Topic: Tascam DR-70D 4-channel audio recorder (Part 5)

[kleiner Rainer]

Hi all,

very interesting. This correlates with information I got from our manufacturers when selecting SPI FLASH for an embedded application. Another point to keep in mind is the difference in performance and write endurance between SLC (1 bit per cell) and MLC (2-3 bit per cell). MLC is cheaper and has less write endurance, but is preferred (size and price) for consumer applications:

http://www.supertalent.com/datasheets/SLC_vs_MLC%20whitepaper.pdf

The performance degradation in SD cards is also discussed here:

http://www.linuxplumbersconf.org/2010/ocw/proposals/387

Formatting FLASH cards can make a difference:

http://3gfp.com/wp/2014/07/formatting-sd-cards-for-speed-and-lifetime/

So to summarize: SD cards are not made equal. And I would not want to design a recorder that is expected to work with any card available 

Greetings,

Rainer

[stevetoney]

...then that indicates that either their performance really degrades badly over a year or two,

I don't see this.  The vertical scales aren't consistent from graph to graph, so one can't rely on visual comparisons of the graphs.  The analytical data on the bottom indicates that all of the Transcend cards except one have an average write speed in the 9mb/s range.  No trend conclusions can be made from that in terms of new vs. old cards. 

or they are simply far inferior to the SanDisk units.

Yeah, I think that's a valid conclusion from this as well, though I'm not sure the test necessarily indicates that Transcend cards shouldn't be used in many/most applications. Clearly the write speeds on the Sandisk cards are superior in all cases over the Trascend.

What I find most interesting about these is how the Transcend cards all drop off after 10% and then trend downward with increasing samples, while the Sandisk cards all process data uniformly throughout the test.  Clearly the two companies use different technology for writing data to their cards. 

Transcend cards seem to show some kind of a bottleneck effect as they fill up while Sandisk cards don't seem to be affected by the amount of data that's been written.  That seems consistent with the results some have reported here...that they didn't experience issues with their card in the DR70D until after some amount of time had passed into the recording session.

[johnmuge]

I just updated my firmware to V1.11, put in a new Sandisk 16GB SDSDUP-016G approved card and set up my mics.  Loaded up the house CD player with some Jerry Garcia Band and hit record.  It's a weekend test house party here !!

[stevetoney]

I just updated my firmware to V1.11, put in a new Sandisk 16GB SDSDUP-016G approved card and set up my mics.  Loaded up the house CD player with some Jerry Garcia Band and hit record.  It's a weekend test house party here !!

Hey John.  Here's a potential timesaving tip for you.  If you have another recorder besides the DR70D, set it up so that it's recording directly from the mics, then set your DR70D to record the output of the first recorder.  Once you've recorded a couple of hours of music, load both files into your DAW and compare the two on the same timeline.  If you have any issues, I think you might see that the length of the files are different.  If not, then they'll be close to the same length. 

On my DR70D, when it had problems, there was digi-noise but it also missed a small amount of music, so there would be a mis-match between the two files in that case.  I found this to be a less cumbersome method of checking for issues on the DR70D files than listening closely to two hours worth of music all over again.

[jcb]

^ ^ jcb, that's a very interesting test.  A couple things:

Did you happen to do an Erase / Full format on those Transcend cards before testing?  If not, that may improve their results.  If you did fully format them, then that indicates that either their performance really degrades badly over a year or two, or they are simply far inferior to the SanDisk units.

No the cards were not formatted before the test. I'll try to do this (full format/erase in the computer then run test) but full format/erase takes time...

Your test was random read / write of 250 10MB samples.  To me, that seems to be how they would be used in a camera.  From what I understand though, audio recording is a much more continuous write.  I wonder if you can find a test that does sustained writes.  I think such tests exist for hard drives, but I'm not sure if they work for flash media.

The test I ran was easy to operate and had easy to copy graphic results. It is indeed writing randomly on the card. I'll have a look to see if I can find some other test that writes sequentially.

(edited to repair the quotes)

[Life In Rewind]

^ ^ jcb, that's a very interesting test.  A couple things:

Did you happen to do an Erase / Full format on those Transcend cards before testing?  If not, that may improve their results.  If you did fully format them, then that indicates that either their performance really degrades badly over a year or two, or they are simply far inferior to the SanDisk units.

