Topic: SD cards performance and testing

[jcb]

This subject was raised in the Tascam DR-70D discussion due to the perceived "pickiness" of this recorder regarding SD cards. It was time to start a dedicated thread to discuss what kind of performance we expect from memory cards and how to check that we get what we need.

For reference, the Tascam DR-70D thread hijack began around here : http://taperssection.com/index.php?topic=174900.msg2163449#msg2163449 and it was my fault.

So what do we know ?

• Most recorders come wit a list of approved/tested media. When using non-recommended cards, problems will be blamed on the card.
• A number of fake or dubious cards are being sold through different channels (ebay, ...)
• Recording one channel at 24/96 with 20% overhead requires writing 24 x 96 / 8 = 345 kB/s to the card.
• So 4 channels require roughly 1.5MB/s bandwith when writing to the card.
• Some recent class 10/UHS-I cards seem to have very good write performance (more than 60 MB/s) whether they have to write sequential or random data and they never slow down temporarily.
• Other recent class 10/UHS-I cards will dislay a decent average write rate for sequential data (35MB/s) that is more than needed but slow down considerably when writing random data (8MB/s) with at times a very limited bandwidth (less than 1MB/s) which may be a real problem.

What can we test (please excuse my ignorance : I don't do windows and never used a mac so I do not know many useful tools) ?

• Fake cards can be identified with H2testw on windows or f3 http://oss.digirati.com.br/f3/ on linux.
• f3 has an uility (f3write) that tests the average write speed of the card for 1GB files. Something similar may exist on windows.
• gnome_disks_utility ("disks" in the gnome menu) allows for random write testing on linux. Something similar probably exists for windows.

What can we (yet) not test ?

• A recorder writes 1 to many (depending on the number of channels) files at the same time. What are the cards performance in this case ? How might we test this ?

[aaronji]

I think fio will do what you need it to do (generating an arbitrary number of sequential files, like audio recording) and benchmark the card's performance.

I think your 20% overhead is probably highly conservative; where does that number come from?

[kleiner Rainer]

Did some tests with CrystalDiskMark Portable today:

Lexar 64GB, Class 10 SDXC:

-----------------------------------------------------------------------
CrystalDiskMark 5.0.3 (C) 2007-2015 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

   Sequential Read (Q= 32,T= 1) :    17.565 MB/s
  Sequential Write (Q= 32,T= 1) :    15.519 MB/s
  Random Read 4KiB (Q= 32,T= 1) :     3.259 MB/s [   795.7 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :     0.005 MB/s [     1.2 IOPS] 
         Sequential Read (T= 1) :    17.198 MB/s
        Sequential Write (T= 1) :    15.309 MB/s
   Random Read 4KiB (Q= 1,T= 1) :     2.987 MB/s [   729.2 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :     0.005 MB/s [     1.2 IOPS]   

  Test : 1024 MiB [I: 8.4% (5.0/59.6 GiB)] (x5)  [Interval=5 sec]
  Date : 2015/11/15 11:32:31
    OS : Windows 7 Starter SP1 [6.1 Build 7601] (x86)
 


Sandisk Ultra 32GB, Class 10 micro SD:

-----------------------------------------------------------------------
CrystalDiskMark 5.0.3 (C) 2007-2015 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

   Sequential Read (Q= 32,T= 1) :    18.742 MB/s
  Sequential Write (Q= 32,T= 1) :    17.801 MB/s
  Random Read 4KiB (Q= 32,T= 1) :     3.394 MB/s [   828.6 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :     1.969 MB/s [   480.7 IOPS]
         Sequential Read (T= 1) :    18.457 MB/s
        Sequential Write (T= 1) :    17.617 MB/s
   Random Read 4KiB (Q= 1,T= 1) :     3.148 MB/s [   768.6 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :     1.841 MB/s [   449.5 IOPS]

  Test : 1024 MiB [I: 0.0% (0.0/28.8 GiB)] (x5)  [Interval=5 sec]
  Date : 2015/11/18 22:30:42
  OS : Windows 7 Starter SP1 [6.1 Build 7601] (x86)
 

Seems like the "Random Write" performance is the limiting factor in uses other than sequential writing of a video/audio stream.

Greetings,

Rainer

[voltronic]

^ Wow!  I second your use of emoji there!  I'm assuming a fresh full format on each card before testing?

Thanks for linking this tool and your results - this looks better than some others out there.

[aaronji]

Seems like the "Random Write" performance is the limiting factor in uses other than sequential writing of a video/audio stream.

You can find some white papers describing this if you poke around on the web a bit...

Nice to see that the cards you tested were way above their speed class.

