from flash is MUCH faster than writing, so you get less benefit from parallelism there.

    A 16GB chip might have, say, two read/write channels, and the 64GB chip might have eight. Maybe the 128GB iPad uses two 64GB chips, for a total of 16 channels. Each channel is writing to a different part of the chip, or a different chip.

    The system can read/write from those channels in parallel, speeding things up significantly. You see this in USB 3.0 thumbdrives as well; the 32 and 64GB versions of the SanDisk Extreme vary by quite a bit in write speed.
     
  18. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #18
    Thanks for your message. I was generally aware of the factors you mentioned, and I've often noticed higher speeds for larger SSD's or flash drives. However, I've also seen many cases in which the speeds for different sizes are the same (as noted above for certain sizes of the Samsung 850 Pro).

    It's still not clear to me whether there's a consistent speed difference between the 64GB and 128GB iPad Air 2. One complicating factor is that there might be variation caused by different flash vendors. For example, there were sometimes huge differences in performance of flash of the same size in past MacBook Airs depending on the flash vendor for that particular unit.
     
  19. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #19
    The tasks you listed sound like they don't demand much of the hardware, so perhaps iOS models from years ago would be satisfactory for you, as they are for millions of users (which is why Apple still sells them).

    However, many people also care more about performance, which is why great effort is expended on benchmarks by sites like Anandtech, and many people perform and submit benchmark results like Geekbench and Passmark, and pay hundreds of dollars for relatively minor CPU upgrades (e.g., 2.5 to 2.8 GHz on the rMBP, a much smaller difference than the alleged NAND disparity).

    Despite claiming to be busy, you've taken time to post useless derogatory commentary ("silly bench tests") on my thread which offered and requested comparative NAND performance data for people who care about the topic. I never claimed it should be important to most users, and you're welcome to ignore it. I wonder if you also enjoy trolling Tesla forums about why the P85D and differences of seconds in 0-60 times are silly.
     
  20. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #20
    Here's an example of an article claiming to show MLC NAND greatly outperforming TLC NAND in the iPhone 6 (referencing the original Korean article about the benchmarks):

    http://www.iphonehacks.com/2014/11/...istently-outperforms-ones-tlc-based-nand.html

    The article seems to have a typo saying that TLC (not MLC) is faster, and I don't know the quality of the benchmarks. Interestingly, even the MLC speeds shown (for random data) are far below the iPad Air 2 results above. Indeed, the read and write benchmark results I see on my MLC iPhone 6 Plus are far slower than all iPad Air 2 sizes.

    There have been many complaints about reboots on the iPhone 6 and 6 Plus allegedly caused by TLC NAND. Apart from performance questions, I'd be interested to know whether people are experiencing more reliability issues with TLC NAND in the iPad Air 2's (presumably only in the 64GB or 128GB models.)
     
  21. Newtons Apple, Jan 23, 2015
    Last edited: Jan 23, 2015

    Newtons Apple macrumors P6

    Newtons Apple

    Joined:
    Mar 12, 2014
    Location:
    Jacksonville, Florida
    #21
    For being the "power user" you "claim" to be, you are spending a lot of time here, too.:p

    Best of luck to you and your slow NAND. I will not waste anothe second of your valuable time.
     
  22. MikeCBR macrumors member

    Joined:
    Dec 5, 2014
    #22
    Widely differing NAND flash speeds for iPad Air 2

    188 mb/s write / 998 mb/s read on my 128gb air 2 cellular space gray from best buy
     
  23. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #23
    I'm glad to spend time discussing iPad NAND speeds with other people who care. The waste of time was your needlessly denigrating the topic simply because it didn't matter to you.
     
  24. userjohn thread starter macrumors member

    Joined:
    Mar 24, 2011
    #24
    Nice. Highest write speed so far.

    When I submit a result from the Passmark app after running the benchmark, and then view those results on the web site, the read and write speeds displayed there (for both disk and memory) are strangely ~21% higher than in the app, but other figures seem to match (even the Disk Mark and Memory Mark, as opposed to the individual read and write values).

    The read and write values I listed were from the app, not the website. I assume all other results in this thread are also from the Passmark app (not the web site, or any other method), unless stated otherwise.
     
  25. rui no onna macrumors 601

    rui no onna

    Joined:
    Oct 25, 2013
    #25
    That's because SSDs like the 850 Pro have sophisticated controllers that can employ tricks such as RAM caching, pseudo-SLC cache, etc with load power consumption that's probably the same as or even higher than the whole A8X chip altogether. Not exactly something you'd find in your typical eMMC package. Also, keep in mind that the difference in speed is just for sequential operation. For random operations (which is more important for the OS), the different capacities will likely post similar performance.
     

Share This Page