Just annotations of little "how to's", so I know I can find how to do something I've already done when I need to do it again, in case I don't remember anymore, which is not unlikely. Hopefully they can be useful to others, but I can't guarantee that it will work, or that it won't even make things worse.
High CPU usage + underclock vs. low CPU + normal/high clock
Tags cpufreq-set, cpufreqd, cpulimit
This is just a question rather than a typical blog post. I know blogs aren't really supposed to be used for that, but I think it's justifiable since it's not that important or urgent, even though it has some practical applications. So I'm not "stealing" attention from other users with more pressing matters over there.
The question is basically "what gets more work done":
a) higher CPU usage per process (limiting less or not limiting the usage at all with tools such as cpulimit) but in the other hand "underclocking" a little bit (with things like cpufreqd or cpufreq-set) so that the CPU does not melt
b) having the CPU frequency next to its maximum or a bit above average, but limiting more the load of some specific applications with cpulimit.
I think that there may be advantages and disadvantages for each strategy, so perhaps it depends on what is that one is using the PC at the moment. For example, I imagine that the "b" strategy leaves the user and other processes with extra responsivity/resources, whereas "a" privileges the high-load CPU process somewhat, and everything else may be a bit sluggish, even though it's still useful for minimizing CPU temperature and maybe power consumption (even though that part is a bit complicated and sometimes a higher performance may result in less consumption in the long run, or something like that - see here and here some indirect statements with links to the original source (apparently offline momentarily) by the Intel's kernel developer, Arjan van de Ven, saying that the "ondemand" governor may be more efficient in terms of power-saving than "conservative" and even "powersave" governors)
But perhaps "a" may be more inefficient even when for some reason you want to privilege one process in detriment of the others (such as running something like folding@home when you're not using the PC), perhaps this process could get more work done even if it were more limited by cpulimit, as long as it could work in a less underclocked environment. But of course could be the other way around, perhaps this extra Mhz don't mean much, what matters most is being less restricted even if the CPU is operating on lower frequencies.
And, of course, could be that it does not make any difference any way or the other, or that it depends more on the fine tuning of "a" and "b" strategies than the strategies per se.
So, if anyone has a more enlightened perspective on the subject, I'd be curious to hear it.
The question is basically "what gets more work done":
a) higher CPU usage per process (limiting less or not limiting the usage at all with tools such as cpulimit) but in the other hand "underclocking" a little bit (with things like cpufreqd or cpufreq-set) so that the CPU does not melt
b) having the CPU frequency next to its maximum or a bit above average, but limiting more the load of some specific applications with cpulimit.
I think that there may be advantages and disadvantages for each strategy, so perhaps it depends on what is that one is using the PC at the moment. For example, I imagine that the "b" strategy leaves the user and other processes with extra responsivity/resources, whereas "a" privileges the high-load CPU process somewhat, and everything else may be a bit sluggish, even though it's still useful for minimizing CPU temperature and maybe power consumption (even though that part is a bit complicated and sometimes a higher performance may result in less consumption in the long run, or something like that - see here and here some indirect statements with links to the original source (apparently offline momentarily) by the Intel's kernel developer, Arjan van de Ven, saying that the "ondemand" governor may be more efficient in terms of power-saving than "conservative" and even "powersave" governors)
But perhaps "a" may be more inefficient even when for some reason you want to privilege one process in detriment of the others (such as running something like folding@home when you're not using the PC), perhaps this process could get more work done even if it were more limited by cpulimit, as long as it could work in a less underclocked environment. But of course could be the other way around, perhaps this extra Mhz don't mean much, what matters most is being less restricted even if the CPU is operating on lower frequencies.
And, of course, could be that it does not make any difference any way or the other, or that it depends more on the fine tuning of "a" and "b" strategies than the strategies per se.
So, if anyone has a more enlightened perspective on the subject, I'd be curious to hear it.
Total Comments 0