RE: PCI 3.3V Supply Rail Tolerance (resend)

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

...  In fact, if
> this specification for a 3Com Server NIC is to be believed:
> http://www.3com.com/products/nics/3cr990svrsp.html , at least one PCI
> product intended for the PC market has been developed that runs off
> 3.3V, but requires a tighter spec than +/- 9%  (+/- 5%, in this example)
 
If those specs are accurate, then technically, it is not a PCI product; it
does not comply with PCI specs.  It just happens to use a PCI-like interface
that isn't quite PCI.

It is also possible that the +/-5% was put there by marketing without full
knowledge of actual product needs or capabilities.  Or that the doc-writers
were working on "cruise control".

> 2. How do people view the approach that 3Com is apparently using (i.e.
> claiming PCI 2.2 compliance, but specifying a tighter 3.3V tolerance)?
 
That would be lying.  It is not PCI compliant if it requires a tighter 3.3V
tolerance than what the PCI spec states it must accept.

> Can anyone confirm that they are actually doing this?
 
That is exactly what their sales docs say.  Whether or not their product
actually does this (i.e., requires +/-5%), I don't know.

> 3. Is anyone aware of other brand-name, high-volume PCI products that
> run off 3.3V and require a tighter tolerance than the PCI spec?
 
Again, such products would not comply with PCI specs and cannot claim to be
PCI compliant.

Many low voltage logic components readily tolerate very wide supply voltage
ranges, well in excess of +/-10%.  A designer of PCI cards shouldn't use
components that require tighter-than-PCI voltage tolerances.  But this might
be difficult with ASICs, where you have fewer choices.  Designers should
avoid ASIC vendors whose products don't meet ALL PCI specs, including this
one.

You never know what environment you will find your card used in, maybe it's
OK in a desktop with ATX supply, but dies in a server, industrial lab box,
laptop expansion unit, etc.  The purpose of the spec is to make it possible
for any card to actually work in any compatible system, and not become
intermittent or something because the card fails to meet all specs.

> 4. Is anyone aware of a movement to change this tolerance spec in the
> future?
 
I am not.

Regards,
Andy