Conference mvblab::alphaserver_4100

This system is running UNIX 4.0a and has all available internal slots utilized
with a requirement for more options.

Thanks
Roy

    PCI expanders are not supported.  There has been some talk about
    bringing a Bit3 system in for evaluation, but I wouldn't count on
    support any time soon.
    
    There will be a system with more PCI slots coming soon (you didn't hear
    it from me though.  Not allowed to talk about unannounced products in
    here.)
    
    Todd

    Just to complete the loop for me, is it true that there are no PCI
    expanders supported for Windows NT on the 4100?

    RE .3
    
    True, no PCI expanders are supported on any operating system.
    
    Todd

reference

>    RE .3
>    
>    True, no PCI expanders are supported on any operating system.
>    
>    Todd

Can someone comment on the salesupdate article which says that a PCI expansion
cabinet, with expansion upto 16 PCI slots total, is orderable now and 
deliverable in April 1997?

UNIX Version support?
NT support?

Any gotchas wrt which cards can be plugged where?

Thanks

Bill

    Re .5, have you called the product manager whose name appeared on the
    article?
    
    --svb

    
    
Bill,

  The previous replies in this string are referring to I/O expansion
cabinets that create additional PCI slots from a single PCI slot by
using PCI-PCI bridge chips.  A typical implementation has a PCI card in
the primary PCI with a bridge chip on it.  This card has a cable set
that goes to a separate box where the signals are received by another
PCI bridge chip on a PCI card that plugs into a four slot PCI
backplane.  This creates a third level PCI bus has three available
slots for PCI cards and a PCI bridge chip to a fourth level PCI bus
that has four slots.  This is a real implementation, I'm not making it
up.  We have not qualified this type of arrangement for the
Alphaserver 4100.  We have tested adapters behind multiple PCI bridges
in our lab and the performance of the adapter under test is
compromised as well as the performance of other adapters on the
primary PCI.

  The sales update article is announcing the I/O upgrade to the
Alphaserver 4000.  Details of the implementation were not provided so
that it is a bit difficult to determine just what is involved. The 16
PCI slot implementation of the 4000 involves a new drawer, motherboard,
I/O adapter and backplane.  Viewed from the back the obvious
difference from the 4100 (Dodge) drawer is that there are an additional
8 PCI slots on the right-hand side.  

  The space consumed by the right-hand I/O  backplane and PCI slots is
given up in the CPU/Memory area at the center of the box.  The new
motherboard for this reduced space has a maximum of two CPUs, two
memory options and two I/O interfaces.  Each I/O interface has two
MCBUS to PCI bus bridge chip sets on it.  This yields four separate
I/O channels each with four PCI slots for the total of 16.  

  The only slot restriction that I am aware of is the same as for the 
8 slot variation, graphics adapters have to be in PCI 0.

Dennis H.

dennis (and others)

thanks for the quick reply, it's saved me from some embarrasment!

It's a shame that the physical limitations of the original 4100 layout mean
that 16 PCI slots will only be available on the 4000, ie with max 2 procs.

Already discussed elsewhere are the justifications for building large disk 
rawhide configurations. IMHO, with the 16 PCI slots there is now the danger that
such configurations may turn out to be processor bound with only 2procs.

Can anything be done to rectify this?

Thanks again 

Bill

    And memory-bound to two memory slots (pairs) for a max of 4GBs of
    memory if 16 top-level PCI bus slots.