Both used Test : 50 MiB [D: 0.0% (0.0/4815.0 MiB)] (x5) [Interval=5 sec] via crystaldiskmark x64
So looking at it the transcend only beats the kingston in speed at the random writing parts, but i find the speed in the pine64 from those 2 cards really different as the transcend loads things up faster and doest hang as much. Is the random write speed really that differentiating?
(04-30-2016, 02:55 AM)xrez Wrote: So looking at it the transcend only beats the kingston in speed at the random writing parts, but i find the speed in the pine64 from those 2 cards really different as the transcend loads things up faster and doest hang as much. Is the random write speed really that differentiating?
Funny, isn't it?
The 'faster' Kingston that clearly shows better sequential speeds is over 4 times slower regarding random I/O and therefore the whole Android system starts to stutter. This is the reason why Android devices normally have onboard eMMC that is magnitudes faster than your (still rather) slow Transcend card.
Combining slow storage (the average SD card pulled out of a camera) with Android/RemixOS is a simple recipe to get worst user experience ever. Well known since ages.
BTW: Same applies to 'Linux as Desktop', when you start to monitor I/O behaviour and watch Firefox for example opening up a couple of SQLite databases and being stuck in I/O it's also obvious that you have to take care of storage that shows high IOPS at random writes with small sector/block sizes. Speed class is close to irrelevant since this is only about large sequential reads/writes. And now take a look into Pine64's wiki or the so called 'freshman guide' what recommendations are written there.
(04-29-2016, 08:40 PM)xrez Wrote: Thanks for that Tkaiser. I'm still new at all of this and has been educational. Over here a 16gb evo card costs about $23 usd vs the cheapo Kingston at about $6 usd.
To test speeds you use H2testw or f3 in a first step (this tells you sequential speeds and whether your card is fake or not) and then CrystalDiskMark to test random I/O too (see the explanations here: http://thewirecutter.com/reviews/best-microsd-card/)
And you should always keep in mind that all Samsungs I tested are pretty slow directly after powering up. They need some time to get faster (maybe some calibration or something like that happens)
Plus shipping to Australia its about usd17 still. Hence why im trying to find an alternative to the samsung microsd cards
(04-30-2016, 03:18 AM)xrez Wrote: Hence why im trying to find an alternative to the samsung microsd cards
One last time: The most important factor for SBC (or Android device) performance is random I/O with small record/block sizes. This is not related to the 'speed class' ratings and cards of the same speed class show differences in the range of being 20 times slower (or even more).
So all you can do is to rely on real measurements done within the last few weeks/months. At Armbian we've done this, others have done this. Forget about any cards listed there that are older than 6 months since in the meantime newer controllers or NAND dies could be used. Here you find the results of different test runs: https://github.com/igorpecovnik/lib/blob...re-sd-card (the tests made by Armbian users being the 'best' for Pine64 since already taken the sequential speed limitation into account)
Both used Test : 50 MiB [D: 0.0% (0.0/4815.0 MiB)] (x5) [Interval=5 sec] via crystaldiskmark x64
So looking at it the transcend only beats the kingston in speed at the random writing parts, but i find the speed in the pine64 from those 2 cards really different as the transcend loads things up faster and doest hang as much. Is the random write speed really that differentiating?
Thank you for this. I now know that my issues are with my card. Here are my results:
This may be a dumb question, but how do we know that these tests, when we each test these micro SD cards at home are testing the micro SD cards and not our equipment's ability to read/write to them? For example. would I get the same results on the same card in a different computer? What if I used the same micro SD in a SD adapter (via my computer's built in card reader) vs in a USB SD card reader (VIA a USB port) on the same computer?
I realize that the point is to source a micro SD card with as fast a random I/O as possible, but how do we know what's fastest to the Pine64+ ?
05-01-2016, 03:39 AM (This post was last modified: 05-01-2016, 03:58 AM by tkaiser.)
(04-30-2016, 10:36 PM)NondescriptMember Wrote: For example. would I get the same results on the same card in a different computer?
Depends. For example: I've a few rather expensive SanDisk Extreme Pro (if you buy SanDisk either buy Pro or Plus, the 'normal' SanDisks show very low random I/O write performance especially at 16K sector size).
When I write OS images to them on my MacBook I get sequential write speeds of up to 80 MB/s (which is nice since burning images is fast). But if I use the card in Pine64 then sequential transfer speeds are limited to the Pine64's SDIO implementation (4 bit @ 50 MHz --> theoretical limit of 25MB/s, in reality it's 22-23 MB/s and sequential writes are a little bit slower than reads).
SEQUENTIAL transfer speeds are close to IRRELEVANT when we talk about SBC useage. So it's both useless to look after any product that is advertised with 'more than 25 MB/s' and sequential transfer speeds don't matter at all.
Get a fresh card that shows superiour random I/O performance and don't buy no-name crap (Kingston, PNY and the like -- they all buy cheap NAND dies and controllers on the spot markets and all you get are slow cards that just fulfil the irrelevant sequential speed class ratings)
(04-30-2016, 10:36 PM)NondescriptMember Wrote: I realize that the point is to source a micro SD card with as fast a random I/O as possible, but how do we know what's fastest to the Pine64+ ?
Random I/O performance depends not that much on the host's SDIO or USB implementation but on the controller used inside the SD card. So while sequential speeds differ between different hosts, random I/O performance varies only a little bit.
(04-30-2016, 08:02 PM)NondescriptMember Wrote: I now know that my issues are with my card. Here are my results:
Yes, very slow random I/O performance: very sluggish Android/RemixOS behaviour. Everything as expected.
The eMMC used in typical Android devices is magnitudes faster than the PNY card you use (even if sequential transfer speeds might look slower -- only random I/O matters)
Quote:If you want to be sure to get a useful test with pine, install Linux to the prospective cards and run the test in Linux on pine.
Thank you for that, hyperlogos; I know that is prudent advice, but I also know that do so far exceeds my capabilities.
Quote:don't buy no-name crap (Kingston, PNY and the like -- they all buy cheap NAND dies and controllers on the spot markets and all you get are slow cards that just fulfil the irrelevant sequential speed class ratings)
tkaiser, I've been following everything you been saying with regard to these SD cards, and I consider you a (if not the) subject matter expert. I know you're getting tired of saying the same thing over and over again. It's my practice to avoid the need for that. I believed that I was approaching the subject from a different perspective (not necessarily different than your own experiences, but different than has been discussed on this forum). I didn't read anything to suggest that PNY was crap (PNY is too small of a search term so I may have missed it). Thank you so much for your new insights and for your willingness to give thorough responses!
While this particular PNY card was the only 64gb card in stock at Walmart, I probably would have selected it by choice because it is marketed for use in Android devices (assuming the standard EVO card wasn't available). I truly thought PNY was a good product.
