Jump to content

Testers wanted: SD card performance

tkaiser

By tkaiser
in Beginners

 Share

Recommended Posts

tkaiser

tkaiser

Posted
Posted

Hi,

 

in addition to some research regarding a few quality SD cards I wanted to ask our users to contribute to some collective benchmarking.

 

In case you bought Samsung, SanDisk, Toshiba or Transcend SD cards within the last few months, have a A20 device and are able to test the card in question, please use this A20 Armbian 5.07 image (vanilla kernel, performance governor, no user creation), burn the image on the card, connect board to network, start it and do the following at least twice

iozone -e -I -a -s 100M -r 4k -r 16k -r 512k -r 1024k -r 16384k -i 0 -i 1 -i 2

Please be aware that some cards seem to need some time for auto calibration when being first used so please do the test at least two times and in case results improve repeat the test until results seem to be stable (please see the test linked above for my experiences with a few Samsung cards that needed 4 test runs and then some performance values were twice as good than before).

 

When you're finished please execute 'armbianmonitor -u' to collect system information and post the support URL armbianmonitor provides, your test results and description of your board and the tested card in this thread.

 

I'm especially interested in results of cards with different capacity so in case you've for example two SanDisk Extreme Plus (one with 8 and another with 32 GB) then results from both cards are interesting since performance might improve with higher capacity cards.

tkaiser

tkaiser

Posted
  • Author
Posted

BTW: Results for no-name cards or from manufacturers that only buy NAND dies and controllers from somewhere and put them together (Kingston, PNY, Sony and so on) aren't interesting since they take what they get at the spot markets and both quality and performance of cards that look identical might differ a lot.

 

So collecting/publishing results for no-name cards or any other brand than the 4 listed above makes no sense (same applies also to buying these brands -- but that's another story ;) )

tkaiser

tkaiser

Posted
  • Author
Posted

You can use an existing image (vanilla, performance, cpufreq 960MHz). I created the test image only to ensure that inexperienced users are able to test with identical conditions (especially to ensure that no background jobs negatively influence performance).

Igor

Igor

Posted
Posted

I prepared those :), test board will be the same - Banana M1. I am getting few Samsung 32Mb EVO tomorrow and I have some more duplicates, some in use and no-names ...
 

IMG_60601.JPG

 

Test follows during evening. Most of cards are used.

Igor

Igor

Posted
Posted

Transcend Premium 300x 16GB (almost new) http://sprunge.us/UcSD

  Reveal hidden contents

 

 

Sandisk Extreme Pro 8Gb (almost new) http://sprunge.us/aDJJ

 

  Reveal hidden contents

 

 

Hardkernel eMMC 8G via SD reader (brand new) http://sprunge.us/SHXF

 

  Reveal hidden contents

 

 

Sandisk Ultra 8Gb (old and very used) http://sprunge.us/OWZf

 

  Reveal hidden contents

 

 

Sandisk 8GB (almost new) http://sprunge.us/iViT

 

  Reveal hidden contents

 

 

Transcend 8Gb (used) http://sprunge.us/NTRT

 

  Reveal hidden contents

 

 

Sandisk 8G (new) http://sprunge.us/XHjj

 

  Reveal hidden contents

 

 

Samsung EVO 32Gb (brand new) http://sprunge.us/bTQh

 

  Reveal hidden contents

 

1
tkaiser

tkaiser

Posted
  • Author
Posted

Interesting. All the 'normal' SanDisk show the same ultra slow 16K random writes. And regarding this (random I/O writes) still the rather cheap Samsung EVOs seem to outperform the others. But the eMMC's random I/O performance is impressive (but it's still quite expensive, isn't it?)

1
Igor

Igor

Posted
Posted

I just got EVO 32G and I'll add one <4 USD value no name which I bought in a bulk and most of cards are still operational. This eMMC absolutely rocks, its 18 USD @hardkernel store, but shipping is another 16 USD. 

 

Two more tests and then I am out of SD media ;)

zador.blood.stained

zador.blood.stained

Posted
Posted

All 4 cards are not new, and since this took quite some time, I collected armbianmonitor link only from 1st run: http://sprunge.us/AQDT

Filesystem is ext4 with default settings.

 

Toshiba, class 4, 16GB

 

  Reveal hidden contents

 

 

Transcend, class 10 UHS-I, 16GB

  Reveal hidden contents

 

 

Transcend, class 10, 16GB

  Reveal hidden contents

 

 

Toshiba, class 10, 8 GB

  Reveal hidden contents

 

1
lampra

lampra

Posted
Posted

I tried the image on three different cards, written by dd on ubuntu and Win32diskimager on win10.

