Folding Forum

Consider a computer with processor-scaling feature disabled. It should run with 100% clockspeed by default. But I don't really understand about stuffs like HLT instructions, whether they really save power or not on such setup. So I want to know: If a computer doesn't have SpeedStep or Cool'and'Quiet, will it consume more power when it's folding up to 100% CPU usage, compared to when idle?

SpeedStep and Cool'and'Quiet save power by changing the voltage & speed of the processor. Any CPU will use a small amount of power when idle but that number increases as you load up the cpu to %100. For example an idle cpu will use 15-20 watts and a fully loaded one will use 90+ watts.

Hi ;

So far, from the few CPU I have owned, the PII Xeon and PIII Xeon were the only one to behave like "processor scaling disabled".

Those prehistoric chips were even cooler with some fah or seti running. Never knew if it was a winXP misfeature or a mobo disorder.

Hello syockit, welcome to the forum.

Folding will ramp up the processor to 100%, so it will use more electricity. Offset that increase by swapping out a couple regular light bulbs with CFLs. Or do 1 of any of a thousand other energy saving ideas to offset the increase.

So no, you can't fold without increasing watt use, but you can balance that out with savings in other places.

The short answer: No.

The longer, more complicated answer: No, not on anything made in the last 10 years.

entraidelec wrote:So far, from the few CPU I have owned, the PII Xeon and PIII Xeon were the only one to behave like "processor scaling disabled".

This is probably due to the chips note equipped with SpeedStep. Other than that, old BIOS which fail to recognize SpeedStep capability on newer processors also cause it to run 100% speed, because it reported to the OS as not being able to change speed.

One thing I still don't know is power consumption when idle: Wikipedia's article on CPU_power_dissipation (sorry, I can't hotlink yet) mentions that "In many applications, the CPU and other components are idle much of the time, so idle power contributes significantly to overall system power usage". Let's say for example, we have a Pentium 4 whose maximum thermal power is 85 W. Is it correct to assume that if the processor is running at 100% speed (not usage) all the time, the thermal power dissipated would be 85 W?

This distinction is important: if the above is true, then I can just configure FAH to use CPU% just below the minimum speed threshold i.e. if my processor increases speed when CPU usage reaches 40%, I can set FAH to use 35%.
But, if the above is false, and a busy CPU will still use more power than an idle CPU even when running at same speed, then I might consider not running it at all.

Speedstep ? not even SSE on those old PII !

But since Pentium II any cpu has the HLT command that is supported by oses.
You must be missing something that causes you this misunderstood.
Idle is idle, consommation is lessen. The maximal TDP is not the idle wattage, and the fixed clock doesn't prevent idle mode.

And again, maximal usage causes maximal consommation.

entraidelec wrote:Idle is idle, consommation is lessen. The maximal TDP is not the idle wattage, and the fixed clock doesn't prevent idle mode.

Okay, now I got it. Thank you for the explanation!

Every generation of processor contains new power saving features. The very early computers were either completely on or completely off. The OS called it idle, but the processor was really executing instructions that did nothing useful while waiting for useful work to show up. That time is long gone, though. In the days of the early Pentium, the chip could shut down portions of the processor without shutting everything down. Some folks overclocked their processors believing that Prime95 which was running at 100% was proof that their overclock was stable, not realizing that when FAH was running, other portions of the chip were activated, increasing the power consumption, increasing the heat, and causing the system to become unstable, even though the OS still said that it was running at 100%. Thus a new benchmark needed to be created that represented FAH's pattern of CPU use rather than Prime95's pattern.

With the soon-to-be-commonplace integration of a CPU and a GPU, we'll face a new hurdle: How to benchmark 100%CPU + 100%GPU as well as what we know very little about: What's the difference between one kind of 100% GPU and another kind. Look at the diagram for the internal construction of a GPU and recognize that not all of the subsections will be equally loaded. Can power be saved by regulating what's happening in the underutilized subsections, and if the answer is "not yet" know that it's going to happen one of these days.