Page 2 of 13

Re: SD Card Benchmarks

Posted: Thu Mar 15, 2012 10:22 pm
by delicious-pi
Posted from my other thread

4GB version of this card

http://www.memorybits.co.uk/sh.....gu46/17800

SanDisk 4GB Ultra (30MB/s) SDHC Class 6 Memory Card

Through memory card reader on laptop



Re: SD Card Benchmarks

Posted: Thu Mar 15, 2012 10:27 pm
by Kernel
delicious-pi said:


Posted from my other thread

4GB version of this card

http://www.memorybits.co.uk/sh.....gu46/17800

SanDisk 4GB Ultra (30MB/s) SDHC Class 6 Memory Card

Through memory card reader on laptop






Some nice results - I'm starting to think I should buy a class 6 card - nice speed/not too expensive and don't feel like buying a class 10 - I already bought 3 class4's though!

Re: SD Card Benchmarks

Posted: Thu Mar 15, 2012 10:30 pm
by delicious-pi
Kernel said:


delicious-pi said:


Posted from my other thread

4GB version of this card

http://www.memorybits.co.uk/sh.....gu46/17800

SanDisk 4GB Ultra (30MB/s) SDHC Class 6 Memory Card

Through memory card reader on laptop





Some nice results - I'm starting to think I should buy a class 6 card - nice speed/not too expensive and don't feel like buying a class 10 - I already bought 3 class4's though!


Get the card i got from play, then you might hit the jackpot like me You could always sell the other cards on.

Re: SD Card Benchmarks

Posted: Thu Mar 15, 2012 10:35 pm
by Kernel
delicious-pi said:

Get the card i got from play, then you might hit the jackpot like me You could always sell the other cards on.
Yeah I could - even if I ended up with the class 4 a class 4 ultra is probably better than a normal class4 anyway

I could - I'll just keep them though - never know when you might need one

Re: SD Card Benchmarks

Posted: Fri Mar 16, 2012 1:53 am
by fakeplastic
Thought I'd post my results for everyone's benefit. I tested these all on the same card reader and same computer. Short version, Sandisk Ultra 8GB is awesome.

512MB Sandisk

Sequential Read :    10.343 MB/s

Sequential Write :     2.960 MB/s

Random Read 512KB :    10.245 MB/s

Random Write 512KB :     1.873 MB/s

Random Read 4KB (QD=1) :     3.210 MB/s [   783.8 IOPS]

Random Write 4KB (QD=1) :     0.042 MB/s [    10.3 IOPS]

Random Read 4KB (QD=32) :     3.472 MB/s [   847.8 IOPS]

Random Write 4KB (QD=32) :     0.041 MB/s [    10.0 IOPS]

Test : 50 MB [E: 0.0% (0.0/483.6 MB)] (x5)

2gb atp max pro

Sequential Read :    18.636 MB/s

Sequential Write :    16.780 MB/s

Random Read 512KB :    18.613 MB/s

Random Write 512KB :     5.079 MB/s

Random Read 4KB (QD=1) :     5.218 MB/s [  1273.8 IOPS]

Random Write 4KB (QD=1) :     0.052 MB/s [    12.8 IOPS]

Random Read 4KB (QD=32) :     6.154 MB/s [  1502.5 IOPS]

Random Write 4KB (QD=32) :     0.055 MB/s [    13.5 IOPS]

Test : 50 MB [E: 0.0% (0.0/1962.1 MB)] (x5)

8GB Retailplus SDX10 Class 10 (SDHC)

Sequential Read :    19.342 MB/s

Sequential Write :    12.488 MB/s

Random Read 512KB :    18.875 MB/s

Random Write 512KB :     1.158 MB/s

Random Read 4KB (QD=1) :     3.716 MB/s [   907.3 IOPS]

Random Write 4KB (QD=1) :     0.014 MB/s [     3.4 IOPS]

Random Read 4KB (QD=32) :     3.790 MB/s [   925.3 IOPS]

Random Write 4KB (QD=32) :     0.014 MB/s [     3.3 IOPS]

Test : 50 MB [E: 0.0% (0.0/7613.0 MB)] (x5)

8gb sandisk Ultra Class 6 (SDHC)

Sequential Read :    19.425 MB/s

Sequential Write :     9.116 MB/s

Random Read 512KB :    19.217 MB/s

Random Write 512KB :     3.779 MB/s

Random Read 4KB (QD=1) :     4.251 MB/s [  1037.9 IOPS]

Random Write 4KB (QD=1) :     1.839 MB/s [   448.9 IOPS]

Random Read 4KB (QD=32) :     4.464 MB/s [  1090.0 IOPS]

Random Write 4KB (QD=32) :     1.562 MB/s [   381.3 IOPS]

Test : 50 MB [E: 0.0% (0.0/7572.0 MB)] (x5)

1gb Kingston  MicroSD

Sequential Read :    10.266 MB/s

Sequential Write :     3.942 MB/s

Random Read 512KB :    10.265 MB/s

Random Write 512KB :     2.557 MB/s

Random Read 4KB (QD=1) :     4.137 MB/s [  1010.0 IOPS]

