SD Card Benchmarks

[Sandbird]

Has anyone tested this card ? Its got 90mb/sec.
I want to get the best card for my pi 2...and also be ready for windows 10 when it comes out.
http://www.ebay.com/itm/Transcend-Micro ... 19f2a8cea4

[Marty]

I think FAT location matters as well since it is optimized in many SD cards, not sure how you change that though, also did you try with a 32K cluster because that seems to be the native write-unit of this particular card ? Not sure if you could also try the page size of 8K.

I tried them all. Fat32 with 16 KB cluster size was best. I can't get the performance back from the store bought format. Would certain locations on the card be faster than others? and if so what locations are fastest?

[Sleep Mode zZ]

Has anyone tested this card ? Its got 90mb/sec.
I want to get the best card for my pi 2...and also be ready for windows 10 when it comes out.

You won't benefit from that 90 MB/s speed because the Pi maxes out at 22 MB/s. (And 'Windows 10' for the RPi is probably not what you expect it to be. Take a look at how the stripped down version of Windows 8.1 is currently used on the Intel Galileo and expect something similar for the RPi.)

[Mikael]

Here's a decent one for you. It's a Samsung Pro 32GB µSDHC UHS-I card fresh out of the box. It was tested in my 2013 MacBook Air running Windows 7, with its built-in card reader (which hangs on USB3 internally).

It performs really well, but it will obviously not be as fast in the Pi. I have a brand new Pi 2 that I will use this with. I'll see if I can return with some actual numbers from using it with the Pi.

[Sandbird]

Has anyone tested this card ? Its got 90mb/sec.
I want to get the best card for my pi 2...and also be ready for windows 10 when it comes out.

You won't benefit from that 90 MB/s speed because the Pi maxes out at 22 MB/s. (And 'Windows 10' for the RPi is probably not what you expect it to be. Take a look at how the stripped down version of Windows 8.1 is currently used on the Intel Galileo and expect something similar for the RPi.)

Thanks for the answer.
I've been reading here that some class 4 cards are better than class 10, when it comes to small files and random read/write. I guess since most people use the pi for a media center, small .xml files from kodi would benefit from this. Am i right?
So its better to get a class 4 instead of a class 10 card?
Does anyone have a suggestion for an 8gb card that they are actually satisfied with ? The benchmarks are a good source of information...but personal opinions are really the best deal.

[Sleep Mode zZ]

Has anyone tested this card ? Its got 90mb/sec.
I want to get the best card for my pi 2...and also be ready for windows 10 when it comes out.

You won't benefit from that 90 MB/s speed because the Pi maxes out at 22 MB/s. (And 'Windows 10' for the RPi is probably not what you expect it to be. Take a look at how the stripped down version of Windows 8.1 is currently used on the Intel Galileo and expect something similar for the RPi.)

Thanks for the answer.
I've been reading here that some class 4 cards are better than class 10, when it comes to small files and random read/write. I guess since most people use the pi for a media center, small .xml files from kodi would benefit from this. Am i right?
So its better to get a class 4 instead of a class 10 card?

I have also read the same and I'm sure that some class 4 cards are better than some class 10 cards - and some are not. My Samsung class 10 was much better than my Kingston class 4. I think it would be best to disregard the class rating completely and try to find some relevant information about their performance. Small random read and write could be the most relevant measurement but I don't think that manufacturers releases such information. Sequential read is good to be at least the 22 MB/s that the Pi can handle. I actually got myself a card with "up to 90 MB/s" read speed because it was not much pricier and its speed might be good to have if/when it ends in some other device that can actually benefit from its speed.

Does anyone have a suggestion for an 8gb card that they are actually satisfied with ? The benchmarks are a good source of information...but personal opinions are really the best deal.

I would trust more benchmarks than personal opinions unless the personal opinion comes from someone who actually has systematically compared the performance of many cards. You can't really know if something is fast or slow without having something to compare it with.

[dom]

I've been reading here that some class 4 cards are better than class 10, when it comes to small files and random read/write. I guess since most people use the pi for a media center, small .xml files from kodi would benefit from this. Am i right?

Yes

So its better to get a class 4 instead of a class 10 card?