Boot on cubietruck always hangs at:

Waiting for root device /dev/mmcblk0p1...
sunxi-mmc 1c12000.mmc: fatal err update clk timeout
sunxi-mmc 1c12000.mmc: fatal err update clk timeout
sunxi-mmc 1c12000.mmc: fatal err update clk timeout
tkaiser

tkaiser

Posted
  • Author
Posted

Thx for the feedback, might work if you can mount the card on another linux host and then replace the contents of /boot/dts/bananapipro.dtb with the cubietruck's.

rufik

rufik

Posted
Posted (edited)

Samsung EVO 8GB, brand new (product code is MB-MP08DA/EU). Tested on Orange PI 2 with Armbian 5.05, result is third pass.
HW info from armbianmonitor uploaded to http://sprunge.us/ZSAe

 

  Reveal hidden contents

 

 

Random write is slow, so I've made another 3 passes:
 

 

  Reveal hidden contents

 

 

More information about sd card:

 

 

  Reveal hidden contents

 

Edited by wildcat_paris
added spoiler tags
tkaiser

tkaiser

Posted
  • Author
Posted
  On 4/18/2016 at 10:05 AM, rufik said:

Samsung EVO 8GB

 

Thanks, can you please provide the output from 'sudo armbianmonitor -u' as well (that shows the card's metadata) and in case this card was brand new, try to repeat the test to see whether 4K writes get faster (at least that was the case with all new EVOs I tested with)

rufik

rufik

Posted
Posted
  On 4/18/2016 at 10:09 AM, tkaiser said:

Thanks, can you please provide the output from 'sudo armbianmonitor -u' as well (that shows the card's metadata) and in case this card was brand new, try to repeat the test to see whether 4K writes get faster (at least that was the case with all new EVOs I tested with)

 

I've edited my post and it includes hwinfo now. I'll try to make another 3 passes of iozone and let you know about result...

tkaiser

tkaiser

Posted
  • Author
Posted
  On 4/18/2016 at 10:36 AM, rufik said:

I'll try to make another 3 passes of iozone and let you know about result...

 

Great! Unfortunately you use an older Armbian image, so please use the following script to extract SD card information and add it to the post.

#!/bin/bash

get_flash_information() {
	find /sys -name oemid | while read Device ; do
		DeviceNode="${Device%/*}"
		DeviceName="${DeviceNode##*/}"
		echo -e "${DeviceName} info:\n"
		find "${DeviceNode}" -maxdepth 1 -type f | while read ; do
			NodeName="${REPLY##*/}"
			echo -e "$(printf "%20s" ${NodeName}): $(cat "${DeviceNode}/${NodeName}" | tr '\n' " ")"
		done
	done
} # get_flash_information

get_flash_information
rufik

rufik

Posted
Posted
  On 4/18/2016 at 10:45 AM, tkaiser said:

Great! Unfortunately you use an older Armbian image, so please use the following script to extract SD card information and add it to the post.

 

The result is:

mmc0:59b4 info:

                 cid: 1b534d3030303030105587061f00f283
                 csd: 400e00325b5900003bcd7f800a400041
                 scr: 0235800300000000
                date: 02/2015
                name: 00000
                type: SD
preferred_erase_size: 4194304
               fwrev: 0x0
               hwrev: 0x1
               oemid: 0x534d
              manfid: 0x00001b
              serial: 0x5587061f
              uevent: DRIVER=mmcblk MMC_TYPE=SD MMC_NAME=00000 MODALIAS=mmc:block
          erase_size: 512

Btw, there is no v5.06 for OPI 2 yet, right?

tkaiser

tkaiser

Posted
  • Author
Posted
  On 4/18/2016 at 10:55 AM, rufik said:

 Btw, there is no v5.06 for OPI 2 yet, right?

 

Nope, but there will be a new release for all boards soon: http://forum.armbian.com/index.php/topic/881-prepare-v51-v201604/

 

BTW: If the EVO shows better values after more tests it would be good if you update the results. I'm still trying to figure out the capacity 'borders' for different storage media (smaller capacity is often associated with lower speeds but this is only true up to a specific capacity. For example the 32GB and 64GB EVO perform identical while those with smaller size show lower write speeds)

rufik

rufik

Posted
Posted
  On 4/18/2016 at 11:24 AM, tkaiser said:

BTW: If the EVO shows better values after more tests it would be good if you update the results. I'm still trying to figure out the capacity 'borders' for different storage media (smaller capacity is often associated with lower speeds but this is only true up to a specific capacity. For example the 32GB and 64GB EVO perform identical while those with smaller size show lower write speeds)

 

Updated. Still the same.

