Kingston SDM4/4GB
4GB miniSD, Class 4

~ $ sync; time sh -c "dd if=/dev/zero of=/media/mmc1/testfile bs=1M count=100; sync"
100+0 records in
100+0 records out
real 0m 31.14s
user 0m 0.03s
sys 0m 2.75s
~ $ time dd if=/media/mmc1/testfile of=/dev/null bs=1M
100+0 records in
100+0 records out
real 0m 19.36s
user 0m 0.00s
sys 0m 2.59s

Also Kingston 4Gb miniSDHC, class 4 (£17.44 including delivery).

~ $ sync; time sh -c "dd if=/dev/zero of=/media/mmc1/testfile bs=1M count=100; sync"
100+0 records in
100+0 records out
real 0m 31.29s
user 0m 0.02s
sys 0m 2.23s
~ $ time dd if=/media/mmc1/testfile of=/dev/null bs=1M
100+0 records in
100+0 records out
real 0m 15.45s
user 0m 0.00s
sys 0m 1.80s

N.B. All times are best of 3 runs.

Dividing 100 MB by the number of seconds (real) yields:

write: 5.26 MB/s
read: 7.47 MB/s
write: 3.32 MB/s
read: 5.17 MB/s
write: 3.20 MB/s
read: 6.47 MB/s

That casts doubt into the validity of the test. I did not see more than a few percent variance on the 770. Sure nothing else was running?

I'll be testing here also in a few days

There is always something else running. That is why when you do timing tests like this you always have to do it several times, take the fastest and assume that the others were slower because some background process kicked in to delay them. Averaging is the wrong thing to do: only the maximum rate counts.

The write time I posted is effectively identical to the time reported by ashes: my other two runs were about 3 seconds slower. The read time is better, but the other two runs were very close to the time ashes gave, so I'm confident that the times for the Kingston card are reasonably accurate and stable. Ima's sandisk times do seem to fluctuate a lot though: perhaps there was more running in the background there than should have been.

Kingston 4go class 6:
sync; time sh -c "dd if=/dev/zero of=/media/mmc1/testfile bs=1M count=100; sync"
100+0 records in
100+0 records out
real 0m 41.66s
user 0m 0.00s
sys 0m 3.42s

and:
time dd if=/media/mmc1/testfile of=/dev/null bs=1M
100+0 records in
100+0 records out
real 0m 16.72s
user 0m 0.01s
sys 0m 1.80s

Please carefully note that these results are insignificant because they highly depend on linux's block cache, and subsequently on the memory used (but mostly available) on the n8x0 at the time of the test. Please also note that the block device layer reads ahead when possible, so clearly, these are not super meaningful results, unless if you flush the buffer cache prior to the test and run every test with the same amount of memory available for buffer cache. The buffer cache can easily be flushed as such:

echo 1 > /proc/sys/vm/drop_caches

I also suggest some random i/o patterns rather than sequential i/o, but most importantly what would be REALLY useful is to know how many writes such or such card or brand or type of card allow, it's flash technology, therefore it has a limited amount of write cycles available, which obviously is clearly not advertised, and these numbers vary quite a lot from constructor to constructor (and often from same cards from the same manufacturer, but different revision numbers). But performing such tests are indeed destructive, although that'd be worth it.

Benchmarks in sequential i/o are somewhat useless, because as long as the card is fast enough to play a movie (the contrary would be surprising) then it's fast enough for anything you would want to do that involves sequential i/o on a flash card (unless if you're intending to put a database on such card and do some table scans). Random i/o is probably closer to what people are using these cards for. If you're just storing movies on such card, then maybe, just maybe you bought the wrong kind of device, and you should have looked for some video ipod or anything (archos, creative zen...?) preferably with a harddrive. Because the point of storing 10 movies on a card is somewhat meaningless (that's when streaming media is usually more valuable than reading from the flash card); and needless to say, the cost of a 12GB/8GB card is almost as high as the cost of 10 dvds.

If you really want to use the N8x0 as a media player (read: PMP), I suggest that you focus on attaching an external harddrive to it, which could be a 2.5in laptop drive thru the usb port, or maybe bluetooth or wifi if you have some money to spend. Obviously, then the issue of power consumption becomes a real concern as well as space occupied by the "Portable" media player. Again, if playing movies is what you're looking for, then maybe the N8x0 is not the device you should have bought.

Now, storing some mp3 on such flash device makes more sense than movies because of the limited size of such device and the fact that it doesn't have an internal hard drive. Books present little interest unless you wish to kill your eyes on such a small screen. Emulated games make some sense, although I doubt there's a high interest in carrying 5000+ games around (and if so, you should have bought a gp2x). Carrying databases such as an astronomy database, or mathematic/physics reference formula or maps or simply a subset of wikipedia does make sense as well, but for such, access time to the media is usually irrelevant because whatever needs to be searched is indexed (in a b-tree I'd hope).

Now, let me ask the N810 community, what do you intend to use your 6GB+ flash card for? Now don't get me wrong, it's fine to have your favorite movie on your flash card (as long as it's "Wargames"), I'm just saying that it somewhat defeats the purpose of flash memory.

@jibanes: bit OT maybe. I would be using it for music, pictures, ebooks, some video, utils (yes even Windows utils), packages (for poss re-install), Squeak images and probably lots of other stuff.

Kingston 2GB microSD (Transflash) SDC/2GB-2ADP
sync; time sh -c "dd if=/dev/zero of=/media/mmc1/testfile bs=1M count=100; sync"
100+0 records in
100+0 records out
real 0m 34.10s
user 0m 0.00s
sys 0m 2.07s

time dd if=/media/mmc1/testfile of=/dev/null bs=1M100+0 records in100+0 records out
100+0 records in
100+0 records out
real 0m 11.08s
user 0m 0.00s
sys 0m 1.02s

best of tree attempts

For me, Emulators and game engines; Random access, 30MB to 3GB game data.

Thanks for the useful info on the cache. Maybe we can get more accurate testing programs ported from linux-land. The dd test at least does show rough correlation between class 4 and class 6 claimed performance and results.

Your point that movie/media streaming does not require a high speed card is a good one and I hope people take it into consideration.

(edit)----------------------------------
Here in Germany, I'm looking at alternate.de's offers for the:

Kingston 4GB mini SDHC Class 6, rated 6MB/s Write, 6MB/s read (the one that ppl have been testing above) for 22 euro.

or

Transcend 4GB mini SDHC 45x, rated 8MB/s Write, 7MB/s read (strange rating!) for 24 euro.

Seeing as the Transcend hasn't been tested here yet, I'm buying that one today so you guys have more data points for comparison. Tests to follow...

(edit)---------------------------------
running the modified speedtest 5x (see shellscript attachment) on internal flash gives me

Write: 47.2, 42.0, 50.5, 59.7, 50.2 sec. best: 2.38 avg: 2.00 MB/s

Read: 11.9, 11.8, 19.3, 21.3, 11.8, sec. best: 8.47 avg: 6.57 MB/s

N810 with 8 GB Sandisk SDHC in Kingston adapter.

Write: 37.12s or 2.70 MB/s
Read: 11.26s or 8.88 MB/s