Random Write 4KB (QD=1) :     0.053 MB/s [    12.9 IOPS]

Random Read 4KB (QD=32) :     4.609 MB/s [  1125.4 IOPS]

Random Write 4KB (QD=32) :     0.055 MB/s [    13.5 IOPS]

Test : 50 MB [E: 0.0% (0.0/982.2 MB)] (x5)

  2gb micro SD

Sequential Read :    16.090 MB/s

Sequential Write :     6.599 MB/s

Random Read 512KB :    14.827 MB/s

Random Write 512KB :     1.611 MB/s

Random Read 4KB (QD=1) :     1.395 MB/s [   340.5 IOPS]

Random Write 4KB (QD=1) :     0.089 MB/s [    21.7 IOPS]

Random Read 4KB (QD=32) :     1.555 MB/s [   379.5 IOPS]

Random Write 4KB (QD=32) :     0.080 MB/s [    19.6 IOPS]

Test : 50 MB [E: 0.0% (0.0/1905.7 MB)] (x5)

Re: SD Card Benchmarks

Posted: Fri Mar 16, 2012 2:25 am
by danielverza
I'm seeing different results for the same card in different readers.

Take a look:





These are both from the same card (a very old 2GB Sandisk Ultra II) on different laptops.

I have repeated the operation several times with different cards and the results are consistent.

So, while this works for comparing your own cards against each other, it's pointless for comparing with another person.

Re: SD Card Benchmarks

Posted: Fri Mar 16, 2012 2:32 am
by liam.lah
As in the Nook Color benchmark thread http://forum.xda-developers.co.....?t=1005633

There were varying results. But with enough results there was a fairly clear result. Note that all but 2 of the sandisk class 2 and 4's were in the top half of the results. With the top half completely stacked with that particular card.

With enough results, these inconsistencies can be smoothed out.

Re: SD Card Benchmarks

Posted: Fri Mar 16, 2012 10:48 am
by Kernel
liam.lah said:


As in the Nook Color benchmark thread http://forum.xda-developers.co.....?t=1005633

There were varying results. But with enough results there was a fairly clear result. Note that all but 2 of the sandisk class 2 and 4's were in the top half of the results. With the top half completely stacked with that particular card.

With enough results, these inconsistencies can be smoothed out.



Interesting I actually have some Sandisk Class 4's aswell - but I have only opened up the Kingston packaging so far - seems I might be better off with the Sandisk's

Re: SD Card Benchmarks

Posted: Fri Mar 16, 2012 2:01 pm
by jamesh
I'm just putting an image on a Sandisk Ultra SDHC vard 8GB, class 4. For demoing on a alpha board. Will see how I get on.

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 7:26 am
by semtex
Went to my local Costco in Canada and got a twin pack Sandisk Ultra 8Gb Class 6 cards (that's 2 cards for $26... which I think is a good deal).

Anyway the performance test results, especial the 4k read/write tests, are very promising for RasPi.



Unlike the 'cheap' brands I normally use, a nice bonus was that these cards came with labels to write distro names etc.

However, this card might be a Costco special. Look at the photo below. It's a Class 6 "Ultra" card but the read speed is 30MB/s!! Normally Ultra is 20MB/s. I also searched the Sandisk site and this card is not listed! Weird.


Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 10:54 am
by AlexPT
From what I seen so far from google search's, the best cards are from Sandisk Class 6.

But we have the Extreme and Ultra series.. Don't know whit best…

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 4:25 pm
by sorinm
4 GB SDHC KINGSTON class 4 

4 GB microsdhc Kingmax class 4 

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 4:44 pm
by poing
Daniel Ferreira said:


I"m seeing different results for the same card in different readers.


I think the energy put in this testing would be more rewarding if it was aimed at creating or agreeing on a standard test to run on a Pi once available.

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 5:56 pm
by Chromatix
People reading this thread might have missed the benchmarks I ran previously.  This has just been updated with a few extra cards in my collection.

http://dl.dropbox.com/u/601111.....hmarks.pdf

I ran these using a TrimSlice, which is a slightly more powerful (and considerably more expensive) ARM based box.  It runs from an internal disk, but has a built-in SD card slot.

On the left you can see the results of a synthetic test similar to the ones posted above.  On the right you can see the results of practical tests using a real ext4 filesystem and the contents of a real R-Pi filesystem.  As you can see, the two don't always correspond very well.

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 6:19 pm
by Kernel
Thanks Chromatix - it seems that from seeing the results from quite a few different people the Sandisk Class 6 Ultra and Extreme cards seem to be very good overall.

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 8:28 pm
by semtex
This is excellent information Chromatix. Thank you for taking the time to do this.

I was surprised to notice such a big difference in the practical test results between the Sandisk blue 4G C4 card and the Sandisk blue 16G C4 card. Apart from capacity was there any other difference between these two cards?