I've ordered Samsung EVO 16GB, so I will check it in next few days. You will have quite complete info through different capacities (8, 16, 32, 64GB). 

lampra

lampra

Posted
Posted
  On 4/17/2016 at 4:51 PM, tkaiser said:

Thx for the feedback, might work if you can mount the card on another linux host and then replace the contents of /boot/dts/bananapipro.dtb with the cubietruck's.

Well, i did the change and cubietruck booted but iozone hangs at the first random read.

I will wait for the next release and i will report back here.

zador.blood.stained

zador.blood.stained

Posted
Posted
  On 4/19/2016 at 10:18 AM, lampra said:

Well, i did the change and cubietruck booted but iozone hangs at the first random read.

I will wait for the next release and i will report back here.

If disk activity LED (green on cubietruck) is lit or it flashes, it didn't hang, it's testing. It may take more than an hour (or even 2 hours) to complete first set of tests.

tkaiser

tkaiser

Posted
  • Author
Posted
  On 4/19/2016 at 10:18 AM, lampra said:

Well, i did the change and cubietruck booted but iozone hangs at the first random read.

 

Well, iozone gives no feedback while doing a single test. I do always in another terminal an 'iostat 5' in parallel since this gives some feedback what really happens (requires an 'apt-get install sysstat' before). And as Zador pointed out. There are cards with very low random I/O performance where this test can take ages (but looking at iostat output you can estimate the duration already)

lampra

lampra

Posted
Posted
  On 4/19/2016 at 10:34 AM, zador.blood.stained said:

If disk activity LED (green on cubietruck) is lit or it flashes, it didn't hang, it's testing. It may take more than an hour (or even 2 hours) to complete first set of tests.

 

  On 4/19/2016 at 11:16 AM, tkaiser said:

Well, iozone gives no feedback while doing a single test. I do always in another terminal an 'iostat 5' in parallel since this gives some feedback what really happens (requires an 'apt-get install sysstat' before). And as Zador pointed out. There are cards with very low random I/O performance where this test can take ages (but looking at iostat output you can estimate the duration already)

Thank you for your patience and for the info.

 

Here are the results for Samsung EVO 16GB, brand new (product code is MB-MP16DA/EU).

Tested on Cubietruck  with Armbian_5.07_Bananapipro_Debian_jessie_4.4.6 from the first post (adapted for cubietruck).

Results from pass No 5, 6 & 7.

 

  Reveal hidden contents

 

 

HW info from armbianmonitor below (due to sprunge overquota)

 

  Reveal hidden contents

 

 

I will post results for a couple more cards later this week

1
Igor

Igor

Posted
Posted

Another interesting device - Opi+ internal eMMC with 3.4.111-sun8i
 

  Reveal hidden contents

 

tkaiser

tkaiser

Posted
  • Author
Posted
  On 4/20/2016 at 10:09 AM, Igor said:

Another interesting device - Opi+ internal eMMC with 3.4.111-sun8i

 

Indeed. Sequential speeds not that impressive but random I/O especially with 16K record size pretty fast :)

tkaiser

tkaiser

Posted
  • Author
Posted

Another eMMC result:

                                                            random  random
              KB  reclen   write rewrite    read    reread    read   write
          102400       4    3836    3989    13389    13370   10816    3727
          102400      16   14491   14452    28656    27941   25246   13844
          102400     512   30696   30482    50109    50102   49613   30575
          102400    1024   30844   30666    51191    51202   50959   30466
          102400   16384   32673   33962    55222    55213   55158   33856

This is the 8GB eMMC of the soon available RK3288 (Cortex-A17) based MiQi Development Board (made by Benn Huang, one of Cubietech's founders). Both single threaded CPU and GbE network performance are impressive but prices irritating low (especially for the smaller variant with 1GB RAM and 8GB eMMC)

rufik

rufik

Posted
Posted

One more brand new Samsung EVO 16GB (MB-MP16DA/EU) result on OPI 2 + Armbian 5.05

