zardoz66
Posts: 77
Joined: Wed Jan 02, 2013 1:02 am
Location: Austin TX
Contact: Website

Re: SD Card Benchmarks

Mon Jan 21, 2013 9:01 pm

Image
-----------------------------------------------------------------------
CrystalDiskMark 3.0.2 (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 : 20.062 MB/s
Sequential Write : 14.945 MB/s
Random Read 512KB : 18.971 MB/s
Random Write 512KB : 1.466 MB/s
Random Write 4KB (QD=1) : 0.194 MB/s [ 47.3 IOPS]
Random Read 4KB (QD=32) : 2.359 MB/s [ 575.8 IOPS]
Random Write 4KB (QD=32) : 0.186 MB/s [ 45.5 IOPS]

Test : 50 MB [F: 0.0% (0.0/7608.0 MB)] (x5)
Date : 2013/01/21 14:27:31
OS : Windows Server 2008 R2 Server Standard Edition (full installation) SP1 [6.1 Build 7601] (x64)
SDHC Card : Patriot LX 8GB Class 10
Zardoz
Raspberry Pi 3 | Raspbian Jessie | 32GB Samsung EVO+ | D-Link 7 port USB HUB | Sabrent CRW-EXT 6-Slot Card Reader | 64GB USB Storage. Other RPi's - 2x RPi1b, 1x Rpi1b+ 1x Rpi2b, 2x Rpi3b
Join us on IRC #raspberrypi at https://webchat.freenode.net/

User avatar
toysareforboys
Posts: 136
Joined: Thu Dec 06, 2012 11:01 pm

Re: SD Card Benchmarks

Mon Jan 21, 2013 9:09 pm

zardoz66 wrote:SDHC Card : Patriot LX 8GB Class 10
I've said it once, and I'll say it again:

Patriot LX cards = slowest on earth.

-Jamie M.
Seagate GoFlex Home, 1.2GHz ARM (kirkwood), 128MB RAM, Gigabit Ethernet, SATA2. Sandisk Extreme 120GB SSD running Arch ARM Linux 3.6.11-0. nginx + php-fpm = LIVE STATUS hosted right on the SGFH!! http://tafb.yi.org

iwl
Posts: 26
Joined: Mon Dec 17, 2012 5:35 pm

Re: SD Card Benchmarks

Mon Jan 21, 2013 11:43 pm

It's just no difference also making sence with 512k Blocks which I think is larger than the write Block Size of an SD Card

iwl
Posts: 26
Joined: Mon Dec 17, 2012 5:35 pm

Re: SD Card Benchmarks

Tue Jan 22, 2013 12:07 am

The Sandisk 45 MB/s 32 GB slows further down on random with larger Files

Code: Select all

                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
  256000     512   19472   20949    21812    21867   21645    1996
The Samsung Es.. 32GB remains constant:

Code: Select all

                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
  256000     512   11256   11276    21919    21946   21703   11040

User avatar
toysareforboys
Posts: 136
Joined: Thu Dec 06, 2012 11:01 pm

Re: SD Card Benchmarks

Tue Jan 22, 2013 12:11 am

iwl wrote:The Sandisk 45 MB/s 32 GB slows further down on random with larger Files
I was lookin more for the 4k numbers with the bigger file ;)

-Jamie M.
Seagate GoFlex Home, 1.2GHz ARM (kirkwood), 128MB RAM, Gigabit Ethernet, SATA2. Sandisk Extreme 120GB SSD running Arch ARM Linux 3.6.11-0. nginx + php-fpm = LIVE STATUS hosted right on the SGFH!! http://tafb.yi.org

cito
Posts: 16
Joined: Mon Nov 07, 2011 3:27 am
Contact: Website

Re: SD Card Benchmarks

Wed Jan 23, 2013 4:13 am

Testing SD Card - Sony Class 10 16 Gig
OS: Raspbian

I bought a new SD Card super fast for my DSLR Camera since I do a lot of photography and wanted the fastest or supposedly one of the fastest cards I could get for also recording 1080p video

So I bought this card for my Camera: http://www.amazon.com/gp/product/B0077S ... 03_s00_i00

