Thanks, that's good to hear there's no problem with SPI communication with the RFM69 if using the internal RC oscillator.
Assuming I'm reading the datasheet properly, there's an additional benefit to using the internal RC oscillator, and that would be much faster start-up times when waking up from powerdown sleep: just 6 clock cycles (750ns, if 8Mhz) if using internal RC oscillator (from Table 9-12) versus 16,000 clock cycles (2 milliseconds if 8Mhz) if using a crystal oscillator (Table 9-4).
Is that significant? Well, if you were relying on Listen-Mode and you were waking up the radio once a minute, then that would equate to a savings of more than 17 minutes of "waking up" time for the moteino, and the same amount of time savings for the radio if it's in Rx mode waiting for the atmega328p to respond to its packet received interrupt.
So, assume current utilized by the radio is 16mA while in Rx mode. So, that would be a savings of about 4.7mah for just the radio over the year, or about 47mah over 10 years.
Well, it's not nothing, but if running off a couple of AA's, I'd struggle to call it significant.
I need to measure how long it takes for the atmega328p to read a packet waiting for it from the radio's queue. If it's 2ms or less at 8mhz, then staying at 8Mhz would save about the same amount (i.e. not much) as compared to running at 4mhz. In that case, for simplicity, I think an argument could be made to just run at 4mhz (so I can chase declining battery voltage all the way down to 1.8v) and not worry about changing frequency.
However, if running from energizer utimate lithium batteries, even that might not be significant either, as I'd only recover about 5% (guestimate) of otherwise wasted battery energy, because the discharge curve is apparently so steep between 2.4v to 1.8v (i.e. 1.2v to 0.9v per cell).
So, unless I'm overlooking something (?), then if running off a non-coincell battery, it seems I'm already at the point of diminishing returns, even if using an atmega328p that's running off a 16Mhz/8Mhz crystal.
On the other hand, if running off a coincell for as long as possible, I can see how these optimizations might prove significant in a frequent usage scenario.
[Edit: On the other hand, if running on Alkaline batteries instead of energizer L91 lithium batteries, maybe running at 4Mhz does buy significant extra time. 8Mhz is good down to 2.4v, and there looks to be meaningful energy between 1.2 to 0.9v (per cell) in the alkaline discharge curve:
I have no idea what the dischare curves would look like for the much lower currents that would be drawn (certainly much lower than a constant 50ma), so it's hard for me to extrapolate. Is it qualitatively similar? Anyhow, I guess it's worth looking into--or perhaps just setting the clock frequency to 4Mhz and moving on--since for indoor use Alkaline's are certainly much cheaper than lithium's. It would be nice to build just once for a probable worst-case and then have a single standardized platform going forward and never need to revisit this].