Far from being an expert on this subject (or any subject for that matter!), 
I have some informative data from one of our board vendors concerning 
dielectric constant of FR-4.  I assume this information is fairly consistent 
across vendors with the differences coming in construction. (I assume - and 
I know this could be trouble - consistency due to the MIL designation of the 
prepreg materials)

Some quick defintions:
Prepreg = not totally cured dielectric
Core = cured dielectric
DK = Dielectric Constant

PREPREGS (listed from thinnest to thickest):
Glass Style 106
Nominal resin content = 72%  DK = 4.2

Glass Style 1080
Nominal resin content = 65%  DK = 4.4

Glass Style 2116
Nominal resin content = 57%  DK = 4.6

Glass Style 7628
Nominal resin content = 44%  DK = 4.8

Cores are made of different combinations (for thickness) of prepregs that 
are then cured.

Boards are made from combinations of cores, prepregs, and copper foil.

Dependent on your layer count and board vendor mfg style (conventional vs. 
foil laminate) you can have varying DK.  I cannot echo enough the comments 
of Doug Brooks that the board vendor needs to be involved with the design 
team early in the process.

Unfortunately for Mitch, trying to use traditional FR-4 with 1.2 GHz 
circuits is gonna be tough.  Perhaps you'll have to look at advanced 
materials (like PTFE or Cyanate Ester) which have lower DK and better loss 
tangents for the high speed circuits.  It will however cost you more $$$$.

Hope this information helps some.

Ben
[log in to unmask]
 ----------
>From: Doug Brooks
>To: designercouncil
>Subject: FAB: Re: Controlled Impedance - Dielectric Constant- Comment
>Date: Tuesday, May 21, 1996 8:41AM
>
>Mitch Morey wrote ...... (in part)
>"Bob,
>
>I only wished it was this easy. If it was I wouldn't have a weekly 
arguement
>about controlled impedence.
>
>I only put #3 above because I can't seem to get a "true" value for the 
dielec-
>tric constant from our three main fab houses. No real problem because we 
allow
>them to modify our trace widths and stackup to give us back our impedence
>required, BUT with speeds in excess of 1.2 Ghz now, we are asking for much 
more
>precise information. Our circuits will just not work correctly with the 
"fidge"
>factor any longer."
>
>
>Exactly ...............
>
>Dielectric "constant" changes from board to board, from layer to layer, and
>even within a layer. And, as Mitch points out, it is frequency dependent, 
so
>it even changes with signal characteristic.
>
>If really tight controlled impedance is necessary, we believe the only
>answer is to have the fab house on board at the BEGINNING of the project
>(rather than after the design has been completed.) A really good fab house
>will know its processes well enough to help you define what you need to do
>to hit your targets.
>
>Another example of where lowest bid doesn't necessarily mean lowest cost!!
>
>Doug Brooks
>President
>UltraCAD Design, Inc.
>http://www.eskimo.com/~ultra/homepage.htm
>
>