According to your figures the ATP Class 6 (in the internal mmcblk0 slot) is about 10% faster than the OCZ in the external/mmcblk1 slot, and both cards are about the same speed as my Class 2 8GB Transcend (~19.5s).

The only thing I can deduce from these figures is that the read performance is not affected in any way by the "Class" of the card and a Class 2 card will transfer (read) data as quickly as a Class 4 or Class 6 card, assuming all other variables are the same (high-speed and wide bus both enabled). Also, based on your figures a "fast" SD card is unlikely to be faster than a standard SDHC Class 2 card (your OCZ is slower than my Transcend).

As for testing "write" performance - this is tricky as the OS will buffer writes giving unrealistic performance figures. Hopefully someone else can suggest a methodology to test write performance.

Until we can get firm figures on write performance, if anyone is looking to buy extra storage my advice would be to buy Class 2 SDHC cards, avoid SanDisk (no high-speed support) and don't spend too much as you're unlikely to see any real benefit for the extra outlay.

I've had a good experience with Transcend - reasonably cheap 4GB and 8GB Class 2 capacities (in the UK £25 and £50 respectively), and they have high-speed and wide bus support.

I've just run this speed test on a couple of cards I've bought for my n800:

The Integral card was listed as ORA on the website (expansys.com) but it's branded Integral on the card itself.

Integral 4gb sdhc run #1 - ext slot:
real 0m 18.14s
user 0m 0.08s
system 0m 3.21s

Integral 4gb sdhc run #2 - ext slot
real 0m 19.63s
user 0m 0.13s
sys 0m 3.25s

Inov8 1gb "3-in-1" sd run #1 - int slot
real 0m 26.49s
user 0m 0.14s
sys 0m 2.94s

Inov8 1gb "3-in-1" sd run #2 - int slot
real 0m 26.20s
user 0m 0.09s
sys 0m 3.35s

From what I understand about the new class ratings: yes, they are referring to the minimum write speeds. All previous SD cards, where they advertise 120X or 150 X were usually referring to reading speeds. I read somewhere that the SD Association mandated that the new format have a minimum write speed - hence the different classes.

I've found a few reviews online which suggest that Sandisk SDHC cards are horrible in terms of price / write performance.

http://www.extremetech.com/article2/...2094734,00.asp

http://techgage.com/article/sd_card_roundup/

None of these reviews that I've listed use an exact scientific approach to benchmarking, but it is interesting reading. The price I paid for the ATP was almost enough for an 8gb Transcend. I would have bought that, but I was unsure about the quality and the read/write speeds. I would guess that the Kingston Class 2 would be a good all rounder and the Transcend a good value buy.

EDIT: Forgot to mention that anybody planning on running the tests should DISABLE virtual memory - I've found that it'll add a couple seconds to results. Test results seem to fluctuate a lot too.

Yes, me too. It is worse when booting from the card. Any write to card causes audio to skip. It is caused by raised priority of mmcqd kernel daemon. I'm not sure if it is intentional or bug. With Nokia kernel mmcqd has normal priority (=0) with sdhc patches it is raised to -5. It can be fixed by running as root after each boot. I notified Philip Langdale about it.

Thanks for searching!
Seems to work.
:-)

To test write speed I use something like that:

$ sync; time sh -c "dd if=/dev/zero of=testw1 bs=1M count=100;sync"

dd will output a useless MB/s, but time "real" output is what you're looking for, 100/"real time" will give you MByte/second.

Hope this helps,

Laurent