So I took my old SD Card the Sony Class 10 16 Gig and ran bonnie++ on it

Screenshot: http://image.dude-suit.net/albums/userp ... rdtest.PNG

Image


just apt-get install bonnie++

or check this out: http://claw.triple6.org/linux/hard-driv ... ark-linux/

nice little benchmarking tool

User avatar
MartyMacGyver
Posts: 32
Joined: Sat Jan 26, 2013 1:37 am

Re: SD Card Benchmarks

Sat Jan 26, 2013 7:26 am

SanDisk Ultra 16 GB SDHC Class 10 (SDSDU-016G-U46) benchmarks

My Pi: Model B 512M with Occidentalis v0.2 (http://learn.adafruit.com/adafruit-rasp ... s-v0-dot-2). I've installed all the updates.

I ordered what was supposed to be the SanDisk Ultra 16 GB SDHC Class 6 (SDSDH-016G-U46) from Amazon. What I got was in a retail package but it was immediately clear it was the class 10 version of this card. On the back was the usual inventory sticker over the UPC code (with this SDSDH number on it), but lifting it confirmed it is a SDSDU-016G-U46. I suspect it's just that the class 6 cards are becoming obsolete now (since most people are buying them for large-sequential-write applications like cameras).

I probably won't keep it, but I got some details for those who may consider this class 10 card.

First, the device info:

Code: Select all

$ head /sys/block/mmcblk0/device/{cid,csd,scr,*rev,manfid,oemid,name,serial}
==> /sys/block/mmcblk0/device/cid <==    0353445355313647800490881500cc00
==> /sys/block/mmcblk0/device/csd <==    400e00325b59000076b27f800a404000
==> /sys/block/mmcblk0/device/scr <==    0235800100000000
==> /sys/block/mmcblk0/device/fwrev <==  0x0
==> /sys/block/mmcblk0/device/hwrev <==  0x8
==> /sys/block/mmcblk0/device/manfid <== 0x000003
==> /sys/block/mmcblk0/device/oemid <==  0x5344
==> /sys/block/mmcblk0/device/name <==   SU16G
==> /sys/block/mmcblk0/device/serial <== 0x04908815
Next, the most common benchmark info (iozone 3.414 which took *forever* to compile on the Pi - then again I've rarely seen a 600K C source file...)

Code: Select all

Command line used: ./iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2
                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
   51200       4    1442    1559     5180     5184    4940     593 (* was as low as 373)
   51200     512    9983    8138    21751    21737   21733    1598
   51200   16384    6260   10029    22343    22341   22347   10699
I actually ran that test three times (this one was the closest to average). The read speeds didn't fluctuate much but the write speeds - particularly for the two larger-file-size tests - fluctuated quite a bit. Some argue this is because the test size is too small... but re-running with -s 1024M takes a very very *VERY* long time and I'm not convinced there is a great benefit to it (in terms of comparing results). I believe the fluctuations we see here are down to other transactional overhead from other things running in the OS (either causing threads to wait or interrupting the writes), and it may also have to do with the card's architecture.

That said, the -s 1024M results were lower across the board (in some cases, particularly 4K random write, depressingly lower).

Code: Select all

Command line used: ./iozone -e -I -a -s 1024M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2
                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
 1048576       4    1432    1491     5006     5007    4198     257  (* 332 on a subsequent run)
 1048576     512    7355    6386    21815    21815   21669    1077
 1048576   16384    6529    6727    22285    22284   22277    6658
In short, this card is, like many class 10s, definitely weak on the smaller writes (though it's not as bad as some out there). However, I've yet to see a solid recommendation for a currently-available card. It'd be nice to see a data source more useful than max throughput in the tables I've found so far for the Pi.

I also wonder how accurate this test is given the CPU is single-threaded, and how much variation that introduces run-to-run? (It'd be nice to be able to have iozone automatically run a few passes and average them.)

