calculate the loss in CPU usage during the shift in CPU caps

[Isaac]

My netbook has CPU power saving settings that range from On demand, Power Saver, and Performance.

Performance claims that it lets me use the full range of cpu power at all times.

On demand claims to cap processing power unless the user's actions demand a higher cap. (My understanding)

Power Saver caps CPU usage to 800Mhz, on my system.


When using the On demand setting, it possible to calculate the loss in CPU usage during the shift in CPU caps/limits?

I'm imagining a supply and demand curve, but I don't know if that loss calculation can be applied in this situation.

[fliptw]

its probably based on sustained CPU utilization.

[Krom]

For all practical purposes "On Demand" should perform identically to "Performance", and have better battery life than "Power Saver" do to the nature of modern power saving features in processors. Forced lower performance power saver modes are actually counterproductive for saving battery life on modern Intel or AMD processors.

The reason behind this requires in-depth understanding of power and leakage in transistors and is really beyond something I can explain in a single post. It is sufficient to say that modern processors save power by method of "hurry up and sleep". Meaning in balanced mode they will wake up to full performance and get the task at hand done as quickly as possible so they can get back in to a sleep state and stay there for as long as possible, while power saving mode artificially limiting the clock speed actually forces the processor to stay awake longer in order to complete its tasks which costs more energy in the long run.

[Isaac]

If they're designed to "hurry up and sleep" is performance mode killing the processor?

[Krom]

No, performance mode only kills the battery (because it prevents the processor from sleeping at all).

[Isaac]

Ah, thank you. This is something was going to ask about on the Ubuntu forums later on. Very interesting!

For my original question, I think I need to install benchmark software to really get an answer in comparing the loss in performance in the ondemand setting, or the dead weight area caused by shifting caps, vs staying in performance mode. Regardless, my system is a magnificent athlete under ubuntu and was able to outperform my friends stronger Windows 7 netbook, using flash and HTML5 benchmarks.

[Krom]

I wouldn't even bother trying to benchmark the difference between performance and on demand. If it is anything like maximum performance and balance in windows, the difference will always be within the margin of error. Yes the processor requires time in order to wake up, but the amount of time is measured in microseconds and most chips can wake up and go back to sleep thousands of times in a single second if necessary. Trying to benchmark it is next to impossible because any benchmark is going to wake up the processor when you push "go" and it is going to stay awake at full performance for the duration.

[VonVulcan]

I actually understood that Krom. Nice job of keeping it simple as always.

[Jeff250]

In the gnome panel, you can add the "CPU Frequency Scaling Monitor" to see if/when your CPU is scaling. For multi-core machines, you might want to add one for each core to get the whole picture.