Why does Zen(+) seem to idle at such a high frequency?

[mplayerMuPDF]

I have two very similar setups software wise (Debian Linux based OS with icewm as window manager/user interface, no heavy background processes running, generally just ultra lightweight with a CPU load of about 0-1.5% per core at idle) but seemingly very different hardware behavior based on monitoring the CPU frequency using the built-in icewm traybar applet on both my Ryzen (Zen+) desktop and Llano (mobile K10(.5) on 32 nm) laptop. The (quad core) Llano laptop idles at roughly 800 MHz, while my (hexacore) Zen+/X470 desktop (Biostar motherboard/American Megatrends UEFI with stock CPU settings/no OC) idles with some cores at about 3.7 GHz and others at 2.4-2.7 GHz, despite those cores obviously having much higher IPC than the practically ancient K10/Stars ones in the Llano laptop. Why don't at least some cores drop down to around, say, 1 GHz? Why do some cores stay at 3.7 GHz, which is even 500 MHz above the base clock of my CPU?!

 

[timta2]

Comparing mobile CPUs with desktop CPUs is probably starting off on the wrong foot. There are a lot of options out there, in terms of price vs. performance vs. energy savings. It often comes down to making the right choice, that's right for you or your use case.

 

[mplayerMuPDF]

Comparing mobile CPUs with desktop CPUs is probably starting off on the wrong foot. There are a lot of options out there, in terms of price vs. performance vs. energy savings. It often comes down to making the right choice, that's right for you or your use case.

Obviously there are different tradeoffs that are made but what sense does it make to have 2 cores 'idling' at their turbo clock? Anyway, I found this:

Ryzen 3600 won't down-clock at idle

(Asus ROG Strix B450-F, latest BIOS, latest chipset drivers, Windows 10) Just installed my new Ryzen 3600 and noticed that it is stuck at 4 GHz at all times and the CPU temperature continually "cycles" up and down, oscillating between high-30's and high-40's at idle. The vcore isn't locked but bo...

linustechtips.com

https://www.reddit.com/r/Amd/comments/hzvesp

looks like the reported clocks in Linux are probably not accurate because with these new (ok maybe not that new anymore...) Ryzen CPUs the frequency is adjusted extremely rapidly internally and the OS cannot detect that apparently. What does remain strange is the fact that that Llano laptop also idles at 26 C (yes, even during the recent California heatwave, admittedly in rooms with some AC but still relatively hot conditions), while my Ryzen desktop idles at 37-40C (I know, nothing to really worry about but still) and that's with an aftermarket Noctua CPU cooler and in a relatively large and pretty empty mATX case...

 

[octaivermatt]

This occurs because AMD Ryzen power profile from AMD Chipset drivers sets Zen+ to 90% min frequency/100% max frequency.
Go to Windows Power Management settings, choose advanced option and change it to 5% min / 100% max frequency.

 

[mplayerMuPDF]

This occurs because AMD Ryzen power profile from AMD Chipset drivers sets Zen+ to 90% min frequency/100% max frequency.
Go to Windows Power Management settings, choose advanced option and change it to 5% min / 100% max frequency.

Only problem is I am exclusively running Linux on this system... (on the Llano laptop as well, by the way)

 

[R0H1T]

On desktop you shouldn't set the idle/min frequency too low, you can still enable all power saving options & get decent power savings from there. Frequency ramp in zen or zen+ is rather slow so you aren't getting too many power savings when you downclock frequently.

 

[HD64G]

The cores not used are deactivated and in sleep state and the ones not used heavily shut down parts of them and do not use more energy than needed. Temperature is the proof of that. Check it out.

 

[ARF]

I have two very similar setups software wise (Debian Linux based OS with icewm as window manager/user interface, no heavy background processes running, generally just ultra lightweight with a CPU load of about 0-1.5% per core at idle) but seemingly very different hardware behavior based on monitoring the CPU frequency using the built-in icewm traybar applet on both my Ryzen (Zen+) desktop and Llano (mobile K10(.5) on 32 nm) laptop. The (quad core) Llano laptop idles at roughly 800 MHz, while my (hexacore) Zen+/X470 desktop (Biostar motherboard/American Megatrends UEFI with stock CPU settings/no OC) idles with some cores at about 3.7 GHz and others at 2.4-2.7 GHz, despite those cores obviously having much higher IPC than the practically ancient K10/Stars ones in the Llano laptop. Why don't at least some cores drop down to around, say, 1 GHz? Why do some cores stay at 3.7 GHz, which is even 500 MHz above the base clock of my CPU?!

The software doesn't report properly. It simply stops at some point after which the CPU lowers even further but it isn't noticed and showed on the display.
Your real idle clocks should be lower than those showed in front of you..

 

[Zach_01]

I do not have any experience with Linux but on windows I had a R5 3600 for almost 3 years.
1. First of all, all Ryzen have very aggressive boost. Depends on power plan/settings too.
2. 3.6GHz is the base clock of the R5 3600 and it’s max boost is 4.2GHz. So 3.7GHz is not really boost clock.
3. Most software is messing with Ryzen core states in their try to read/poll operation parameters so the software can make Ryzen to boost beyond their true state according to their true load. Also causing them to raise voltage.