Also, in the real world caching is usually present (even Android sees great improvements when SD card caching is enabled).

http://itsacleanmachine.blogspot.com/20 ... ry-pi.html

I think I'm going to give the SDSDX-016G-X46 a try next (are there any better suggestions in the 20 MB/s actual high-end throughput range that the Pi seems to top out at anyway?)... it appears to have similar low-end performance but better high-end performance, and the Pi appears to be able to make good use of that headroom. And better write-caching would probably remove most of the differences currently seen... and would be far more effective than tilting at the mythical "perfect" SD card windmill (because short of an actual SSD, you either must lop off large-r/w sequential throughput or suffer poor small-r/w performance when dealing with flash memory...)

tl;dr - The Pi seems to top out around 20MB/s in terms of best possible uncached throguhput (from what I've read so far), and caching makes a BIG difference in real-world performance on any such system (also saving wear-and-tear!) so if your card can hit that speed on large-block reads and writes and is decent on the uncached low-end throughput (0.5-1MB/s appears to be "decent" in this regime) you're probably going to get good bang for your buck - unless you do a *LOT* of tiny random writes. The question is, which happens more in real-world usage, and how well is Raspbian optimized for SD cards? Could be a lot of room there to close the speed gap...

User avatar
toysareforboys
Posts: 136
Joined: Thu Dec 06, 2012 11:01 pm

Re: SD Card Benchmarks

Sat Jan 26, 2013 9:21 pm

MartyMacGyver wrote:SanDisk Ultra 16 GB SDHC Class 10 (SDSDU-016G-U46) benchmarksThat said, the -s 1024M results were lower across the board (in some cases, particularly 4K random write, depressingly lower).
4k random
read: 4198
write: 257
Thanks for running the "big" test! The 4k random read speeds are awesome, but the 4k write is pretty bad. It's not super horrible as most of the cards I tested are around 150 but for sure not as good as say the Patriot EP at 707 4k write (but only 3218 4k read).

Thanks again for the great test!

Be interesting to see, if they swap it out with a class 6 card, how much better (or worse?!? lol) the 4k write numbers will be!

-Jamie M.
Seagate GoFlex Home, 1.2GHz ARM (kirkwood), 128MB RAM, Gigabit Ethernet, SATA2. Sandisk Extreme 120GB SSD running Arch ARM Linux 3.6.11-0. nginx + php-fpm = LIVE STATUS hosted right on the SGFH!! http://tafb.yi.org

User avatar
MartyMacGyver
Posts: 32
Joined: Sat Jan 26, 2013 1:37 am

Re: SD Card Benchmarks

Sat Jan 26, 2013 10:55 pm

toysareforboys wrote:Thanks for running the "big" test! The 4k random read speeds are awesome, but the 4k write is pretty bad. It's not super horrible as most of the cards I tested are around 150 but for sure not as good as say the Patriot EP at 707 4k write (but only 3218 4k read).
Thanks again for the great test!
Be interesting to see, if they swap it out with a class 6 card, how much better (or worse?!? lol) the 4k write numbers will be!
-Jamie M.
I'm not planning to downgrade to a class 6... they're difficult to find as it is and it's not worth having poor overall speed - not to mention I've yet to see a solid winner. We'll see how the x46 works out: one's already on the way.

User avatar
MartyMacGyver
Posts: 32
Joined: Sat Jan 26, 2013 1:37 am

Re: SD Card Benchmarks

Wed Jan 30, 2013 9:42 am

Followup: Rpi benchmarks for the Sandisk Extreme 16 GB 45 MB/s Class 10 SDSDX-016G-X46

Code: Select all

head /sys/block/mmcblk0/device/{cid,csd,scr,*rev,manfid,oemid,name,serial}
==> /sys/block/mmcblk0/device/cid    <== 03534453553136478004b6004400cc00
==> /sys/block/mmcblk0/device/csd    <== 400e00325b59000076b27f800a404000
==> /sys/block/mmcblk0/device/scr    <== 0235800300000000
==> /sys/block/mmcblk0/device/fwrev  <== 0x0
==> /sys/block/mmcblk0/device/hwrev  <== 0x8
==> /sys/block/mmcblk0/device/manfid <== 0x000003
==> /sys/block/mmcblk0/device/oemid  <== 0x5344
==> /sys/block/mmcblk0/device/name   <== SU16G
==> /sys/block/mmcblk0/device/serial <== ---------------

