There are a couple things to consider.
An obvious one, would be if the AGP digital interface on
a motherboard, was non-compliant, or wasn't capable
of working at high transfer rates.
There is no evidence Intel chips ever suffered from this.
I have one motherboard, with an ALI AGP chipset, where
that seemed to be the problem. That's how I became a
fixture here, when I spent three weeks trying to get it
to work, and nobody could help me. That was one of the
few motherboards, which was a dismal failure for me.
And it seemed to be a chip design issue on the motherboard
(ALI Northbridge). The chip simply wasn't fast enough on
the I/O.
*******
A second kind of failure, is related to TYPEDET. That's
a simple logic signal, that sets the voltage on the
AGP bus interface. TYPEDET is supposed to be grounded
for 1.5V cards. A motherboard slot circuit, has to switch
between 1.5V and 3.3V when commanded to do so. And TYPEDET
is the signal controlling that power selection.
There were a few card designs, where some idiot designer
used a 50 ohm resistor to ground, instead of a good solid
ground. (Someone discovered this on a Matrox card, and
some factory engineer probably told the design engineer
to do it for "design for test" reasons.) The motherboard
designer on the other hand, assumed when doing their transistor
level switch, to work with a solid ground as an indication.
On boards which used 50 ohms to ground on TYPEDET, the
motherboard AGP slot voltage might be around 2V or a bit more.
This isn't what the chips on either end were expecting, and
could lead to problems. If, say, the switching threshold
doesn't have an automatic threshold adjustment.
(Slightly more details from my answer back in 2007...)
http://www.techiehq.net/pc-hardware/re-building-shuttle-sn95g5-63235.html
To check that 1.5V powered video card is properly designed,
you check for a zero ohm reading between pin 2A and 5A.
Rather than say, 50 ohms.
There are two supply voltages, but three signaling levels.
http://www.playtool.com/pages/agpcompat/agp.html
On that page, you can see the cards keyed for 1.5V or 3.3V.
But the signaling levels are 0.8V, 1.5V, and 3.3V. A way
to make a smaller amplitude signal, is to do something like
a double termination. And that can allow an interface with
a 1.5V power supply, to end up with 0.8V amplitude signals on
it. As far as I know, no "true" AGP 8X cards were made, using
0.8V pad drivers and 0.8V signals. The cards continued to be
powered by 1.5V power source. I don't know any of the details,
of how they made the chips work on that interface (switchable
termination or whatever).
As far as I know, my 845 board didn't cause me problems.
It might have had an old ATI 8500 or the like in it. I don't
know if I ever tested my 9800Pro in there. I have tested
my 9800Pro in my 440BX motherboard (for fun) - it should
have worked, based on slot keying, but the screen would
not light up and I had to remove it. The card continued
to work just fine after that, no damage.
The only card back then, where any concerns were raised,
was ATI 9700, and I don't remember the details of what
was said at the time. (It could be, there was no way to
qualify the card when it was made, so every motherboard
it was plugged into, would be "a new test case".) That's
the only card I might be tempted to do more Googling for.
Everything else should have worked.
And I don't think the practice of using a 50 ohm resistor
for TYPEDET is that common. It might have been seen on
a couple Matrox cards. Early in the life of AGP,
there were a few failures related to gimpy 3.3V supplies
on the AGP slot, and that was a motherboard issue. After
about TNT2 or so, that stopped being a common issue observed
by users. I doubt your slot is "winking out" on an overload.
That just doesn't happen any more.
The only other really bad mistake, was the SIS305 cards keyed
for both voltages, when the card was fixed at 3.3V. Those
could damage a 1.5V AGP slot, and it is because of that card,
that "AGP Protect" circuits were added to 845 motherboards
like the one you're working with. My P4B 845 motherboard
revision 1.05, was the first revision of that motherboard,
to be protected against SIS305 cards or similar (improperly
keyed cards). A red LED on the motherboard, checks TYPEDET,
and if no ground is present, that indicates a damaging
SIS305 could have been plugged in. The protection circuit,
prevents the ATX power supply from being turned on. And
the red LED glows as a hint as to how you screwed up.
