Max seep of the SD slot?

Andy Green andy at
Fri Jul 25 12:35:42 CEST 2008

Hash: SHA1

Somebody in the thread at some point said:
| On Thu, Jul 24, 2008 at 08:49, Andy Green <andy at> wrote:
|> Well, a good rule of thumb for generic 1-bit transfer is divide by 10
|> for the max speed in Bytes/sec you could expect, here we have 4-bit so
|> the peak raw trasfer speed is ~10MBytes/sec *BUT* after Glamo pulls the
|> data from the card, we have to sit there dragging it into the CPU
|> memory, on top of card latencies and command setup the speed is way
|> slower, still a decent ~2MBytes/sec each way IIRC.
|> - -Andy
| Thanks.
| This is something I also wanted to know for the choice of a new card.
| So we may say that up to 10MBytes/s, max speed of the microSD matters.
| After it should not much. (?)
| (and looks like 10Mbytes/s is currently what offer rather high end

The MCI / MMC stack in Linux negotiates the clock speed with the card,
all the microSD cards I saw say they can handle 16MHz and that's what
they get.

We can't sustain the actual throughput that implies right now, although
soon we might be able to get a little better in the driver.

| One more question though : if the card is slow, throughput will be
| lower, but will the CPU / glamo graphic bus be able to benefit of more
| resources while the card transfer occur ?
| Or does it just had to time to wait and CPU / glamo bus are as much
| busy (or waiting) ?

Have to make clear what we mean by "slow", it means latency to NAND
inside card as I understand it.  So every time you ask for some data it
is slower to start sending it and chokes more often waiting on getting
next block from NAND. --->

| Same question if we lower the SD clock rate. Is it equivalent to
| having a slower card ?

If it's slow at the card it will block Glamo memory less (probably...
arbitrator in Glamo might mess that assumption up) and definitely block
the CPU bus less.

- -Andy
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora -


More information about the community mailing list