Code: Select all

Command line used: ./iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2
Output is in Kbytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 Kbytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
   51200       4    1666    1825     4720     4897    4781     751
   51200     512   20729   20910    21911    21907   21822    4458
   51200   16384   14807   21030    22359    22326   22346   21019

Code: Select all

Command line used: ./iozone -e -I -a -s 1024M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2
Output is in Kbytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 Kbytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
                                                    random  random
      KB  reclen   write rewrite    read    reread    read   write
 1048576       4    1771    1881     4906     4908    4551     430
 1048576     512   20788   20813    21859    21858   21644    1528
 1048576   16384   21023   21091    22331    22333   22322   14983
The SDSDX-016G-X46 I just got performs better than the SDSDU-016G-U46 in almost all speed categories, though random write and 4K I/O in general is still pretty poor compared to 512K and 16M readings. Again, I'd think caching should smooth out such bumps.

iwl
Posts: 26
Joined: Mon Dec 17, 2012 5:35 pm

Re: SD Card Benchmarks

Fri Feb 01, 2013 4:46 pm

I have to mention that an other user has problems with the card I recommended
Samsung MB-SSBGA Class 10 Essential SDHC 32GB
He says his pi freezes after some hours with this card and it's the same with the two cards of he bought.
He got the messages i/o error writing journal shortly before freeze.
I never had any problems and it still is the fastest card for my use.
I have 4 other brands of class 10 cards and never such problems, performance differences only.

leexgx
Posts: 10
Joined: Fri Feb 15, 2013 3:50 pm

Re: SD Card Benchmarks

Fri Feb 15, 2013 4:07 pm

Hello i got my PI 2-3 days ago so this is my first post on here been playing with XMBC at the moment but was looking at getting faster Class SDHC card so found this topic

really i was not expecting the results i got below for this "TeamGroup Class 10 32GB" (think my SD card reader mite be maxed out at 17-18MB/s but i lack any thing els that is faster then it to test it)

i was about to order an Kingston 32GB Class 10 but after the reading on this forum its very shocking results from Kingston and the test i just did on my class 4 Kingston SDHC card at the bottom
-----------------------------------------------------------------------
CrystalDiskMark 3.0.1 x64 (C) 2007-2010 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 17.651 MB/s
Sequential Write : 13.668 MB/s
Random Read 512KB : 17.556 MB/s
Random Write 512KB : 7.812 MB/s
Random Read 4KB (QD=1) : 3.928 MB/s [ 958.9 IOPS]
Random Write 4KB (QD=1) : 0.839 MB/s [ 204.7 IOPS]
Random Read 4KB (QD=32) : 4.821 MB/s [ 1177.1 IOPS]
Random Write 4KB (QD=32) : 1.156 MB/s [ 282.2 IOPS]

Test : 50 MB [H: 0.0% (0.0/29.9 GB)] (x3) <All 0x00, 0Fill>
Date : 2013/02/15 15:53:36
OS : Windows 7 Ultimate Edition SP1 [6.1 Build 7601] (x64)
TeamGroup Class 10 32GB in USB Adapter, stated speed R:20MB/s-W:16MB/s
32GB Class 10 TeamGroup SDHC in USB adapter.png
32GB Class 10 TeamGroup SDHC in USB adapter.png (45.44 KiB) Viewed 10775 times
is it me or are Kingston really skimping on random access (look at the IOPS!!) could it be an fake ? last time i seen IOPS that low was when SSDs first came out from One maker that was utter garbage for any type of writes

this 4GB SD card is very useless (mite take it back as its way under performing)
-----------------------------------------------------------------------
CrystalDiskMark 3.0.1 x64 (C) 2007-2010 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 15.842 MB/s
Sequential Write : 3.999 MB/s
Random Read 512KB : 15.763 MB/s
Random Write 512KB : 0.629 MB/s
Random Read 4KB (QD=1) : 3.005 MB/s [ 733.6 IOPS]
Random Write 4KB (QD=1) : 0.007 MB/s [ 1.7 IOPS]
Random Read 4KB (QD=32) : 3.136 MB/s [ 765.7 IOPS]
Random Write 4KB (QD=32) : 0.018 MB/s [ 4.4 IOPS]