No the cards were not formatted before the test. I'll try to do this (full format/erase in the computer then run test) but full format/erase takes time...

Your test was random read / write of 250 10MB samples.  To me, that seems to be how they would be used in a camera.  From what I understand though, audio recording is a much more continuous write.  I wonder if you can find a test that does sustained writes.  I think such tests exist for hard drives, but I'm not sure if they work for flash media.

The test I ran was easy to operate and had easy to copy graphic results. It is indeed writing randomly on the card. I'll have a look to see if I can find some other test that writes sequentially.

I'd like to see this done using the FULL FORMAT function on DR-70D. (or even series done with QUICK vs FULL format...)

[jcb]

...then that indicates that either their performance really degrades badly over a year or two,

I don't see this.  The vertical scales aren't consistent from graph to graph, so one can't rely on visual comparisons of the graphs.  The analytical data on the bottom indicates that all of the Transcend cards except one have an average write speed in the 9mb/s range.  No trend conclusions can be made from that in terms of new vs. old cards. 

I read it a bit differently :

• You have to look to the instant response on the chart and not only to the average figure below as the bandwidth is much better at the beginning of the test than later. If you look towards the end of the graphs, with the most recent Transcends, the instant throughput is regularly in the 1MB/s range (on the downwards spikes).
• The access time scatter plot (the green goo like thing on the graph) is also interesting when you search out of the homogeneous cloud for the extreme values : less than 2msec for the SanDisks, up to 8msec for the Transcends.

In a point and shoot camera, this is no big deal, the image is buffered and the only consequence is that you wait a little longer before the camera is ready for the next picture. In a video camera or sound recorder this could result in a real problem when you hit a slow write patch.

or they are simply far inferior to the SanDisk units.

Yeah, I think that's a valid conclusion from this as well, though I'm not sure the test necessarily indicates that Transcend cards shouldn't be used in many/most applications. Clearly the write speeds on the Sandisk cards are superior in all cases over the Trascend.

I agree with this : the Transcends are just inferior. What is strange is that their read performance increases from one generation to the next while their instantaneous write performance degrades...

What I find most interesting about these is how the Transcend cards all drop off after 10% and then trend downward with increasing samples, while the Sandisk cards all process data uniformly throughout the test.  Clearly the two companies use different technology for writing data to their cards. 

Transcend cards seem to show some kind of a bottleneck effect as they fill up while Sandisk cards don't seem to be affected by the amount of data that's been written.  That seems consistent with the results some have reported here...that they didn't experience issues with their card in the DR70D until after some amount of time had passed into the recording session.

I suspect that there is some buffering inside the card. When this buffer is full the performance falls.

(Edited to repair the quotes structure)

[jcb]

I tried to do a full erase format to see if it really changed the results. The test program remains the same. The card is the most recent Transcend in the former sample so a class 10 UHS-I card.

The first test with the card (as it was after use) gave the following results :



After a full erase and format in my computer (it does seem logical to format in the machine used for the test) :



I ran a second same test just after but without the green plot so that the red plot can be seen better :



The improvement is not evident. The read seems smoother but the write throughput remains very irregular and there are still patches where the card does not write faster than 1MB/s. Remember that each point on the graph is an average write speed for a 10MB data write sequence. 10MB is more than 6 seconds of recording at 24/96 with all four channels meaning that one can have really slow writing on the card for more than 6 seconds of recording...

[vanark]

Wow, pretty dramatic difference between cards. It seems the recorder would need a more generous (and forgiving) buffer to deal with such an erratic write performance.

[stevetoney]

... No trend conclusions can be made from that in terms of new vs. old cards. 

I read it a bit differently :

• You have to look to the instant response on the chart and not only to the average figure below as the bandwidth is much better at the beginning of the test than later. If you look towards the end of the graphs, with the most recent Transcends, the instant throughput is regularly in the 1MB/s range (on the downwards spikes).
• The access time scatter plot (the green goo like thing on the graph) is also interesting when you search out of the homogeneous cloud for the extreme values : less than 2msec for the SanDisks, up to 8msec for the Transcends.

In a point and shoot camera, this is no big deal, the image is buffered and the only consequence is that you wait a little longer before the camera is ready for the next picture. In a video camera or sound recorder this could result in a real problem when you hit a slow write patch.

