Best performing 64GB cards for Kobo?

[lonku]

I'd like to get a dedicated 64GB card for my kobo aura HD (which would contain android and a partition containing pdfs and cbz files which average around 50MB-200MB, and also contain 1000s of images averaging 100kb-1MB)

I've found threads here mentioning that class 4 cards perform better than class 10 cards, but as far as I'm aware, class 4 64gb cards don't exist? or is it a matter of brand in this case?

[JSWolf]

delete post

[DNSB]

Originally Posted by lonku

I'd like to get a dedicated 64GB card for my kobo aura HD (which would contain android and a partition containing pdfs and cbz files which average around 50MB-200MB, and also contain 1000s of images averaging 100kb-1MB)

I've found threads here mentioning that class 4 cards perform better than class 10 cards, but as far as I'm aware, class 4 64gb cards don't exist? or is it a matter of brand in this case?

In my rather limited testing, I've found that the SanDisk and Samsung 64GB cards are consistently among the best for small block performance. The problem is that most high capacity cards are optimized for reading/writing large blocks (large files, multimegabyte images, videos, etc.) and their performance on small blocks is pathetic.

If you have a Windows computer, I'd suggest trying CrystalDiskMark to test any card you purchase. Take a close look at the 4K and 4K QD32 results.

The reason for the earlier recommendation was that some Class 10 cards would have incredible large block read/write results but drop to using kilobytes to measure their performance on small block read/write operations.

This originally came up when people were modifying their Nooks and wondered why some cards worked well and others make the Nook almost unusable. It turned out Android on the Nooks was doing a lot of small block read/write operations. When I tested some cards I had on hand, a SanDisk class 4 card gave me a 1.81MB/sec write speed while a Kingston class 10 card gave me a 0.019MB/sec write speed -- and, yes, that was not a typo, it was 19 kilobytes/sec write speed.

Regards,
David

[hawhill]

Does a Kobo actually work with 64GB cards (thus being SDXC, not SDHC)?

As for Class 4 being faster than Class 10 - I think that would be pure luck. SD cards differ so much, even within the same series. Performance will also depend on the interface, filesystem, driver code and much more. It only makes limited sense to check from a different unit.

[lonku]

Originally Posted by DNSB

If you have a Windows computer, I'd suggest trying CrystalDiskMark to test any card you purchase. Take a close look at the 4K and 4K QD32 results.

I'm on a computer running Debian at the moment, Is there a Linux alternative/command to do the same type of test?

[JSWolf]

I recommend going wth a 32GB SDHC card at class 4.

http://www.amazon.com/SanDisk-MicroS.../dp/B004G7D0EG

[DNSB]

Originally Posted by lonku

I'm on a computer running Debian at the moment, Is there a Linux alternative/command to do the same type of test?

I haven't used any but one friend used a program called vdbench(?) to do some benchmarking. Sorry I can't be any more help to you.

Regards,
David

[JSWolf]

Originally Posted by lonku

I'm on a computer running Debian at the moment, Is there a Linux alternative/command to do the same type of test?

Surely you must know someone who uses Windows where you could have access to run the benchmarks.

Of course, the smart thing to do is go buy a copy of Windows and install it in a dual-boot setup.

[DNSB]

Originally Posted by hawhill

Does a Kobo actually work with 64GB cards (thus being SDXC, not SDHC)?

As for Class 4 being faster than Class 10 - I think that would be pure luck. SD cards differ so much, even within the same series. Performance will also depend on the interface, filesystem, driver code and much more. It only makes limited sense to check from a different unit.

I was checking the cards on a single system and the results were pretty consistent over multiple cards. Whether I tested using the Ricoh interface SD card slot or a USB uSD card adapter, the cards that had good small block performance were consistently better than the cards that had poor small block performace.

It had to be admitted that the performance item I was most interested in was not one that most producers of high capacity SD storage cards optimized for, I could see why some of them had very poor performance. However for my uses, the small block performance was a fairly critical measurement and having a good small block read/write performance made a noticeable difference.

I do have some difficulty understanding your belief that a card that performed poorly for small block read/writes in my test setup was going to suddenly perform well in another system. The results in the performance database were pretty consistent. Slugs were slugs regardless of what system they were tested on and a Class 10 card may have blown a Class 4 card out the water for sequential and large block performance but in small block performance, the 4 Class 4 Sandisk cards I tested were one (expletive deleted) of a lot faster than any of the Class 10 cards I tested.

BTW, what got me started on this finding a thread in the XDA Developers conference http://forum.xda-developers.com/show....php?t=1005633. You can also check the GoogleDocs database of the results at https://docs.google.com/spreadsheet/...lRDJQTEE#gid=0

Regards,
David

[Lucas Malor]

Here are some other benchs:

http://www.tomshardware.com/charts/m...marks,169.html
http://us.hardware.info/reviews/4544...s-a-difference

You can bench your sd card directly on kobo, using dd with a telnet connection. An image backup is strongly suggested.

[diskette]

4K is the best test block size according to DNSB

seems the simplest way to test write speeds is

sudo dd bs=4K count=20000 if=/dev/zero of=/mounted/directory/of/SDcard conv=fdatasync
source: https://romanrm.net/dd-benchmark

i have now done 4K write speed tests of five different SD cards, two of which were microSD cards in a verbatim adapter. i kept the filesystem [ext3], partition table type [dos], computer testing it with, and USB SD card reader itself the same throughout the tests

a brand new verbatim class 10 card performed 4K write speeds 2x worse than one of the much older non-micro SD cards. though it was 50% faster than the original kobo microSD card

here's some up-to-date 4K benchmarks for a larger variety of brands names
http://www.jeffgeerling.com/blogs/je...i-microsd-card
http://www.tomshardware.co.uk/charts...marks,169.html
http://www.custompcreview.com/review...pdate/22978/2/

considering the benchmarks above i am guessing DNSB's kingston card was an anomaly ?

[Lucas Malor]

You can also use fio, a Flexible I/O Tester. Recommended by L. Torvalds in his magnificent benevolence

[diskette]

i actually did the tests again with iozone and that really threw me for a loop with unpredictable outputs that i didn't really understand, despite putting in reasonable options. the verbatim may be about 50% faster than the stock kobo SD card when tested with iozone, and that also tests read speeds, which are obviously more important for an... ereader !

i'm planning to just buy a sandisk ultra given the benchmarks i posted from elsewhere and stop trying to figure out this mind bending I/O testing. though fio does look alot simpler than iozone

[Lucas Malor]

Anyway, can I say that probably ebooks are loaded into memory once opened, so this kind of benchmarks are completely useless? :P

[cramoisi]

Originally Posted by hawhill

Does a Kobo actually work with 64GB cards (thus being SDXC, not SDHC)?

I put a 128 go Sandisk SDXC in my gloHD once, and nothing was wrong. I even did a partition in exFAT for backup purpose (since i always have my reader with me).

After that i tested some 16/32 go, always Sandisk since it’s the brand that comes with the reader. I’m going now with a 32go Sandisk, which is perfect for a mixed collection of epub and cbr.

My experience shows that when reading the variation in speed is inexistant. The only change I detected (and perhaps it was in my mind) is the duration of the device for processing the loading of a huge collection on it.