AtTiny85 128khz clock and power consumption

Question

I am wondering about the Tiny85 power consumption and methods.
I cannot find answers for the following questions in the datasheet, maybe I will find it here..

The 128Khz internal watchdog clock should use less power than the 8Mhz clock [choosing 128khz will not use the 8Mhz clock, which save power].
I am assuming 8Mhz with prescaler of 64 (gives 128khz) will be waste of power, right? I can just use the 128khz internal oscillator?

Theoretically, can I use the clock prescaler even with the 128Khz clock (CLKPR register)? If yes, I can set it to /256 = 500hz clock, which should save me some more power. In the real life I cannot do it, I need ADC, which the minimum prescaler is 2, so 128Khz will give me 64kHz at max, while the range of the ADC is 50-200khz which fit well. Am I right?

Another aspect is the output PORTs. Setting all of the output ports to be input should save some more. but Should I set them to 1 or 0 (in terms of pull up resistance) to save more power?

And last thing I need to clarify - Before powering down, I am disabling ADC in order to reduce consumption. I am setting ADEN to 0 on ADCSRA.
I read about PRR register on the datasheet and I saw I can shut TIMER0, TIMER1, USI, ADC.
Will it change the power down consumption?
Will PRR ADC shut down differ than ADCSRA disabling ADC?

Thanks.

Answer 1

If you select the internal 128kHz oscillator as the clock source, the 8MHz internal oscillator should be disabled, at least according to Atmel.

Assuming the 128kHz oscillator itself (not the core running off the clock) uses less power than the 8Mhz oscillator (a pretty safe bet), then the chip should use less power when running off the 128Khz oscillator than it does running off the 8MHz oscillator divided down to 128kHz by the prescaler.

But of course all this is conjecture! If the answer is important, then this is not a hard thing to test. Program a chip with an infinite loop and then try both fuse settings and see how much power it uses in each config. If you don't have a sensitive enough current meter, you can try running the chip off a capacitor charged to a known voltage and then measure how long it takes to drop to another known voltage in each case.