Test : 50 MB [H: 8.7% (6.0/68.9 MB)] (x2) <All 0x00, 0Fill>
Date : 2013/02/15 15:53:20
OS : Windows 7 Ultimate Edition SP1 [6.1 Build 7601] (x64)
Kingston SDHC 4GB class 4 in USB Adapter, Product speed not stated (from the Write speed no wonder they do not show it)
4GB Class 4 Kingston SDHC in USB adapter.png
4GB Class 4 Kingston SDHC in USB adapter.png (45.44 KiB) Viewed 10775 times

leexgx
Posts: 10
Joined: Fri Feb 15, 2013 3:50 pm

Re: SD Card Benchmarks

Fri Feb 15, 2013 4:48 pm

also (as i cant post right away yet) ok can now :)

i was getting 1300-700ms of response time on the kingston 4GB class 4
where as the teamGroup 32GB class 10 was around 50ms (depending what test was running the QD32 really pushes it high response time [was was 120ms] but that's to be expected but unlikely under norm use)

monitored this just using windows resource monitor (windows 7 and vista have it, not perfectly but it gives you aprox access time speeds, had an look at AS SDD as well but test takes long time ended up just doing the ACC time only in the end got 1.059 ms and 8.160 ms (dread to know what my Kingston kicks up will post when i get back as it likely take an hour to run the test)

Fraoch
Posts: 135
Joined: Thu Mar 07, 2013 11:53 pm
Location: Cambridge, Ontario, Canada

Re: SD Card Benchmarks

Tue Mar 19, 2013 3:56 pm

I bought two cards for my still-to-be-delivered Pi using this list as a reference:

http://elinux.org/RPi_SD_cards#SD_card_performance

Unfortunately this was before I saw this thread and others indicating class 2, class 4 and class 6 cards may have better 4K random QD32 writes...dang.

Anyway I bought a SanDisk Ultra 16 GB Class 10 (part SDSU-016G-U46S) and a SanDisk Extreme 16GB Class 10 UHS-1 (part SDSX-016G-X46). Yes, I know testing them in the Pi itself is the best way, but in advance I've been testing them in Windows and Ubuntu.

Ultra:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.2 (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 : 16.004 MB/s
Sequential Write : 14.181 MB/s
Random Read 512KB : 15.801 MB/s
Random Write 512KB : 2.943 MB/s
Random Read 4KB (QD=1) : 4.594 MB/s [ 1121.7 IOPS]
Random Write 4KB (QD=1) : 0.646 MB/s [ 157.7 IOPS]
Random Read 4KB (QD=32) : 4.305 MB/s [ 1051.1 IOPS]
Random Write 4KB (QD=32) : 0.619 MB/s [ 151.0 IOPS]

Test : 50 MB [F: 0.0% (0.0/14.8 GB)] (x5)
Date : 2013/03/19 8:41:05
OS : Windows XP Professional SP3 [5.1 Build 2600] (x86)

Extreme:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.2 (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 : 15.967 MB/s
Sequential Write : 15.036 MB/s
Random Read 512KB : 15.823 MB/s
Random Write 512KB : 3.929 MB/s
Random Read 4KB (QD=1) : 5.153 MB/s [ 1258.0 IOPS]
Random Write 4KB (QD=1) : 0.614 MB/s [ 149.9 IOPS]
Random Read 4KB (QD=32) : 4.460 MB/s [ 1088.7 IOPS]
Random Write 4KB (QD=32) : 0.553 MB/s [ 135.1 IOPS]

Test : 50 MB [F: 0.0% (0.0/14.8 GB)] (x5)
Date : 2013/03/19 8:50:44
OS : Windows XP Professional SP3 [5.1 Build 2600] (x86)

Ultra, iozone testing on Ubuntu (using iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2)

Code: Select all

random  random    bkwd   record   stride                                   
              KB  reclen   write rewrite    read    reread    read   write
           51200       4    1512    1526     4510     4527    3756     820
           51200     512    9099    9235    18625    18675   18524    2244
           51200   16384    8117    8798    18990    18937   18956    9158
Extreme:

Code: Select all

random  random    bkwd   record   stride                                   
              KB  reclen   write rewrite    read    reread    read   write
           51200       4    1607    1616     3848     3839    3827     826
           51200     512   11160   11514    17739    17738   17727    3379
           51200   16384   11392   11276    18080    18077   18077    9988
The extra money spent on the Extreme seems wasted so far. The 4K random QD32 writes aren't bad but aren't as good as the results of some class 2/4/6 cards I've seen in this thread and others. Maddening as there were SanDisk class 4 and class 6 cards sitting beside these on the store shelf, priced lower...

I don't think you can go wrong with SanDisk though. It's obvious they use class-leading controllers in their SD cards. Their warranty is generous and the prices aren't bad considering what you get.

We'll see how it performs in my Pi - Newark bumped up my delivery date from April 15th to March 26th - next week! Yay!
Pencoed-made Model 1B, Samsung memory
2B 1.1
3B+
4B 2GB

User avatar
MartyMacGyver
Posts: 32
Joined: Sat Jan 26, 2013 1:37 am

Re: SD Card Benchmarks

Wed Mar 20, 2013 8:59 pm

Fraoch wrote:I bought two cards for my still-to-be-delivered Pi ...
Unfortunately this was before I saw this thread and others indicating class 2, class 4 and class 6 cards may have better 4K random QD32 writes...dang.

Anyway I bought a SanDisk Ultra 16 GB Class 10 (part SDSU-016G-U46S) and a SanDisk Extreme 16GB Class 10 UHS-1 (part SDSX-016G-X46). Yes, I know testing them in the Pi itself is the best way, but in advance I've been testing them in Windows and Ubuntu.
See my reviews earlier in this thread - I benchmarked both on the Pi itself and the "Extreme" SDSDX-016G-X46 was better in the benchmarks and feels better as well IMHO.

Yes, the 4K random performance is kinda poor in general on faster cards (though this particular card seems to do better than the rest), but it depends on what you want to do... if you rarely do large sequential writes or reads then sacrificing top-end speed on a lower-class card may be worth the low-end speed gain.

One benchmark I've never seen is a performance benchmark (versus the rather artificial memory benchmarks we've all seen and done)... what's the overall I/O usage pattern for real-world applications?

portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: SD Card Benchmarks

Thu Mar 21, 2013 12:02 am

The only real world test I've done is boot speed. I have two cards that are similar in all speed tests, except for 4k read/write. One is about 4200 read, the other 3200 read. The faster one is 72 write, the other is 38 write.(pretty slow cards)

The faster one boots in 28 seconds and the other boots in 34. So it looks like 4k read/write has a major real-world impact, at least on boot performance.

I have a USB 3.0 flash drive coming in the mail today, it should be good. I'll be using this instead of the SD card(aside from boot partition). Will post boot time and iozone results.

One last thing, you can overclock the SD card! Add

Code: Select all

init_emmc_clock=110000000
to /boot/config.txt
This overclocks it by 10%. 120000000 is 20% and so on. It WILL cause corruption if set too high though. So make a backup first.

portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: SD Card Benchmarks

Thu Mar 21, 2013 10:51 pm

So my new USB drive is much faster than the SD card I was using. Boot time went from 28 seconds to 18 seconds.

But in many tests on the raspberry pi it says it's slower than the SD card. On my computer with a USB 2.0 port it's much faster. Why would this be? A bug?

User avatar
toysareforboys
Posts: 136
Joined: Thu Dec 06, 2012 11:01 pm

Re: SD Card Benchmarks

Thu Mar 21, 2013 10:59 pm

portets wrote:But in many tests on the raspberry pi it says it's slower than the SD card. On my computer with a USB 2.0 port it's much faster. Why would this be? A bug?
The raspberry Pi's USB port is driven by the CPU and shares it's bandwidth with the network card and other USB devices. Running a "load test" on USB sticks is problematic due to everything it's shared with.