Well, it depends. A class 4 with high 4K read/write scores will likely perform better than a class 10 with low 4K read/write scores.
But not all class 4 cards have good 4K read/write scores, and not all class 10 cards have low 4K read/write scores.

Basically ignore the class (which define continuous write speed which is largely irrelevant) and just look for fast 4K read/write numbers.

[Sandbird]

Yeah true, I forgot to mention that i do have a Raspberry pi model B (512mb) and since i want to upgrade to the pi 2, i was gonna get a micro sd for it.
Obviously i will try both cards on the pi 1 first and compare the speeds...thats why i thought someone else might have done this, and has something to say about it.
Unfortunately i cant use amazon...so i have to find something on ebay...and cross my fingers its not a fake card

[Sandbird]

Based on this review here: http://www.tomshardware.com/reviews/mic ... 11-12.html
I guess the Samsung Plus (8GB, Class 6) looks like the most promising card. I think i'll go with that one.

I am going for the Kingston microSDHC Card (Class 10, 16 GB)

[BenPope]

My iozone3 scores look like this:

Code: Select all

                                                            random  random
              KB  reclen   write rewrite    read    reread    read   write
           51200       4    1442    1733     5617     5626    5548     783
           51200     512    6810    7519    21921    21923   21898    2118
           51200   16384    8114    8826    22408    22407   22405   10626

8GB Sandisk Ultra 30MB/s Class 6

That was on the Pi B.

Retested today (iozone3 4.30):

Code: Select all

                                                   random    random
   kB  reclen    write  rewrite    read    reread    read     write
51200       4     1308     1527     4825     4823     4350      400
51200     512     5709     6912    18680    18707    18623     1813
51200   16384     6138     8504    18956    18988    18983     8169

This is on Pi 2 with Samsung 64GB Evo (MB-MP64D)

Code: Select all

                                                   random    random
   kB  reclen    write  rewrite    read    reread    read     write
51200       4     2288     3105     7691     7694     7618     3427
51200     512    16959    16833    19079    19115    19110    18139
51200   16384    18353    18415    19276    19297    19300    18336

That makes me happy.

[guilhermino]

As I was trying to say some cards have the area reserved for the FAT table optimized, as explained on one of the links I've posted, however the only way that I know to find out where it resides is using the flasbench linux utility, though I haven't been any fortunate at it, but in case you want to try, the utility and some info is on github, you just have to google it.

[jimmymcmahon]

Hi guys.

So I tried an sd only install of kodi this weekend.
Using an old class 4 micro sdhc card I had.

On my previous pi I have always used an sd card + usb3.0 setup.
In terms of responsiveness through the menus etc I agree the pi (sd only installation) performs well (albeit with a few temporary lock ups when add ons are updating in the background)

The biggest and most frustrating difference I've noticed from SD only vs usb 3.0 install is the time taken to scan a series in my library from tvdb etc.

In both the below I tested with no overlcock and then the same again with turbo overclock enabled.

If I use a well known series as an example. If I use the "add to library" function on this series for example on the usb 3.0 setup it will add them to the library and scan tvdb for all the info etc in about 10 seconds flat.

On the sd card, the same operation is taking about 10 minutes to complete the library scanning for the same series. I'm talking significant differences in speed. It also causes the pi to become pretty much unresponsive during that process.

Anyone else noticing this behaviour ?

Thanks

[dom]

So I tried an sd only install of kodi this weekend.
Using an old class 4 micro sdhc card I had.
...
The biggest and most frustrating difference I've noticed from SD only vs usb 3.0 install is the time taken to scan a series in my library from tvdb etc.
...
Anyone else noticing this behaviour ?

Yes, as keeps getting mentioned here the random access 4K read/write benchmark is key to performance.
I suspect your old sdcard has very poor 4K writes, as that is key when updating the library in kodi.

There is no necessity for USB - a good sdcard like the NOOBS card will have good database update times.

[user007]

Here's a decent one for you. It's a Samsung Pro 32GB µSDHC UHS-I card fresh out of the box. It was tested in my 2013 MacBook Air running Windows 7, with its built-in card reader (which hangs on USB3 internally).

It performs really well, but it will obviously not be as fast in the Pi. I have a brand new Pi 2 that I will use this with. I'll see if I can return with some actual numbers from using it with the Pi.

