CPU Clock monitoring on Zen

[Grestorn]

Hi, 

after a very long time of using my old 5960x system and AIDA64 passively, I'm now building a new machine based on Zen 2, and because of that I'm interested in the latest developments once again. 

I've just read this article about issues with AMD Zen's CPU clocking mechanics: 

https://www.techpowerup.com/review/1usmus-power-plan-for-amd-ryzen-new-developments/

What stands out there is the statement of HWInfo's author, he claims that the "traditional" way of measuring a CPU (core) clock has several issues, which is something I've been wondering for quite some time myself.

So HWInfo implemented "a new approach needs to be used called the Effective clock. This method relies on hardware's capability to sample the actual clock state (all its levels) across a certain interval, including sleeping (halted) states. The software then queries the counter over a specific polling period, which provides the average value of all clock states that occurred in the given interval." 

I certainly don't want to change to HWInfo, I'm used to AIDA64 and it's working very well in my system. So what's your stance on that statement? How is AIDA64 measuring the cpu core clocks? Is that approach already used there, too? If not, are you planning to? Or do you have a completely different view on the issue?

[Fiery]

On ‎12‎/‎3‎/‎2019 at 7:42 AM, Grestorn said:

Hi, 

after a very long time of using my old 5960x system and AIDA64 passively, I'm now building a new machine based on Zen 2, and because of that I'm interested in the latest developments once again. 

I've just read this article about issues with AMD Zen's CPU clocking mechanics: 

https://www.techpowerup.com/review/1usmus-power-plan-for-amd-ryzen-new-developments/

What stands out there is the statement of HWInfo's author, he claims that the "traditional" way of measuring a CPU (core) clock has several issues, which is something I've been wondering for quite some time myself.

So HWInfo implemented "a new approach needs to be used called the Effective clock. This method relies on hardware's capability to sample the actual clock state (all its levels) across a certain interval, including sleeping (halted) states. The software then queries the counter over a specific polling period, which provides the average value of all clock states that occurred in the given interval." 

I certainly don't want to change to HWInfo, I'm used to AIDA64 and it's working very well in my system. So what's your stance on that statement? How is AIDA64 measuring the cpu core clocks? Is that approach already used there, too? If not, are you planning to? Or do you have a completely different view on the issue?

Our stance is still that AIDA64 measures the immediate value.  I don't think measuring an average clock frequency instead of that is a good idea.  But, if you're looking for a long-term measurement to find out an average CPU clock frequency for a timespan, you can use the Logging facility of AIDA64 or the Statistics tab of the AIDA64 System Stability module.

[Grestorn]

I think this is not about averaging. It's about the issue that using a discreet multiplier to calculate the frequency is simply wrong for Ryzen. 

That's why Ryzen Master shows such a vastly different core frequency than AIDA64. 

In HWInfo shows an "effective CPU clock" for each core, which actually reflects the value Ryzen Master is showing. See this: https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/

[Fiery]

On ‎12‎/‎13‎/‎2019 at 12:33 PM, Grestorn said:

I think this is not about averaging. It's about the issue that using a discreet multiplier to calculate the frequency is simply wrong for Ryzen. 

That's why Ryzen Master shows such a vastly different core frequency than AIDA64. 

In HWInfo shows an "effective CPU clock" for each core, which actually reflects the value Ryzen Master is showing. See this: https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/

Martin's explanation there specifically talks about effective clock being an average, while the classic clock frequency measurement being instant (discrete).  IMHO it should be the responsibility of the CPU maker to make sure that by simply detecting the core multiplier the core shouldn't wake up or change its power state.  In fact, we've been requesting a more user-friendly and more accurate way of measuring CPU core clock frequency at both AMD and Intel for years, one that wouldn't require elevation (administrator privileges) as well.  We're hoping that eventually it will be implemented in all x86 processors in a proper way.

Till then, I still reckon AIDA64 is correct, simply because it reports the CPU multiplier and resulting core clock frequency as it is, without any tricks, conversions or averaging.

[dlder]

But isn't Aida for now only reporting the current (possible) "boost clock" of each core?

I write "possible clock", because most cores's a sleeping or even parked and Aida reports like 3.8GHz...

So if HWinfo is actually averaging and not pulling the current frequency from some register or what not (I'm not a programmer^^), then wouldn't it be feasable for Aida to at least show "0" MHz for Threads that are parked?

 

[Fiery]

1 hour ago, dlder said:

But isn't Aida for now only reporting the current (possible) "boost clock" of each core?

I write "possible clock", because most cores's a sleeping or even parked and Aida reports like 3.8GHz...

So if HWinfo is actually averaging and not pulling the current frequency from some register or what not (I'm not a programmer^^), then wouldn't it be feasable for Aida to at least show "0" MHz for Threads that are parked?

Yes, I guess it would be useful, but no cores would report that using normal CPU core clock measurement techniques.  AMD (and Intel as well for that matter) should come up with a better solution to provide information on the actual state of each core and their working frequency (if not sleeping).

[dlder]

You're absolutely right; they should. Esp. the more cores you have, it'll be even more interesting what each one does (thread-swapping) or how high they clock...

[f2bacon]

I can't for the life of me figure out why aida64 shows my lowest cpu clock speed as 3.60ghz.
i have a ryzen 5900x and i get as low as 800mhz and as high as 5ghz.

all the other system readings i get are close enough for me, but the cpu reading just isnt right.

if i change the windows power plan to power saver, i can clock down as low as 400mhz but aida only shows it going down to 1.72ghz. nothing lower. i don't get the mismeasurement, and can't figure out why im getting such drastically different results.

 

 

[f2bacon]

strangly enough it doesn't seem like there's an upper limit