Re: [SI-LIST] : AGP Buffer strengths

D. C. Sessions (dc.sessions@tempe.vlsi.com)
Thu, 20 Nov 1997 07:41:01 -0700

Rick Tan wrote:
>
> Has anyone looked at the strengths of AGP Buffers?

You could say that :-)

They're 50-ohm impedance minimum.

> Since AGP is a point to point bus, shouldn't the drivers be a lot weaker
> than they are currently?
>
> Looking at the specs, it seems that AGP buffers are 75% the strength of the
> weakest PCI 3.3 buffers, and 55% the strength of the strongest PCI 3.3
> buffers.
>
> Simulating a AGP bus 9.6 inches long (5.6 on the motherboard, and 4 on the
> add-in card) with the strong AGP drivers causes tremendous overshoot and
> undershoot (this is with the minimum clamping currents).

Last year we went up to Hillsboro for the final review before AGP 1.0
was released & I had a box full of simulation slides showing that
the (then proposed) PCI driver strengths were WAY excessive and would
require either cooling off or very strict clamping specs -- neither in
the spec as proposed. It turns out that Intel had also realized that
the driver spec was too strong and cooled it off (to 2/3 of PCI) and
that sort of short-circuited the discussion, although I would have
preferred to push for symmetrical drive (as is, N is stronger than
P, for no good reason) and about a 60-ohm line.

Either way, receiver clamping is critical for anything beyond
worst case.

-- 
D. C. Sessions
dc.sessions@tempe.vlsi.com