From the benchmarks posted on this forum, the slowest SDHC Class 2 card typically outperforms even the fastest SD card (when reading data) so unless you need SD for compatability I'd recommend SDHC every time as Nokia will surely add official SDHC in the next firmware (clowns if they don't).

Well.. I haven't got the 8GB SDHC card yet. The 4GB I have now has enough space for my needs for now. I'm waiting the 8GB prices to go down some more...

Well, that depends on card price and importance of write speed. I am happy with transcend 4gb 150x. It was cheaper even than Transcend 4GB SDHC class 2 and write speed is excellent and much better than the class 2 card (and even slightly better than class 6 card), see also test here.
http://www.hjreggel.net/cardspeed/sp...cards-sdh.html

Write speed can be measured either by copying some big video over USB which is practical test where it really matters or by doing synthetic test via dd i.e. something like
then just divide size (80 MB?) by the result in seconds

True that some SD cards may be of poor quality with pathetic write speed but fast sd cards are faster when writing then SDHC class 2 and may be cheaper since SDHC is still novelty and sell in lower volumes. Well at least this was true at the time I bought my card. When checking same shop today I see Transcend SDHC class 2 price dropped to be now lower than the 150x one (by approx $8, 4GB SD 150x ~= $58, 4GB class2 SDHC ~= $50) so this may not apply anymore.

When you are planning use cards for storing video and expect to write often to it, fast writes are important. In normal usage writing is not so frequent and read speed of all current cards is similar and good enough.

I have a Kingston 8GB CL6 card on order and plan to benchmark it against my Patriot 8GB CL4. Even if there's no significant difference in the N800, I expect the speed bump will help bulk transfers via a PC hosted SDHC reader.

On a related note, Patriot's 16GB CL6 availability has been pushed back to this summer. My query in their support forum eventually prompted the following response:

It might be safer to move it all to your linux desktop and then write it back to the chip. Once it is on the ext.xx or reiser FS it will be defragged.
My wife has been writing to SD cards so she can take music to work and play it on her Palm. She has had no trouble after almost 2 years. Flash cards will wear out in time, but we have not experienced any problems yet.
Also, I found a nice card reader which plugs into a USB port for less than 10 dollars US. It's made by "pqi genie" and will save your battery a bit as file x-fer is a power hungry process on a portable

Glen

Did the write test with Transcend 150x 4GB SD. When writing 160MB it took 25 seconds
This means N800 write speed including FAT filesystem overhead is approx. 6MB/s with this card.

Fanoush - in this post I wrote 128MB in 20.64s to a Transcend Class 2 8GB card (6.4MB/s - same as your result 160/25=>6.4). However I'm not convinced this test is reliable due to the OS buffering writes, though I could be wrong.

the sync command flushes write buffers, you can also mount before test, measure start, do the write, umount (which flushes device buffers too) and measure end

EDIT: the difference is that I am writing to file so sync/umount should definitely make sure blocks are written. But anyway I would guess direct device i/o does not cache writes, that would be quite dangerous as devices can be removed anytime, it does cache reads however. BTW there is some file in /proc or /sys to flush such read buffers, I can dig it in mail, used it to debug my 'broken' mmcpluc 4GB card that works in everything else but N800.

My Connect3D 4GB non-SDHC card reads something in the 8-10MB/sec range, compared to my Kingston 2GB MMCMobile which reads in the 9-12MB/sec range.

Worth noting, I think, is that these cards are on sale again at ECost for $20. I'd be happy to post specific benchmarks and such: I'll also point at Dealram as where I keep finding cheap memory.

Write ~160MB: 65 seconds, 2.5MB/sec
Read ~160MB: 17 seconds, 9.4MB/sec
Output in dmesg:
Somewhat related, today at work I plan to build a new kernel incorporating MPPE as well as the SDHC patches --- I'm also going to get a full suite of modules together for cifs, nfs, ext3, so forth, and so on. If anyone is interested, I'd be happy to post it up somewhere.

Also somewhat related, I might put together a Python script that will test your cards and upload benchmarks for you to a special site, if I'm bored enough at work

Ouch - that WRITE speed is pretty slow! At least it confirms what Fanoush is saying that direct IO writes aren't buffered in which case the write speed can be guesstimated with a resonable degree of certainty.

Has anyone tested the write speed of a Class 4 or Class 6 SDHC card running in high-speed (48Mhz) mode?

The benchmark app might be quite useful!