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 10:29 pm
by AlArenal
SanDisk Ultra 4GB Class on Windows 7, running in Parallels 6 on Mac OS X 10.7.3


Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 10:58 pm
by semtex
Hi Alex
If you get time maybe you could rerun the test and edit your post with the new results. Set the file size from 1000MB to 50MB... to be consistent with the other posted results. The 4k write speed is much lower that what other tests have indicated. (I"ve used Parallels myself and know it is very good... But it may be significantly influencing the results in your case).
What class is your card?

Re: SD Card Benchmarks

Posted: Sun Mar 18, 2012 11:17 pm
by Chromatix
semtex said:


This is excellent information Chromatix. Thank you for taking the time to do this.

I was surprised to notice such a big difference in the practical test results between the Sandisk blue 4G C4 card and the Sandisk blue 16G C4 card. Apart from capacity was there any other difference between these two cards?


Internally, I suspect that the 4GB card uses 2-bit MLC flash, perhaps from an older fab which is cheap to run, while the 16GB card uses 3-bit MLC - certainly the price I got it for strongly suggests that.  This results in a considerably lower sequential throughput for the larger card, because 3-bit MLC takes *much* longer to erase than 2-bit MLC or (1-bit) SLC flash.  You'll notice that it obtains Class 4 throughput with only a little to spare.  It's a lot faster than that for reading.

The non-sequential performance of the larger card is still relatively good, suggesting that SanDisk have still used their advanced controller chip with it - a good way of attaining the other requirements of the Class 4 spec.  This would also help a great deal with the longevity of the back-end flash, since the advanced controller can reduce write amplification by a lot compared to the more basic ones out there.

You see this a lot with SD cards.  Notice that the 32GB Transcend Class 6 is similarly slower than the 4GB Transcend Class 6.

This is in direct contrast to other drives, which tend to increase in performance as they get bigger.  With mechanical disks this is because there is more data per inch on each track, which is read at a constant *physical* speed, and the tracks are closer together too so the head has less far to move to each file on average.  With SSDs the effect is smaller but is due to having more physical flash chips able to operate in parallel, so multiple erase and write operations can be going on at the same time.

Re: SD Card Benchmarks

Posted: Fri Mar 23, 2012 3:09 pm
by yaconsult
These results were disappointing.  4K write speeds are terrible!  This is a brand new AmazonBasics SDHC Class 10 16 GB Secure Digital Card.  Tests were rerun several times.  Going to stop by Costco to pick up an Extreme card to compare.


Re: SD Card Benchmarks

Posted: Fri Mar 23, 2012 3:36 pm
by semtex
Hello yaconsult. You may want to consider Sandisk Ultra cards and not Extreme cards. We saw most recently at http://www.raspberrypi.org/archives/837 that there are issues with class 10 cards and microSD cards with adapters.

Class 4 and class 6 seem best. Have a look at the earlier post in the thread from Chromatix.

Regards

Re: SD Card Benchmarks

Posted: Fri Mar 23, 2012 5:48 pm
by yaconsult
Yes, that's exactly what I'm going to do.  Did I say Extreme?  If so, I meant Ultra.  Costco has the 16 GB Sandisk Ultra MicroSD Card with Adapter for $17.99 currently.  Gonna grab one today and will post results later.  There seemed to be some discussion about whether micro-cards with adapters would be an issue in the Pi or not but I believe the consensus was that it should work fine.  Nice thing about Costco is their return policy!

Re: SD Card Benchmarks

Posted: Tue Mar 27, 2012 5:51 am
by semtex
Duracell 4Gb Class 4 SD card. Look at the 4k write speeds. The test took so long to run I thought my PC had crashed!

This currently might be the worst card to use with a RasPi!!


Re: SD Card Benchmarks

Posted: Wed Mar 28, 2012 5:53 pm
by fakeplastic
semtex said:


Duracell 4Gb Class 4 SD card. Look at the 4k write speeds. The test took so long to run I thought my PC had crashed!

This currently might be the worst card to use with a RasPi!!


Whoa! I am glad you posted this. I saw these 4gb Class 4 cards for $5 and was thinking of grabbing a couple to have for multiple distributions, but now I think I will pass.

I might go pick up another of the 8gb Sandisk Ultras from Costco (2 pack). They seems the best so far.

Re: SD Card Benchmarks

Posted: Wed Mar 28, 2012 6:03 pm
by Enlightenment
Conceptually, the SD / miniSD / microSD are the same.  Technically, they could put the same dice / IC chips in any of the packages.

All miniSD to SD and microSD to SD adapters are passive and nothing more than a connector for the smaller format connected with wires/PCB to the larger SD.

Using an adapter to conver miniSD or microSD to SD form factor would change the electrical characteristic slightly because of an additional connector and connection in between them slightly.  Impendance differences and characteristics should be slightly different but just slightly.

The write-protect switches don't have anything to do with the data signals.  The detection for WP is simple leaf-switch in the socket.  It is up to the host to decide what to do with the WP.