                                               random    random     bkwd    record    stride
              kB  reclen    write  rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread  freread
          102400       4     2388     2172     6462     6658     6543     1195
          102400      16     5634     6580    14451    14747    14232     3749
          102400     512    11254    10524    21463    21467    21317     8986
          102400    1024    11650    11604    22049    22044    21947     9091
          102400   16384    11874    11383    22860    22877    22848    10566
                                 		              random    random     bkwd    record    stride
	      kB  reclen    write  rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread  freread
          102400       4     2236     2346     6664     6483     6547     1220          
	  102400      16     5664     6571    14427    14712    14413     3677          
	  102400     512    11227    10935    21470    21463    21318     9073          
	  102400    1024    11596    11621    22055    22068    21961     8974          
	  102400   16384    11893    11916    22886    22885    22830    10923
Card info:

mmc0:59b4 info:
                 cid: 1b534d303030303010f06d55a000f887
                 csd: 400e00325b590000775d7f800a4000a3
                 scr: 0235800300000000
                date: 08/2015
                name: 00000
                type: SD
preferred_erase_size: 4194304
               fwrev: 0x0
               hwrev: 0x1
               oemid: 0x534d
              manfid: 0x00001b
              serial: 0xf06d55a0
              uevent: DRIVER=mmcblk MMC_TYPE=SD MMC_NAME=00000 MODALIAS=mmc:block
          erase_size: 512
Eng-Shien Wu

Eng-Shien Wu

Posted
Posted

Orange Pi Plus2 (3.4.110-sun8i)

 

Internal 16GB eMMC (ext4 default)

 

  Reveal hidden contents

 

 

Samsung EVO+ 32GB

 

  Reveal hidden contents

 

 
The 16GB eMMC on the OPI Plus2 appears to be a step up from the 8GB eMMC on the OPI Plus and the 32GB EVO+.
rufik

rufik

Posted
Posted

1 year old Sandisk 32GB SD Card (model SE32G, so it's Extreme Plus AFAIK) used in DSLR camera. Test on OPI 2 + Armbian 5.05, 2nd and 3rd pass:

                                                              random    random     bkwd    record    stride
              kB  reclen    write  rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread  freread
          102400       4     3168     3093     8178     8355     7134     2833
          102400      16     6661     7139    15621    15832    15794     7021
          102400     512    17313    18136    21632    21643    21655    20083
          102400    1024    20643    20714    22152    22176    22160    20491
          102400   16384    21284    21342    22944    22944    22942    21334

                                                              random    random     bkwd    record    stride
              kB  reclen    write  rewrite    read    reread    read     write     read   rewrite      read   fwrite frewrite    fread  freread
          102400       4     3169     3158     8242     8322     7017     2842
          102400      16     6664     7133    15748    16077    16029     7055
          102400     512    15479    20580    21659    21652    21658    20145
          102400    1024    20600    20669    22155    22163    22163    20521
          102400   16384    21165    21300    22933    22934    22931    21267

Card info:

mmc0:aaaa info:

                 cid: 035344534533324780242e28a200e27b
                 csd: 400e00325b590000edc87f800a4040c3
                 scr: 0235800300000000
                date: 02/2014
                name: SE32G
                type: SD
preferred_erase_size: 4194304
               fwrev: 0x0
               hwrev: 0x8
               oemid: 0x5344
              manfid: 0x000003
              serial: 0x242e28a2
              uevent: DRIVER=mmcblk MMC_TYPE=SD MMC_NAME=SE32G MODALIAS=mmc:block
          erase_size: 512
tkaiser

tkaiser

Posted
  • Author
Posted

Since Igor added support for ODROID C1+ to Armbian I tested the C1+ with the well known Samsung Pro (64 GB):

 

 

  Reveal hidden contents

 

 

Results do not differ with 1.7GHz instead and funnily the Samsung Pro card was slow as hell when I started the tests (the above is the 6th test run -- especially 4K writes were below 1000 in the beginning so it seems the card's controller always needs some time when the card is used again).

 

It seems on ODROID C1+ sequential transfer speeds are limited to 16.5 MB/s so choosing eMMC with this board is an even better idea compared with other boards that show higher sequential speeds. Also Hardkernel's eMMC modules are known to be blazing fast especially when it's about random I/O.

lampra

lampra

Posted
Posted

Here is a used Transcend 8GB class 10 card

HW info from armbianmonitor uploaded to http://sprunge.us/hDGd

Photo

 

  Reveal hidden contents

 

 

 

  Reveal hidden contents

 

Guest
This topic is now closed to further replies.
 Share
Go to topic listing
×
×
  • Create New...

Important Information

Terms of Use - Privacy Policy - Guidelines

  I accept