The FFT plots (there are several available in GR) don’t have a bandwidth
of their own, they just display the FFT magnitude of the input sample.
Frequencies are only for display and are usually calculated by the
sampling rate parameter you set. That has no effect on the plot itself,
but only on the axes’ labels.
If you want more bandwidth in your plot, you have got to have a higher
sampling rate. If this relationship is new to you, I may recommend the
recommended reading page on the GNU radio wiki, which, ironically, is
not reachable for everyone right now, but that should be sorted out in a
few hours (I hope).
So: The Bandwidth you see is the bandwidth you put in; there’s no way
around that. If you’re having something that smaples a complex signal at
100kHz sampling rate, you cannot expect your FFT to show higher
frequencies than that.

This is a completely different problem, is it? Bandwidth is orthogonal
to the concept of dynamic range and amplitude values.

Anyway, your FFT plot knows about complex samples; therefore, it
displays dB relative to the numeric value of “1.0”; this has nothing to
do with your input signal power. But since USRP/daughterboards tend to
work linearly over most of their available power range, you can just add
a multiply_const to your signal processing flowgraph and get your value.
The factor, however, depends on your ADC/daughterboard combination, on
the gain you set for the RF frontend, a little bit on temperature, the
tuner, the impedance matching of your RF input to the Daughterboard etc.
So if you already know what you’re putting in, just use this as a
reference and “calibrate” yourself.

Thanks Marcus, my problem is how to calibrate the scale on the FFT plot.
e.g:
I’m feeding a signal of -30 dBm to the usrp and on the plot the value
becomes about -10 dBm