Geekin' out on SD Card speeds!

Discussion in 'Accessories' started by jff1625, May 7, 2012.

  1. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    Hi,
    I'm trying to guesstimate what kind of continuous shooting speed I could get on my E-PM1 after the buffer fills up, when using various SD Cards that I don't have. Hopefully this will help me to decide whether or not I want to get a faster card, and if so which one. I'm posting my research here in the hope that some kind or curious person or persons might like to confirm or correct some of my numbers if they happen to have the cards in question, or their equivalents. If this info proves helpful to someone, so much the better. Also I find this kind of thing fun. I must be nuts :dash2:

    The resources I'm using for this are:
    1) my own testing with my E-PM1 and Sandisk 8GB Class 4 card
    2) dpreviews review of continuous shooting performance of the E-M5 an E-PM1
    3) Giulio Scorios test on his blog of his E-M5 including the card I have and some that I'm interested in. Giulio didn't record his "buffer-full" shot rate but I think I can extrapolate it from his results by referring to and comparing with my and dpreviews numbers.
    4) amazon.co.uk for pricing

    I'm making a few assumptions, namely:
    - E-M5 jpegs are 11MB
    - E-M5 buffer takes 16 jpegs
    - E-PM1 jpegs are 8MB
    - E-PM1 buffer takes 9 jpegs


    Same 4 cards as on Giulio's test minus the eye-fi.
    Remember the main thing I'm looking for is a shots per second speed when the buffer is already full.
    Numbers and maths follow from here, lets dive in!

    Sandisk 8GB Class 4
    My test:
    10 shots in 15.99sec -> 0.63 shots/sec when buffer full on E-PM1
    14.37 sec to clear buffer
    Giulio:
    Buffer cleared just under 56 seconds.
    11MB/shot * 16 shots = 176MB / 56sec = 3.1MB/sec = 0.3 shot/sec on OMD, or 0.5 on PM1 :crying:

    So for this card our numbers are much the same, when allowing for timing inaccuracies and rounding errors and whatnot.
    Price per speed for this works out at £4.48 / 3.1MB/sec = 1.40£/MB/sec


    Sandisk Extreme 4GB 30MB/s
    Giulio:
    Buffer cleared just under 13 seconds.
    11MB/shot * 16 shots = 176MB / 13sec = 13.5MB/sec = 1.2 shots/sec on OMD, or 1.6 on PM1
    DPReview:
    8MB/shot @ 1.6 shots/sec = 12.8MB/s = 1.1 shots/sec on OMD, or 1.6 on PM1

    Again the two sources match up pretty well. Interestingly the buffer-full-shooting write speed is considerably less that the speed written on the card label, but greater that the 'class rating' speed (which is 10 for this card)
    Price per speed £8.79 / 13MB/s = 0.67£/MB/sec :clap2:


    Sandisk Extreme Pro 16GB 45MB/s
    Giulio:
    Buffer cleared just under 17 seconds.
    11MB/shot * 16 shots = 176MB / 17sec = 10.3MB/sec = 0.9 shots/sec on OMD, or 1.2 on PM1
    I don't have other data to compare for this card so I'll have to take Giulios word for it, but he looks like a trustworthy bloke. As G noted, this card tested a bit slower than the smaller but slower-rated 4GB 30MB/s card.
    Price per speed £14.40 / 10.3MB/s = 1.40£/MB/sec


    Sandisk Extreme Pro 8GB 95MB/s
    Mine - this one was really tested - not just guessing the numbers!
    7.8MB/shot * 18 shots = 141MB / 10sec = 14.1MB/sec = 1.8 shots/sec on E-PM1


    Sandisk Extreme Pro 16GB 95MB/s
    Giulio:
    Buffer cleared just under 10 seconds.
    11MB/shot * 16 shots = 176MB / 10sec = 17MB/sec = 1.6 shots/sec on OMD, or 2.1 on PM1
    Price per speed £32.30 / 17MB/s = 1.90£/MB/sec


