>
> 1) If you go the Microdrive route, you will be serving up files that you
>    access over what is effectively an 8-bit, ISA, PIO-only IDE interface.

only REAL IDE make sense.

> 2) If you somehow get PCI IDE onto the box, it won't do you a whole lot of
>    good, because the machine's CPU is a truly stupid design -- a 133MHz
>    486 core connected internally to a 133MHz, 64-bit-wide SDRAM controller
>    by a 32-bit-wide, 33MHz pipe!  AMD's documentation is somewhat unclear
>    about this point but you can easily test the memory bandwidth for

STUPID. i though it drives SDRAM at 66Mhz.


>    yourself and see what they did: hooked up existing "486 core" and "SDRAM
>    controller" cells from their library without bothering to do any design
>    work on the path between them.  Sigh... that's right, the machine has no
>    more memory bandwidth than a 33MHz 486 would, and this turns out to be

than real 486/133 which had 33Mhz memory bus too :(

>    *the* limiting factor for its performance even in routing applications
>    where all you do is move data from a network controller, into memory, and
>    then back to another network controller without any copies.  For file
>    service, even if you use NFS (where at least the data isn't repeatedly
>    copied across the user/data boundary) instead of Samba (where it is) this
>    box is particularly ill-suited because of its cripplingly narrow memory
>    pipe.

25MB/s (486 memory speed) isn't too bad. but it won't saturate 3 100MBps
eths. but will 3 10Mbps.

> I really wish Soekris made a similar machine with a CPU that didn't lose

strongarm could be good for it :)