RE: 32 bit / 3.3V PCI slots?

["Ingraham, Andrew" <Andrew.Ingraham@compaq.com>]

> Considering that 5V-tolerant devices are increasingly difficult to make,
> it seems to me the question for a board manufacturer should instead be
> whether to design a board to be Universal or 3.3V-only. 
 
Indeed.  However, it may be easier on the whole to make the chips on PCI
adapter cards (at least on some PCI cards) 5V-tolerant, than it is to make
the motherboard's host bridge 5V-tolerant.  The latter is more closely
aligned with the CPU process and the CPU's buses and speeds.


> It is a chicken and egg question, but building Universal boards may not be
> possible for much more longer and it is imperative that PCs with 3.3V
> slots become available soon, or for the 5V PCs to dissapear. 
 
Yup.  That's why the PCI spec added wording in 1998 to (hopefully) eliminate
5V-only PCI cards.  Now motherboard vendors need to make the next step and
introduce 3.3V slots in mainstream products.  This should have happened
years ago.


> Just for curiosity, I checked the voltage on these signals on several PCs
> using 5V connectors and I found 5V as expected, but also some 3.3V and
> even a couple of machines with 4V pull-ups!
 
I can understand the 3.3V pull-ups.  I could be mistaken but I don't think
the spec says you must pull up to 5V on a 5V bus; after all, 2.4V is an
acceptable output signal level.  A motherboard designer could decide to tie
his pull-ups to 3.3V, remember it's his responsibility to make it all work.

But the pull-ups to (apparently) 4V bothers me!

I can't imagine that they added a 4V supply just for this purpose.  If they
have pull-up resistors going to 5V, but a PCI chip on the motherboard that
has clamps to 3.3V (GAH!!), then the clamps would conduct and limit the
pull-up to about 4V.  This means the chip isn't 5V tolerant, and it has no
business being used on a 5V PCI bus.

If that's what happened, someone take the PC designer out back and shoot
him.

Alternatively, the pull-ups might be something like internal FETs or ? which
peter out around 4V, and that is probably OK.

Either way, the devices, whether 5V- or 3.3V-powered, should tolerate full
5V levels plus overshoots to 7V or so.


> Considering the presence of pull-ups on all signals that are not driven by
> a 3.3V device, is 5V-tolerance a requirement for today's PCI devices?
 
Not sure exactly what you are asking ... but as long as you design a device
for use on a 5V PCI bus, it is a requirement for that device to be 5V
tolerant, meaning (in part) that it must tolerate +5.25V indefinitely, and
overshoots that might go to +7V or 11V or so (depending on the presence or
absence clamps to the 5V rail).

If you think, signals will never go that high because MY chip only drives to
3.3V, think again.  Your device shares a bus with other 5V PCI devices, and
this can include ones that drive full 5V rail-to-rail signals.  Nothing
prevents someone from plugging in the first PCI adapter card ever made,
which probably drives full 5V levels.  Nothing says new adapter cards can't
do this too (though it may be hard for them to find an ASIC house that still
has a slow, 5V-capable process).

Regards,
Andy

application/ms-tnef