    Sandisk Extreme Pro 64GB 95MB/s
    DPReview didn't post numbers for Large/SFine jpeg/11MB, so from the numbers they did post I guesstimate the E-M5 buffer full rate at 2.2shots/sec so:
    11MB/shot * 16 shots = 176MB / 6sec = 29MB/sec = 2.4 shots/sec on OMD, or 3.6 on PM1 :whoo:
    Price per speed £128.87 / 29MB/s = 4.44£/MB/sec :eek:

    It looks like DPRs 64GB card is faster than Giulios 16GB one.
    Also DPRs card can apparently write at pretty close to the PM1s continuous shooting speed of 5fps!


    The price / performance sweet spot among these is owned by the Sandisk Extreme 4GB 30MB/s card, which for £8.79 should give a buffer-full shooting rate of 1.2 shots/sec on OMD, or 1.6shots/sec on PM1. Above this the law of diminishing returns starts to kick in.


    What do you think? Is this kind of thing helpful? Does anyone have any corrections or extra info to add? Did anyone read all the way down this far?

    cheers, :drinks:
    -J
     
    • Like Like x 4
  2. atomic

    atomic Mu-43 Veteran

    224
    Nov 3, 2011
    Alright, I own an E-PM1, a few SD cards, and I am enough of a geek to appreciate this. Can you detail your testing procedure and math so I can duplicate it as exactly as possible?
     
  3. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    The testing procedure is actually pretty simple.
    Set the camera to the fastest continuous shutter speed (5) to minimise the effect of the camera writing to the card while filling up the buffer, and hold down the shutter. When the shooting speed slows down the buffer is full. Any time now start a stopwatch and count off ten shots. At the end of the ten shots stop the stopwatch. Then take your finger off the shutter button. Repeat several times and take the average time if you care about accuracy (I mostly didn't.)

    now the maths.
    10 shots divided by the number of seconds it took gives you a shots per second figure. In my first example above it took 15.99 (lets call it 16) sec so 10 / 16 = 0.63 shots/sec.

    Assuming the shots are 8MB we can also get a MB/sec rate. The total amount written to the card during the test was 10 shots * 8MB each = 80MB total.
    80MB / 16sec = 5MB/sec
    That's actually pretty good for a class 4 card!


    I'm finding it interesting that the MB/sec rating on the front of the card doesn't equal the write speed in continuous shooting, especially for the faster cards. They typically claim that the number on the front is the "sequential write speed" which I had thought would be exactly the thing that would be going on when shooting in burst mode. I have to assume that when it stops writing one file and starts on the next one that's when it stops being a "sequential" write.

    I've noticed that over at amazon some kind soul has posted performance charts for many of the more popular cards, under the "customer images" for each card. I'll have to go see which of their stats matches up with my results. I'm guessing it's somewhere between "sequential" write and "random" write. Maybe sequential + random / 2?

    -J
     
    • Like Like x 1
  4. Pyro451

    Pyro451 Mu-43 Regular

    162
    Apr 18, 2012
    Massachusetts, USA
    Steve
    Nice work OP. I expect an in-camera test is the way to go. If anyone doesn't want to take the time to repeat the testing, Tom's Hardware has an extensive set of performance charts on SD cards.
     
  5. atomic

    atomic Mu-43 Veteran

    224
    Nov 3, 2011
    Thanks for the info. I'll try to run some tests later today and post results.
     
  6. Promit

    Promit Mu-43 All-Pro

    Jun 6, 2011
    Baltimore, MD
    Promit Roy
    Personally I've been finding that for my largely non-sports photography, the speed benefits of cards beyond 20 MB/s just don't show up much in real shooting. The burst buffers don't seem to get any deeper in most cases, possibly due to limited write bandwidth in camera regardless of card speed?
     