[voltronic]

Just tested my two cards with CrystalDiskMark.  Quite the difference:

Transcend 16GB SDHC Class 10 16GWACA-MG

-----------------------------------------------------------------------
CrystalDiskMark 5.0.3 x64 (C) 2007-2015 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

   Sequential Read (Q= 32,T= 1) :    21.761 MB/s
  Sequential Write (Q= 32,T= 1) :    20.292 MB/s
  Random Read 4KiB (Q= 32,T= 1) :     3.100 MB/s [   756.8 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :     0.018 MB/s [     4.4 IOPS]
         Sequential Read (T= 1) :    21.604 MB/s
        Sequential Write (T= 1) :    18.457 MB/s
   Random Read 4KiB (Q= 1,T= 1) :     3.030 MB/s [   739.7 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :     0.011 MB/s [     2.7 IOPS]

  Test : 1024 MiB [H: 6.1% (0.9/15.1 GiB)] (x5)  [Interval=5 sec]
  Date : 2015/11/19 20:52:48
    OS : Windows 7 Ultimate SP1 [6.1 Build 7601] (x64)
 
---

SanDisk Ultra Plus 16GB SDHC Class 10 UHS-1 SDSDUP-016G-A46

-----------------------------------------------------------------------
CrystalDiskMark 5.0.3 x64 (C) 2007-2015 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes

   Sequential Read (Q= 32,T= 1) :    37.332 MB/s
  Sequential Write (Q= 32,T= 1) :    18.430 MB/s
  Random Read 4KiB (Q= 32,T= 1) :     5.409 MB/s [  1320.6 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :     0.299 MB/s [    73.0 IOPS]
         Sequential Read (T= 1) :    36.916 MB/s
        Sequential Write (T= 1) :    19.716 MB/s
   Random Read 4KiB (Q= 1,T= 1) :     5.305 MB/s [  1295.2 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :     0.228 MB/s [    55.7 IOPS]

  Test : 1024 MiB [H: 0.0% (0.0/14.8 GiB)] (x5)  [Interval=5 sec]
  Date : 2015/11/19 21:03:05
    OS : Windows 7 Ultimate SP1 [6.1 Build 7601] (x64)
 

[F.O.Bean]

I honestly dont understand what any of that data means I just used FakeZFlashTest and H2TestW with success on a 32gb PNY High Performance 90MB/s Class 10! Trying my 64gb Elite Performance 95MB/s Class 10 here next I'll report back! This will certainly come in handy when I get my new 64gb PNY Elite 95MB/s Class 10 coming this Friday, for my 2nd 70d

[n3mmr]

I think, and the tascam dr680mkII tested media list seems to support this notion, that the most serious problems will occur if the card has been filled up with lots-of-little-files-all-the-same.

So I would suggest testing first on a fresh-format card, and then, if the card passes muster in the fresh state,  fill the card up to near full with varying size sequential files. Aim for like 5-10% or 1GB free, whichever is larger. And then test again. That seems like reasonable for a state of "just one more take, please".

[DancyGeorgia]

I think, and the tascam dr680mkII tested media list seems to support this notion, that the most serious problems will occur if the card has been filled up with lots-of-little-files-all-the-same.

So I would suggest testing first on a fresh-format card, and then, if the card passes muster in the fresh state,  fill the card up to near full with varying size sequential files. Aim for like 5-10% or 1GB free, whichever is larger. And then test again. That seems like reasonable for a state of "just one more take, please".

Lots of small files in a single folder is a problem for any system. Of course, the number of files at which lots of small files becomes a problem is dependent on the operating system, file system, and processor. It seems reasonable that filling up a large GB disk with lots of small files in one folder would cause problems. Does the tascam dr680mkII allow multiple folders?

[n3mmr]

I think, and the tascam dr680mkII tested media list seems to support this notion, that the most serious problems will occur if the card has been filled up with lots-of-little-files-all-the-same.

So I would suggest testing first on a fresh-format card, and then, if the card passes muster in the fresh state,  fill the card up to near full with varying size sequential files. Aim for like 5-10% or 1GB free, whichever is larger. And then test again. That seems like reasonable for a state of "just one more take, please".

Lots of small files in a single folder is a problem for any system. Of course, the number of files at which lots of small files becomes a problem is dependent on the operating system, file system, and processor. It seems reasonable that filling up a large GB disk with lots of small files in one folder would cause problems. Does the tascam dr680mkII allow multiple folders?

Yes it but Tascam seems to say something indicating there is an effect that persists even if you create and work in  a new empty folder on an otherwise near-full card.
They do warn against going beyond 100 files in a folder, but they also say some " bad" cards may slow down from just being used, even if you delete old takes.

[daspyknows]

I always use new 8 or 16gb sandisk extreme cards in my dr2d