[Tfug] USB3 *PCI* card

[Bexley Hall]

Hi Zack,

--- On Fri, 10/19/12, Zack Williams <zdwzdw at gmail.com> wrote:

> > http://www.newegg.com/Product/Product.aspx?Item=N82E16815201060&name=Add-On-Cards
> 
> Wow... per specs, that's a 32-bit PCI-X 2.0 (266Mhz) card.
> 
> I don't think I've ever seen a logic board that supports more than
> PCI-X 133Mhz, and even then that was a high end server board.

Yes, though actual performance depends on a lot more than
*just* the card.  I.e., the device it is talking to influences
overall throughput.  As does the host's access patterns for
that device and the number of devices actually "active" at
any given time.

Most devices have some sort of elastic store within.  So, the
trick is matching the buffer depth to the device's requirements
along with the host.

And, average/sustained throughput can always be lower than
instantaneous throughput to still yield beneficial results.
I.e., transferring a sector off a rotating medium at the
*interface's* rate (instead of the medium's) reduces the
latency for that data -- even if you're forced to wait for
the *next* sector!

> I'm betting that with that card you're unlikely to get more than about
> 120MB/s, unless you happen to have a 66Mhz slots, and even then any
> PCIe 1.0 1x board would do much better.

I'm (currently) looking to use some small SBC's (FlexATX-ish) for
this sort of thing -- not a full-fledged "PC" -- in the hope that
I can later repurpose this stuff for my multimedia system (e.g., to
serve media files).  If I opt to design a new board, later, then
I would reevaluate just how much functionality I actually *need*.

(e.g., serving up video is typically < 1MB/s/program -- actually
within the realm of USB1.1 -- for a single program!  OTOH, you
wouldn't want to *move* large amounts of video between media
at that rate!)
 
> Nearly all USB 3 host cards should be supported in linux,
> provided they're XHCI compliant.