    • Like Like x 1
  7. gsciorio

    gsciorio Mu-43 Top Veteran

    636
    Dec 29, 2011
    Miami, FL
    Good info here! Next time I test I'll do a full buffer test too.
     
  8. dhazeghi

    dhazeghi Mu-43 Hall of Famer

    Aug 6, 2010
    San Jose, CA
    Dara
    Sounds reasonable.

    One less obvious benefit of faster cards in my experience has been startup speed for the camera. Switching from a generic 16GB SD to a Sandisk Extreme Pro cut the off-to-ready time for my E-PM1 from 3 seconds to around 1. A worthwhile gain in my case as I am forever turning the camera on and off (to save battery life).

    DH
     
  9. drewbot

    drewbot Mu-43 Top Veteran

    702
    Oct 21, 2011
    Toronto, ON
    I recently switched from my Class 10 SD card to the Sandisk 95 MB/s UHS card for my E-M5.

    I must say that it is really impressive how fast the buffer clears.
     
    • Like Like x 1
  10. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    Every now and then I like to shoot my 1 year old kid in burst mode and and hope maybe one or two of the shots come out ok. I'm finding two seconds worth of buffer just isn't enough for this. I already set the burst rate lower, now I get three seconds - still not really enough, especially compared to the likes on Canikon where you typically get 100+ shots worth of buffer (a friend has a D90).

    So the point of this exercise is to see if it's possible to minimise the problem with lack of buffer size by getting a faster card. If I can shoot at 3fps for 3sec, then it reduces to 2fps until the card is full, that's not bad.
    At the moment I get 0.5 fps and feel like I'm missing some shots. Babies move faster than that!

    Giulio, please test the same cards on the PEN Mini! I'm also curious to see if PM1 and EM5 will write at same speed to same card.


    -J
     
  11. atomic

    atomic Mu-43 Veteran

    224
    Nov 3, 2011
    Okay, I got started in this today. I'll post preliminary results. I should have reread the procedure before I started as I misremembered and counted shots for 10 seconds after full buffer instead of timing 10 shots. I have the data (recorded the audio so I could be sure to get my count right), so I can redo the numbers later. It's late and I'm tired, so here is the rough data. I'll add pictures of each card for identification purposes later.

    Each card was empty, formatted. Camera is E-PM1, set to manual focus, 1/4000 shutter speed. I started the count on the second post buffer shot and counted for 10 seconds. In most cases the pause between the buffer filling and first, and between first and second shots was longer than the rest, so I eliminated these from my results. Math assumes 8Mb per shot, as above.

    Sandisk Extreme Pro 45 Mb/sec 16GB - 30 shots in 10 sec = 3.0 shots per second = 24 Mb/sec
    Lexar Professional 133x 32GB - 29 shots in 10 sec = 2.9 shots per second = 23.2 Mb/sec
    Transcend class 10 SDHC 16GB - 28 shots in 10 sec = 2.8 shots per second = 22.4 Mb/sec
    PNY Professional class 10 20Mb/sec 16GB - 17 shots in 10 sec = 1.7 shots per second = 13.6 Mb/sec
    PNY SDHC class 4 (old) 4GB - 16 shots in 10 sec = 1.6 shots per second = 12.8 Mb/sec
    Patriot LX series class 10 8GB - 12 shots in 10 sec = 1.2 shots per second = 9.6 Mb/sec
    Eye-fi SD card 2GB - 9 shots in 10 sec = 0.9 shots per second = 7.2 Mb/sec

    One thing I noticed is that while with all the other cards I got a fairly steady shot rate once the buffer filled and the first shot recorded, the PNY 16GB class 10 card would shoot 3 or 4 quickly, then pause, then 3 or 4 more. I have no theory as to why.
     
    • Like Like x 1
  12. Chrisnmn

    Chrisnmn Mu-43 All-Pro

    Apr 26, 2012
    Auckland, New Zealand
    Chris
  13. heedpantsnow

    heedpantsnow Mu-43 Veteran

    332
    Jul 24, 2011
    Thanks for the information! IMHO one suggestion to make it even more useful (and maybe you already thought of this): could you modify your cost per speed to be cost per speed per GB? It seems like it would be more 'fair'.
     
