Re: PCI slots

[Andy Ingraham <ingraham@wrksys.ENET.dec.com>]

> More that 4 slots requires a PCI bridge, as the PCI bus only allows 10
> loads ...
> ...  There are motherboards out there that violate this spec...
> I wouldn't use them.

I see this sort of thing stated from time to time.  I don't believe
that it is true.  If there is a "10 load" limit in the PCI spec, I
missed it.  (Someone please point it out to me.)

What is true is that the PCI spec talks about "typical configurations"
of 10 effective loads, where each slot counts as two.

But the spec also says (section 4.1.2), "... it is possible to achieve
different or larger configurations depending on the actual equipment
practice, layout arrangement, loaded impedance of the motherboard, etc."

One should also mention the clock frequency, since a somewhat slower
clock provides more bus settling time.  A slower PCI clock is not
so bad; a number of Pentium systems out there run the PCI clock at
30 MHz, some as low as 25 MHz, simply to have the clocks synchronous
with the CPU.

See section 4.3.3, where the table of pull-up resistor values is based
on "16 loads (believed to be a worst case)."  And these are 16 actual
loads, not effective loads ... each slot counting as one.

So why does the PCI spec recommend 10 loads but not set it as a limit? 
Ten effective loads is rule-of-thumb and is relatively easy to
implement without pain, without rigorous simulation of electrical
characteristics using SPICE or similar.  PCI was intended to be a
mainstream bus anybody can build to, even without exhaustive SPICE
simulations (which Intel already did to prove the concept).  But if
you go beyond the rules-of-thumb, you're truly on your own and you
must prove that it works within the electrical and timing limits of
chapter 4.

So I wouldn't make a blanket statement that motherboards with more
than four slots without a bridge shouldn't be used.  They may have
designed them very carefully and they may work fine.  Then again,
maybe not.

Regards,
Andy Ingraham
ÞàÐ