My point, to which you responded, was that the data doesn't support a conclusion that the older cards show a degrading trend against newer cards.  Your response along with the data from the Transcend cards actually seems to indicate that the newer cards have a lower write speed than the older cards.  It's reasonable to assume that the difference is due to design features, since it wouldn't be logical to expect an older card to be faster simply because of its age.

There's been alot of discussion in this thread  (and Tascam has hung their hat) on a concept that older cards develop bad sectors and degrade over time.  I don't doubt that's true, but I also don't see any trending in this test that supports that concept either.  That said, the age of a card can clearly be expressed in a couple of ways...a) the amount of times it's been written to, and b) the actual age in time of the card since it was new.  For this test, you've provided a bit of data for both a) and b), but I'm not seeing enough data to see any trends that indicates a strong correlation can be made between age and performance.  There's just not enough data to reach that conclusion.

For the Tascam cards, I'd bet that the downward trend in write speed as more data is written to the card (someone suggested that those cards don't have any buffering capability as the design difference between the Tascam and the Sandisk cards) is probably be a far more important factor in terms of card performance than how old the card is or how many bad sectors it has.

[jcb]

Your test was random read / write of 250 10MB samples.  To me, that seems to be how they would be used in a camera.  From what I understand though, audio recording is a much more continuous write.  I wonder if you can find a test that does sustained writes.  I think such tests exist for hard drives, but I'm not sure if they work for flash media.

Testing continuous write proves to be more of a challenge as the tools that I found will give only average data on huge file writing.

The Transcend card tested above (Transcend 32GB, used but not abused, Class 10, 600x, UHS-I bought in 2015) gave average write rates of 34 to 36MB/s when tested for continuous files writing with the f3 program (f3write). So we get the following profile for this card :

• continuous file writing average rate over several GB (seven tests in a row without reformatting between tests) : 34 to 36 MB/s
• random file writing average rate over several GB : 7 to 9 MB/s
• random file writing average rate over 10 MB : less than 1 to 40 MB/s

As I understand it, this card will work fine in a recorder until it meets a situation where writing slows down to a point that it will not provide sufficient throughput for the recorder. When such a situation occurs, the recorder buffer may provide a solution up to a point, however if one supposes that the recorder buffer size is more or less what is needed to provide the pre-record feature (5 seconds on most machines), then very low write rates over 10 MB of data may be a real problem.

Trying to put it simply an average 35MB/s bandwidth is not exactly the same thing if the is no instant when the throughput was less than 30MB/s and when there were times when it was close to 0MB/s.

Summary : average rates on which the card ratings (C10, UHS-I for instance) are based are fine but we may be more interested in worst case scenarios as shown by random file writes rates and we should prefer cards that guarantee a better minimum rate under any circumstances.

[stevetoney]

^ Great work, man.   

[phil_er_up]

Nice JCB. Thank you for your time and effort. It is appreciated.

[kleiner Rainer]

Hi all,

Wikipedia article:

https://en.wikipedia.org/wiki/Secure_Digital

Quote:

"In addition, speed may vary markedly between writing a large amount of data to a single file (sequential access, as when a digital camera records large photographs or videos) and writing a large number of small files (a random-access use common in smartphones). One study found that, in this random-access use, some Class 2 cards achieved a write speed of 1.38 MB/s, while all cards tested of Class 6 or greater (and some of lower Classes; lower Class does not necessarily mean better small-file performance), including those from major manufacturers, were over 100 times slower.[28]"

Writing multichannel or "safety" audio files simultaneous looks like random access to me - more than one file open at any instant.
The reference cited makes very interesting reading:

http://static.usenix.org/events/fast/tech/full_papers/Kim.pdf

Any chance to test random access writes with more than one file open with your tool, jcb?

Greetings,

Rainer

Concerned because using a Tascam DR44WL with a Lexar 64GB SDXC Class 10 card.

[jcb]

Any chance to test random access writes with more than one file open with your tool, jcb?
Greetings,
Rainer

I have not found any tool that would write multiple continuous files at the same time and time operations while doing so.

It might not be difficult to write a program to do this (open n files for writing > get time_0, write p kB to file_1, get time_1.1, write p kB to file_2, get time 1.2, ..., write p kB to file_n, get time 1.n, and back to writing to file _1, get time 1.n, ... > close files). Write file strategy would however have to be close to what the recorders do in real life (how many kB do they write at a time ? Is the routine as simple as this ?)

It might be better to open another thread if this discussion has some kind of future.