  14. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    Wow, great results Atomic! Your real cards are faster than my imaginary ones :cool:
    I would guess that your PNY card has a slow 'random' read/write rate (try saying that 5 times quickly!) So it kinda loses the plot when trying to figure out where to start writing the next file.
    I don't think the exact procedure matters too much as long as you're getting a buffer-full-shots-per-second figure and the speed is fairly consistent.

    yeah, the cost per speed figure was just a bit of fun - all the cards bar one are under £20, so it hardly seems worthwhile unless you;re comparing the extreme high-end cards like that 64GB UHS-1 beast.

    -J
     
  15. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    Thanks Kiwi-Chris. Mr DaiSuki doesn't seem to say exactly what it was he was testing. I have to assume he just clicked the button once and waited till the 'card' light stopped flashing. It's not the same thing as buffer-full write speed but it should be reasonably comparable. His result for the "16GB Transcend class 10" and Atomic's, earlier in this thread, are roughly the same after adjusting for differences in file size, even tho the card capacities are different.
    His relatively "meh" results for the UHS-1 cards seem to be explained by the specs for the E-PL2 not listing UHS-1 among its many talents.


    Can anybody find a coherent way to explain or predict the difference in speed seen between two cards of different capacities but are otherwise identical? I've been told by various people that smaller cards are faster than bigger, and that bigger cards are faster than smaller. At least one of those opinions must be wrong! I suspect it's just random, and a side-effect of the "binning" process where are faster part might be badged as a slower one just to fill a gap in the market.

    -J
     
  16. jnewell

    jnewell Mu-43 All-Pro

    Jun 23, 2011
    Boston, MA
    Based on the results in the first post and Giulio's blog, I bought a couple of Extreme Pro 95MB/s cards and tested them against my Extreme 45MB/s cards (both 16GB). I did a little different test routine (shoot continuous low till buffer fills, time to clear) so I got more pics off before the buffer filled, but a little to my surprise the number of pics to fill the buffer was actually either the same or only one greater with the 95MB/s card (due to greater read/write speeds) and the time to clear the buffer was only slightly faster with the 95MB/s card. I expected a greater difference...will be re-testing.
     
  17. jff1625

    jff1625 Mu-43 Regular

    100
    Jan 14, 2012
    London
    Hi All,
    I finally bought my own Sandisk Extreme 8GB 95MB/s UHS-1 card. Test results added to the first post. It came out a little faster than the Sandisk 16GB 45MB/s and a little slower than the Sandisk 16GB 95MB/s. 1.8 shots/sec with buffer full on E-PM1.
    I also tested in RAW mode with 11.1MB ORF files, and it wrote about 150MB worth of them in ten sec = 1.3 RAW shots/sec, so it seems the card writes faster with bigger files, altho not so much faster than you get a higher shots per second count.
     
  18. Papayoyo

    Papayoyo New to Mu-43

    8
    Nov 1, 2012
    Just received a Transcend 32GB class 10 and pretty disappointed by its 12.5 MB/s write benchmark, which is not even fast enough for the 13 MB/s needed for HD video on my E-PM1. Wondering why it's performing at half the speed of Atomic's 16GB Transcend. Should've gone for the Sandisk Extreme for £5 more. :(
     
  19. kinlau

    kinlau Mu-43 Top Veteran

    836
    Feb 29, 2012
    13Mb/s, not MB/s, note the lowercase vs uppercase, 13Mb/s is only 1.6MB/s. HD video typically only needs class 4 or 6, not even class 10.
     
  20. Papayoyo

    Papayoyo New to Mu-43

    8
    Nov 1, 2012
    Ah that's a relief. Haven't had a chance to try it out yet as expecting the camera tomorrow. Will update with in-camera performance.