A question about CPU frequency control

[Light-Yagami]

Hello everyone,

 

I'm interested about how CPU frequency is controlled in windows. We hear about "boost clocks" and all, and I'm very familiar with terminology and how it applies, but I've never necessarily (or very rarely) seen a CPU boost its clocks when doing simpler operations like browsing through the file explorer or browsing online. 

 

It seems to randomly float along its multiplier table, without any regards on what's happening on screen. It's only when I subject it to constant load or more difficult operations that I see a response - higher clocks.

 

I run a number of programs to monitor system behavior and I wonder  - since they only refresh a few times a second, is it possible that a CPU does in fact boosts to its max available multiplier for a few milliseconds to complete a given operation and returns to lower clock by the time the monitoring program refreshes the frequency value.

 

In simple terms - does CPU boost up and down too fast to be properly monitored and shown to the user.. can anyone confirm that? 

 

I also know about software controlled Speed Shift EPP and hardware controlled EPP. Windows seems to behave differently at same EPP values compared to hardware controlled SS. Hardware controlled SS is a lot more eager to jump to higher frequencies (and of course, you can force max multiplier with hardware control at all times).. So a question comes along - does hardware controlled SS improve performance? I've heard it's faster than if Windows is in control, but I have no way of checking. Also, a subsequent question - if Windows default EPP doesn't jump to higher frequencies as eagerly, does that mean that I'm effectively loosing speed because of it? In which case - should I manually lock it to a lower value? 

 

Thank you for your input!

[svmlegacy]

18 minutes ago, Light-Yagami said:

is it possible that a CPU does in fact boosts to its max available multiplier for a few milliseconds to complete a given operation and returns to lower clock by the time the monitoring program refreshes the frequency value.

Yes. AMD's marketing of the Ryzen processors praise it on it's 1 ms response time. Intel CPU's also change frequency in the order of hundreds of times per second.

 

Use the hardware clock control unless OC'ing. The OS's power plan only suggest certain actions, like less aggressively pursuing turbo clocks.

[Light-Yagami]

8 minutes ago, svmlegacy said:

Yes. AMD's marketing of the Ryzen processors praise it on it's 1 ms response time. Intel CPU's also change frequency in the order of hundreds of times per second.

 

Use the hardware clock control unless OC'ing. The OS's power plan only suggest certain actions, like less aggressively pursuing turbo clocks.

Thought so. I'll switch to hardware controlled SS then. I'm on a laptop, can't really lock the frequency high enough to not also loose single thread performance in the process. But I can force it to run at the highest available multiplier given the number of cores active. It's just that I haven't really noticed the appreciable difference in the past. Maybe Windows is missing on that extra 5% performance on purpose to save on battery power. We'll see how it goes.