Interesting factoid: I was looking to experiment with 100MHz
oscillators largely as a clock source for my own AD9951 experimentation
(using the AD9951's built-in PLL multiplier at 4x). I was hoping to
experiment a bit with 20 MHz crystals I already had in hand before
ordering some "real overtone" crystals cut just for me. I've been
looking at AD app note AN-419 and it's Butler oscillator, in
particular, although the clock input of the AD9951 probably has
different requirements than the AD9850 targetted in AN-419.

The built-in multiplier is quite noisy and makes the 9951 run terribly hot.

Does the AD9951 really work at 660MHz? I thought it was only good to
400MHz...

Yes, if you DON'T use the on board multiplier. I've had it to 750 MHz just
to check it since I had heard of some DL's overclocking it to that
frequency. Properly heat sunk to the eval board, and without the multiplier,
it's cool as a cucumber. AD rates it only to 400 MHz but a sample of 6 units
all operate well at 660 MHz.

So far my experimenting has used the on-chip oscillator at 25MHz and
the PLL at 16x to get to 400MHz.

Actually, it's better
than my 200 MHz 7th overtone tripled to 660

We bandied about "non-harmonic" relations here but how you get from 200
to 660, I don't know.

Well, this one is a 220 MHz 7th overtone from ICL specially surface treated
for low noise and operating in a Stephensen bipolar/FET Butler. But as I
mentioned to Doug, afraid my MMIC tripler makes a bad job of the 660 output
despite a 3 pole final filter. The SAW is not near the Q of the crystal, but
the SNR is much better.

Regards

W4ZCB