I would prefer ATTO Disk Benchmark Tool for testing SD cards.
Here are some micro SD card testing results (Sony VAIO Pro 13, Windows 8.1, built-in card reader, FAT32 formatting):

Transcend microSDHC 8GB class10


Samsung microSDXC 32GB class10 EVO


Samsung microSDXC 64GB class10 EVO

[jimmymcmahon]

Just tested a new card and got some crazy good numbers...

Samsung 16GB Class 10 Grade 1 48MB/s Micro SDHC Plus (MB-MPAGCAEU)
http://www.amazon.co.uk/Samsung-Class-M ... B00D6ENF7M £10.64

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 46.010 MB/s
Sequential Write : 36.343 MB/s
Random Read 512KB : 45.393 MB/s
Random Write 512KB : 36.277 MB/s
Random Read 4KB (QD=1) : 12.400 MB/s [ 3027.4 IOPS]
Random Write 4KB (QD=1) : 3.445 MB/s [ 841.0 IOPS]
Random Read 4KB (QD=32) : 11.771 MB/s [ 2873.9 IOPS]
Random Write 4KB (QD=32) : 3.591 MB/s [ 876.8 IOPS]

Test : 50 MB [F: 0.0% (0.0/14.6 GB)] (x5)
Date : 2014/07/26 19:10:27
OS : Windows 7 Enterprise SP1 [6.1 Build 7601] (x64)

Just got this card after the above figures were quoted
I can confirm I am seeing great results too! I have managed to install openelec 5.0.3 image on my rpi2 without issues and navigation around kodi / library scanning appears to be better than my previous USB 3.0 + SD installation.

Great card for Kodi - would recommend to replace any existing USB 3.0 + SD card setup as the USB stick really isnt required anymore!

below are my CrystalDiskMark results on my i7 running windows 8 for reference.

[jimmymcmahon]

So I tried an sd only install of kodi this weekend.
Using an old class 4 micro sdhc card I had.
...
The biggest and most frustrating difference I've noticed from SD only vs usb 3.0 install is the time taken to scan a series in my library from tvdb etc.
...
Anyone else noticing this behaviour ?

Yes, as keeps getting mentioned here the random access 4K read/write benchmark is key to performance.
I suspect your old sdcard has very poor 4K writes, as that is key when updating the library in kodi.

There is no necessity for USB - a good sdcard like the NOOBS card will have good database update times.

Many thanks for the response, as per my previous post I can now confirm brilliant kodi response / scanning times with this card

http://www.amazon.co.uk/Samsung-Class-M ... B00D6ENF7M - Happy!

[Phills]

Hello All,

What software do you recommend for Mac to run SD Card Benchmarks?

[Glenn N. Davis]

Somewhere I read online that the issue with SD sped is the marvels of backward compatabillity. If u use a card with a speed u can't use it will backward compatible read and write to the earliest standard! Wonder full that it works at all! But don't expect speed in this mode!

So what to do?

My raspberry pi2B claims in the instructions something about class 4 referring to minimums.
It makes no claims to read or write any faster in class 10 or UHS so reading that some user has troubles with class 10. I bought class 4 16GB from microcenter store branded. I haven't fired it up yet.

I hope to find it out what is what is what soon:-) .

[Glenn N. Davis]

Somewhere I read online that the issue with SD sped is the marvels of backward compatabillity. If u use a card with a speed u can't use... .

My question is whas software would get to the bottom of this question?

Any recommendations?

[dom]

Just tested a new card and got some crazy good numbers...

Samsung 16GB Class 10 Grade 1 48MB/s Micro SDHC Plus (MB-MPAGCAEU)
http://www.amazon.co.uk/Samsung-Class-M ... B00D6ENF7M £10.64

I've tested the 32G version of this card and performance is great with Kodi.

[jimmymcmahon]

For anyone interested here is the model / packaging I was sent to make sure you tet the same card if you order one

[LitterBugs]

It would make sense to use a "standard" sd test method run on the Raspberry Pi itself. I see a wide variety of benchmarks on here and not many look to have been generated on the Raspberry Pi. Is there a "standard" tool that we could use to create a database of real Pi statistics?

