Powerd adaptive vs high adaptive

Just sharing some real-world data in case anyone is interested...

I've always wondered how much difference I would actually see on my electric bill between between powerd modes.

I recently picked up a watt meter and have been tracking power use on my appliances for a couple days each.

Besides the fact that it's time for a new refrigerator, I discovered that on my iMac Core 2 Duo running FreeBSD 10.3, adaptive mode saves about 10% over high adaptive. About $40/year in electricity for high adaptive, and $36 for adaptive.

Of course, your mileage may vary depending on use patterns...
On my home-server I use adaptive mostly to keep temperatures low. It does only serve files and acts as music player on my local network so there is no need for max responsiveness all the time.

I used highadaptive before, but noticed that e.g. zpool scrub made the cpu run at it's max frequency (2200MHz) as long as the scrub took. With adaptive setting it bounces around between the idle 800MHz and 1400MHz with cpu temperatures being much lower. Same with backups using rsync.

Though scrubbing and rsync is mostly dependent on the disks how long it takes to get the job done. If I would be compiling a lot I would go with adaptive.

If you are only and really concerned by your powerbill, then finding the cpu's sweet spot between power-usage and processing-power is the way to go. I had a Core Duo T2700 years ago and the max 2300MHz were quite powerhungry, but at 1900MHz it didn't loose much processing-power while power-usage dropped a lot (if I remember right it needed an additional 0,25V for the last 400MHz to reach 2300MHz.

So clocking the cpu down while measuring it's processing-power and power-usage is fun to play with :) I had my Core Duo pin-modded at that time to a lower frequency and lower voldtage-id since powerd wasn't as advanced as it is now.

Today you could just use powerd -M nnnn to set your cpu's max frequency to it's sweet spot if you have a really powerhungry processor and you don't care for maximum processing-power.
I have played with that on my server and on my workstation but there is no need to reduce clock speed since I only have a Xeon-D and an Haswell i3 which are both far from power hungry.
All my hardware is Win7 or Vista vintage laptops that I leave plugged in 24/7 so I use the following in /etc/rc.conf and let the machine decide what it needs and when it needs it:

powerd_flags="-a hiadaptive -b adaptive"

Leaving 4 laptops up and running all the time does not have a noticeable effect on the power bill compared to what having 2 Dell towers with PIV processors running 24/7 did.
So I spoke a little too soon. For reasons I can't fathom, the energy use average has risen since the first day. It appears to be more like a 5% difference.
It really depends on the job you throw at your computer what saves you more energy.

Say you use adaptive mode and you compile a big port like firefox. That will fetch a lot of dependencies while the CPU ramps up when compiling and clocking down when files are beeing fetched.
That could make the overall build time longer compared to highadaptive mode where the CPU will hardly clock down on such a mixed load. The overall power used to get the job done can therefor be higher. When there is a lot of RAM in use during the build that will be even more noticable since high ram usage chews energy up as well.

For me, the adaptive mode is the way to go when you prefer low temperatures on a laptop or the fans not ramping up that much on a HTPC in your living rom.

So to say it with a few words only: adaptive mode does not necessarily use less power from the wall, but will definitely keep CPU temperatures lower in most cases.
I'm aware of all that. The usage patterns on this machine are very stable and did not vary much between the first couple days on the meter and after. This is an old iMac Core 2 Duo that I use mainly to browse the web, read email, and ssh into various servers where I do my development and other heavy work. It's a very pleasant machine to sit in front of, but not that useful for computationally intensive things, so the load is generally light.

At any rate, switching from high adaptive to adaptive did help a little and has had no noticeable impact on responsiveness.