Conference ricks::dechips

You also left out the alternative of multiple PCI Host bridge (Each
giving a complete PCI bus heirarchy).

    The 4000-series already get 500MB/s (measured), on its two 64bit 33MHz
    (?) PCI buses. What else do they need ?

    Re .0, the Miata has two 64-bit PCI slots (albeit marred by issues that
    are being addressed at the moment by engineering).  I suspect future
    Alpha Windows NT systems will all have 64-bit slots.  We are actively
    working with 3rd-party vendors to introduce 64-bit PCI adapters on
    Alpha Windows NT.  I suspect outside of the PCI SIG plug-fest, there
    are not that many Intel NT systems being shipped yet.
    
    Re .1, i have not heard from our partners in the PCI Expansion business
    about their plans for 64-bit PCI Expansion box.
    
    --svb

Have you checked in NESBIT::MODULAR?  The PCI passive backplanes have 64
bit PCI as the primary and the DMCC Alpha modules do 64 bit pCI.  The
PBs support 3.3v  and I believe there are plans for more 64 bit support in
the future.

ac

At present, all our 21164 based workstations have 5V signalling 64-bit
33 MHz PCI slots, with both 5V and 3.3V power. These slots can do
over 200 Mbytes/sec of sustained bandwidth. These slots can also accept
ordinary 32-bit PCI devices.

Our next generation of workstations will also have this type of slot, so this
is a safe design target for at least the next couple years.

--RS

.-1   True multiple PCI Host bridge would be good but only if available in a
      system with cost falling in there price band (AS500, AS1000a).

.-2   The 4000 series is outside there priceband for the moment.
      Most of there system usage could be described as pseudo embedded
      eg : - use as Raster Image Pocessor with a hw module that drives a
             laser exposer or a digital print engine
           - use as a host platform for accelerator boards in there             
             printstreamer

.-3   They are not really interested in Intel NT for the moment (they need all
      the speed they can get for there RIP's)

.-4   We are looking at these

.-5   ok. They see two potential changes from there current design center 
      (AMCC S5933) both gated by new components becomming available from AMCC in
       CY98. 
          a) 64 bit support by AMCC in which they are interested for performance
          b) 3.3V only design support by AMCC in which they are interested for
             power disipation reasons and because they need 3.3v on there       
             modules anyway (I guess this would require 3.3V signalling and is  
             not in our future yet?)
      

For the moment they are looking where to go to in the near to medium future and
not solving an immediat design problem. They wanted to now where we digital
would go to in the PCI space so that they could start looking for replacements
of the current interphase chip is needed. 

For there current designs and those that are in progress they use 
the AMCC S5933 matchmaker. The QB release seems to be OK but they had
lots of problems with earlier releases.

None of there designs in progress (that i know of) needs more that 100Mbytes/sec
aggregated for there components of course you have to add disk & network load
which may push to total over 100. 

Some of them do need at least 4 equal (latency, arbitration, throughput) pci
slots for accelarator boards. We are currently looking if 1000a would be usable
putting the boards after the bridge but have concerns on the impact of the
bridges impact.




 

    re .0 - "I have seen the existance of a PCI notes file on TECRUS but it
    is members only, would it contain info pertinent for thses kind of
    issues?" ...
    
    Sam Duncan ( poboxb::duncan ) is listed as the moderator of the
    restricted RICKS::PCI conference.
    
    -- Kenny House

    Re: .6
    
    There are some quick-turn AlphaServer 4000-platform, significantly
    cost-reduced variants in the works (which may be attractive for your
    customer's particular application). Contact Jeff Robie or Vaughn Mackie
    for more details. 

The PCI notesfile is not very active lately.  You won't find the
answer to your question already there, but asking it might start a
new discussion.

AGP is another possibility, for high speed graphics....

You can do 5V signalling with 3.3V components. Some of our motherboards
and graphics modules do this. The 3.3V components must be 5V tolerant,
since other components on the bus may drive the signals above 3.3V.

--RS

> You can do 5V signalling with 3.3V components. Some of our motherboards
> and graphics modules do this. The 3.3V components must be 5V tolerant,
> since other components on the bus may drive the signals above 3.3V.

  Well, you can do it until you find the occasional signal that
  requires more than 3.3v to reach Vih. (For example, the PCIRST
  input on the Intel SIO chip requires 3.7v Vih.)

                                   Atlant

>>  Well, you can do it until you find the occasional signal that
>>  requires more than 3.3v to reach Vih. (For example, the PCIRST
>>  input on the Intel SIO chip requires 3.7v Vih.)

A PCI adapter is not permitted to drive PCIRST. The Miata workstation
drives its PCI bus (including PCI RST) from a 3.3V part. It uses the
Intel SIO ISA bridge. We have seen no signal-level problems with
interoperability on the SIO or ANY PCI adapters. 

--RS

Reinhard:

> A PCI adapter is not permitted to drive PCIRST. The Miata workstation
> drives its PCI bus (including PCI RST) from a 3.3V part. It uses the
> Intel SIO ISA bridge. We have seen no signal-level problems with
> interoperability on the SIO or ANY PCI adapters. 

  Maybe you'd better look again. Intel changed the spec a while
  ago (on the 82378ZB part) from requiring 2.0v Vih to 3.7v Vih.

  But if you'd rather not check, well, feel free.

                                   Atlant

>  Well, you can do it until you find the occasional signal that
>  requires more than 3.3v to reach Vih. (For example, the PCIRST
>  input on the Intel SIO chip requires 3.7v Vih.)

I've run into just this sort of thing already.  And it was a Digital
branded product!  In that case, it was the PCI CLK that was non-
compliant.

The simple fact of the matter, is that THESE COMPONENTS ARE NOT PCI
DEVICES!  They do not comply with the PCI electrical specs, so it is
incorrect to call them PCI.

> A PCI adapter is not permitted to drive PCIRST. 

That is correct.  PCIRST is a system-level signal driven by the
motherboard.

>>  Maybe you'd better look again. Intel changed the spec a while
>>  ago (on the 82378ZB part) from requiring 2.0v Vih to 3.7v Vih.

I'll follow up on this with the component engineer. The purchase spec
says 2.0V, but of course the spec doesn't mean much if the parts don't
meet it. 

In any case, this isn't relevant to the original question about PCI futures
since the adapter isn't permitted to drive PCIRST anyway.

--RS

Reinhard:

> I'll follow up on this with the component engineer. The purchase spec
> says 2.0V, but of course the spec doesn't mean much if the parts don't
> meet it.

  I know what the purchase spec says and it's wrong -- Intel
  changed their spec back in September of '95. We have this
  information direct from the Intel engineers and our engineer
  (Nelson MILKWY::Chan) can confirm this. And I'm not going to
  quible about what drives PCIRST; the LCA and LCA45 both do.

                                   Atlant