-Jamie M.
Seagate GoFlex Home, 1.2GHz ARM (kirkwood), 128MB RAM, Gigabit Ethernet, SATA2. Sandisk Extreme 120GB SSD running Arch ARM Linux 3.6.11-0. nginx + php-fpm = LIVE STATUS hosted right on the SGFH!! http://tafb.yi.org

Fraoch
Posts: 135
Joined: Thu Mar 07, 2013 11:53 pm
Location: Cambridge, Ontario, Canada

Re: SD Card Benchmarks

Fri Mar 22, 2013 12:11 am

MartyMacGyver wrote:See my reviews earlier in this thread - I benchmarked both on the Pi itself and the "Extreme" SDSDX-016G-X46 was better in the benchmarks and feels better as well IMHO.
Whoops, sorry I missed that MartyMacGyver! :oops:
Pencoed-made Model 1B, Samsung memory
2B 1.1
3B+
4B 2GB

portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: SD Card Benchmarks

Fri Mar 22, 2013 12:37 am

toysareforboys wrote:The raspberry Pi's USB port is driven by the CPU and shares it's bandwidth with the network card and other USB devices. Running a "load test" on USB sticks is problematic due to everything it's shared with.
But shouldn't that only effect the peak bandwidth? 16M sequential read speeds are 37MB/s on my computer and 32MB/s on the Pi, which makes sense for the Pi's limited bus. But 4K random reads are 5MB/s on my computer and 2MB/s on the Pi.

Oh, maybe this is a latency issue instead of a raw bandwidth issue? That would effect all speeds.

Morganleij
Posts: 2
Joined: Sun Apr 21, 2013 9:54 am

Re: SD Card Benchmarks

Sun Apr 21, 2013 10:14 am

Hi
I am new here.
I intend to order a Model B and SD card coming week. And a book...
I design cirquit boards and other machine parts, and am a Linux user since 2006.
Now i want to know the RasPi and learn perl, when spare time admit...

To save on spare time a quick SD card is one measure, and i have read the tests in this thread with interest.
There seem to be two main alternatives: a Sandisk SDHC Ultra class 6, because of MUCH higher performance on small writes than class 10 or SDXC cards. Even a class 4 is faster than the latter in that case.

The other method is to use the very quickest. Some wrote that it is no use to have a card with more than 45M B/s because of speed limit on USB. That may be true if the card really do 45 MB/s sustained, but i doubt it will and USB2 480 Mbit/s is 60 MByte/s including protocol overhead). So i guess 90 MB/s will make a *small* advantage. And possibly the 90 MB/s card is faster than the 45.

Maybe i will get both and compare, and get back - in a couple weeks

Then there is also the matter of reliability and endurance, and i guess the extreme / pro variants then is the choice?

There are some aspects regarding th etests i have not seen adressed:
If I understand correctly all tests are done at the filesystem level.
What filesystem? What modes? Are partition aligned? And use optimal block size?
* for every test it need to be optimised for the card tested *
- if results are to be compared.

Morganleij
Posts: 2
Joined: Sun Apr 21, 2013 9:54 am

Re: SD Card Benchmarks

Mon Apr 22, 2013 5:57 pm

also... write speed depends on if flash blocks are cleared and available, or the the flash need erase when writing. So a newly formatted card will be faster than one that have been in use a while, even if blocks are deleted the conventional way and looks unused to the user.
Googleing, i found SD cards probably do not support TRIM (?)
But here is an interesting read on a clearing tool, and evedence of some erasing command:
http://us.generation-nt.com/answer/impr ... 07771.html
And a flash drive tester/eraser: http://www.vconsole.com/client/?page=page&id=13

WanaGo
Posts: 21
Joined: Wed Oct 10, 2012 1:04 am

Re: SD Card Benchmarks

Sat Apr 27, 2013 10:32 am

Found these from a little while ago, just more of the same type of thing:

http://www.tomshardware.com/reviews/mic ... 11-12.html

http://www.tomshardware.com/reviews/sdx ... 40-10.html

DanielJacks0n
Posts: 1
Joined: Wed May 08, 2013 2:59 pm

Re: SD Card Benchmarks

Wed May 08, 2013 3:11 pm

Hi there,

I've tested two Sandisk Ultra Class 10 cards on RasPi running Raspbmc (both tested with same image). The test was done in the shell after exiting XBMC. Here are the results:

Sandisk Mobile Ultra MicroSDHC 16GB Class 10:

Code: Select all

==> /sys/block/mmcblk0/device/cid <== 035344535531364780054b3bc700d100
==> /sys/block/mmcblk0/device/csd <== 400e00325b59000076b27f800a404000
==> /sys/block/mmcblk0/device/scr <== 0235800100000000
==> /sys/block/mmcblk0/device/fwrev <== 0x0
==> /sys/block/mmcblk0/device/hwrev <== 0x8
==> /sys/block/mmcblk0/device/manfid <== 0x000003
==> /sys/block/mmcblk0/device/oemid <== 0x5344
==> /sys/block/mmcblk0/device/name <== SU16G

-----------------------------------------------------------------------

        Command line used: iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2

                                                            random  random
              KB  reclen   write rewrite    read    reread    read   write
           51200       4    1233    1261     5095     5074    4950    1324
           51200     512    9048    5775    21724    21839   21819    2950
           51200   16384   11340    8503    22321    22316   22328   11394

           51200       4    1270    1396     5171     5191    5066     835
           51200     512   11367    8679    21878    21883   21864    2950
           51200   16384   10301    8734    22336    22337   22305    3340

           51200       4    1180    1386     5098     5113    4992    1370
           51200     512   10293    9080    21823    21837   21771    3751
           51200   16384   12859   10305    22216    22217   22170   10115
Sandisk Ultra SDHC 32GB Class10:

Code: Select all

==> /sys/block/mmcblk0/device/cid <== 035344535533324780066002b500d400
==> /sys/block/mmcblk0/device/csd <== 400e00325b590000edc87f800a404000
==> /sys/block/mmcblk0/device/scr <== 0235800100000000
==> /sys/block/mmcblk0/device/fwrev <== 0x0
==> /sys/block/mmcblk0/device/hwrev <== 0x8
==> /sys/block/mmcblk0/device/manfid <== 0x000003
==> /sys/block/mmcblk0/device/oemid <== 0x5344
==> /sys/block/mmcblk0/device/name <== SU32G

-----------------------------------------------------------------------

 Command line used: iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2
    
                                                            random  random
              KB  reclen   write rewrite    read    reread    read   write
           51200       4    1361    1436     5350     5373    4851    1358
           51200     512    9050    8703    21869    21793   21864    5992
           51200   16384   11427    9568    22235    22287   22343    9877

           51200       4    1259    1419     5357     5385    4810    1120
           51200     512   11299    8894    21837    21793   21808    6652
           51200   16384   12410    9688    22331    22284   22339   10035

           51200       4    1247    1355     5310     5321    5173     874
           51200     512   10923    9530    21867    21884   21843    5945
           51200   16384   10638   10009    22315    22306   22316    9888

phmanzano
Posts: 1
Joined: Tue May 28, 2013 8:22 am

Re: SD Card Benchmarks

Tue May 28, 2013 8:24 am

-----------------------------------------------------------------------
CrystalDiskMark 3.0.2 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 : 23.029 MB/s
Sequential Write : 5.600 MB/s
Random Read 512KB : 22.539 MB/s
Random Write 512KB : 2.998 MB/s
Random Read 4KB (QD=1) : 5.498 MB/s [ 1342.3 IOPS]
Random Write 4KB (QD=1) : 1.977 MB/s [ 482.7 IOPS]
Random Read 4KB (QD=32) : 4.888 MB/s [ 1193.3 IOPS]
Random Write 4KB (QD=32) : 1.219 MB/s [ 297.5 IOPS]

Test : 50 MB [H: 0.0% (0.0/29.7 GB)] (x5)
Date : 2013/05/28 10:15:10
OS : Windows 8 [6.2 Build 9200] (x64)
SanDisk 32GB class 4 (SDSDB-032G-B35) 5yrs

Happy to see that this Class 4 is pretty fast! Good buy for Raspi.

Return to “General discussion”