Greetings!
Karl

Hackin' since the 70's, new to Pi

[mentat]

It would make sense to use a "standard" sd test method run on the Raspberry Pi itself. I see a wide variety of benchmarks on here and not many look to have been generated on the Raspberry Pi. Is there a "standard" tool that we could use to create a database of real Pi statistics?

Greetings!
Karl

Hackin' since the 70's, new to Pi

I agree... and find that 4K write speeds (reported by CrystalDiskMark) are not realized on the Pi or Pi2
So, after reading several benchmark guides, including this one: http://elinux.org/RPi_SD_cards#Terminal_Commands
I use the following commands (I just saved them to file mmctest.sh in /storage and made it executable):

systemctl stop kodi
sync
dd if=/dev/zero of=test_500M_512K.tmp bs=512K count=1000
dd if=test_500M_512K.tmp of=/dev/null bs=512K count=1000
sync
dd if=/dev/zero of=test_500M_004K.tmp bs=4K count=128000
dd if=test_500M_004K.tmp of=/dev/null bs=4K count=128000
sync

I am attemtping to duplicate the 512K and 4K tests referenced by CrystalDiskMark users.

My results, using the stock NOOBS micro SD Card (8GB, Class 6) shipped with the RPi2 look like this:

512K Write
524288000 bytes (500.0MB) copied, 85.968726 seconds, 5.8MB/s
512K Read
524288000 bytes (500.0MB) copied, 28.402366 seconds, 17.6MB/s
4K Write
524288000 bytes (500.0MB) copied, 80.650916 seconds, 6.2MB/s
4K Read
524288000 bytes (500.0MB) copied, 28.477256 seconds, 17.6MB/s

Any suggestions or improvements would be appreciated.

Thanks.

[hglm]

It would make sense to use a "standard" sd test method run on the Raspberry Pi itself. I see a wide variety of benchmarks on here and not many look to have been generated on the Raspberry Pi. Is there a "standard" tool that we could use to create a database of real Pi statistics?

I am sure there are better tools, but a while ago I wrote a Linux-based flash benchmark tool called "flash-bench" for the purpose of measuring (Micro-)SD card and USB stick performance. It is available at https://github.com/hglm/flash-bench. Run

Code: Select all

git clone https://github.com/hglm/flash-bench.git
cd flash-bench
make

to install and compile it. Then run

Code: Select all

sudo ./flash-bench

to perform a benchmark run. It reports sequential read and write, and random (4K) read and write throughput in a consistent manner. Run with the ' -h' option for a list of options, the README file contains a detailed description of all options.

Some benchmarks I performed on a "high-performance" Kingston Class 10 UHS 1 Micro-SD card:

Code: Select all

                                          Seq.    Seq.    Random  Random
                                          Read    Write   Read    Write
    PC (using USB SD card adapter)        18.1    15.7     3.55   15.3
    Raspberry Pi, optimized fs options    17.2    13.3     4.35    0.96
    Raspberry Pi (turbo overclock)        17.2    14.4     4.83    0.86

    Raspberri Pi 2                        17.4    14.5     5.09    1.47

Obviously the high random write performance when used on a PC does not translate to the Raspberry Pi for this card. I am not quite sure what causes it but I think there are cheaper cards that perform much better on the Pi.

I actually have a recent Samsung MicroSD card very similar to the one mentioned above as a good performer, but it currently resides in my mobile phone. I will have to do some experiments with it.

[BenPope]

It would make sense to use a "standard" sd test method run on the Raspberry Pi itself. I see a wide variety of benchmarks on here and not many look to have been generated on the Raspberry Pi. Is there a "standard" tool that we could use to create a database of real Pi statistics?

I am sure there are better tools, but a while ago I wrote a Linux-based flash benchmark tool called "flash-bench" for the purpose of measuring (Micro-)SD card and USB stick performance. It is available at https://github.com/hglm/flash-bench.

Using your benchmark, the results for my Samsung 64GB Evo (MB-MP64D) (my previous iozone3 benchmarks above were also recorded on the Pi2)

Code: Select all

                      Seq.    Seq.    Random  Random
                      Read    Write   Read    Write
    Raspberry Pi 2    17.7    13.9    5.49    3.65