From what I know HWiNFO64 (sensors mode) has the ability to use “Snapshot CPU Polling” method to minimize this “Observer Effect”.
Also HWiNFO64 has the ability to report effective clocks and those are more accurate clocks containing sleep and power down states also.
Discreet clock only reports clock when a core is in active state and this is not how you monitor a CPU today.
Discreet clock reporting will stop polling clock when a core is entering sleep or power down state. This clock is different according to power settings and could be around 3GHz or 2GHz or even below 2GHz. Sleep state can drop clock to even a single digit (MHz).

You can’t really compare a simple CPU like the K10 with the one of the most dynamic CPU on the market today. It’s way more than apples and oranges…

EDIT:
If we are talking about R5 2600 then base clock is 3.4GHz and max boost is 3.9GHz
For R5 2600X is 3.6GHz for base clock and 4.2GHz for max boost.

 

[mplayerMuPDF]

I do not have any experience with Linux but on windows I had a R5 3600 for almost 3 years.
1. First of all, all Ryzen have very aggressive boost. Depends on power plan/settings too.
2. 3.6GHz is the base clock of the R5 3600 and it’s max boost is 4.2GHz. So 3.7GHz is not really boost clock.
3. Most software is messing with Ryzen core states in their try to read/poll operation parameters so the software can make Ryzen to boost beyond their true state according to their true load. Also causing them to raise voltage.

From what I know HWiNFO64 (sensors mode) has the ability to use “Snapshot CPU Polling” method to minimize this “Observer Effect”.
Also HWiNFO64 has the ability to report effective clocks and those are more accurate clocks containing sleep and power down states also.
Discreet clock only reports clock when a core is in active state and this is not how you monitor a CPU today.
Discreet clock reporting will stop polling clock when a core is entering sleep or power down state. This clock is different according to power settings and could be around 3GHz or 2GHz or even below 2GHz. Sleep state can drop clock to even a single digit (MHz).

You can’t really compare a simple CPU like the K10 with the one of the most dynamic CPU on the market today. It’s way more than apples and oranges…

To be fair, this is not a Phenom II that we are talking about, but a Llano K10-based CPU. Llano introduced many of the power management and performance boosting technologies that we now take for granted: https://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/4 "Llano can also fully power gate the x86 CPU cores or both the CPU and GPU if the entire APU is in a deep sleep state. Being able to completely power gate CPU cores or the GPU is an important part of enabling the next major feature of Llano: Turbo Core."

EDIT:
If we are talking about R5 2600 then base clock is 3.4GHz and max boost is 3.9GHz
For R5 2600X is 3.6GHz for base clock and 4.2GHz for max boost.

I have a "2600 lite" (1600 AF, which is Zen+). Its base clock is 3.2 GHz and I think its turbo clock is 3.7 GHz.

 

[RJARRRPCGP]

To be fair, this is not a Phenom II that we are talking about, but a Llano K10-based CPU. Llano introduced many of the power management and performance boosting technologies that we now take for granted: https://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/4 "Llano can also fully power gate the x86 CPU cores or both the CPU and GPU if the entire APU is in a deep sleep state. Being able to completely power gate CPU cores or the GPU is an important part of enabling the next major feature of Llano: Turbo Core."

I have a "2600 lite" (1600 AF, which is Zen+). Its base clock is 3.2 GHz and I think its turbo clock is 3.7 GHz.

I can see why the OP believes it's based on a variant of Phenom II, because they seem to have higher IPC than Zambezi and possibly Vishera as well.

 

[Ruru]

I have a "2600 lite" (1600 AF, which is Zen+). Its base clock is 3.2 GHz and I think its turbo clock is 3.7 GHz.

You're correct. 1600AF is Zen+ and practically a 2600 with a slightly lower clocks.

 

[outpt]

2600 is a 12nm chip that will boost to 3.9ghz. While the 1600af boost to 3.7ghz and I think it is a 14nm chip thus the probable lower clock speed
Edit it is on a 12nm process.

 

[R0H1T]

1600af is also 12nm from what I remember, just sold as regular (first gen) Zen probably because they stopped making them.

 

[mplayerMuPDF]

You're correct. 1600AF is Zen+ and practically a 2600 with a slightly lower clocks.

In fact, I bought it knowing this because I could not find a 2600 for a reasonable price anymore in 2020 (when the 3600 was already available). I got it from Walmart (yes, really) online for $150 because I did not (and still don't really) have confidence in the durability/longevity of the 7 nm process that Zen 2 relies on. According to TSMC their 6 nm (really 7 nm+) is a bit more durable. Reports of rapid degradation with PBO on Zen 2 CPUs and AMD being forced to tweak their firmware only confirmed my suspicions. Only time will tell how long those 7 and 5 nm CPUs with ultra aggressive boost behavior will last but I doubt it will be as long as, say, 32 nm and 22 nm CPUs and I for one am not going to be a guinea pig. With a three year warranty there certainly is zero incentive for AMD to make them last 10 years or more; planned obsolescence and squeezing the highest performance possible out of a piece of silicon is much more profitable.

1600af is also 12nm from what I remember, just sold as regular (first gen) Zen probably because they stopped making them.

I believe that is the consensus. Its clocks match those of the original 1600. I could probably OC it to match the 2600 so I can say in my system specs that I have a real 2600 but I have no real reason to do so. I care about stability and efficiency rather than maximum performance.

 

[Dr. Dro]

In fact, I bought it knowing this because I could not find a 2600 for a reasonable price anymore in 2020 (when the 3600 was already available). I got it from Walmart (yes, really) online for $150 because I did not (and still don't really) have confidence in the durability/longevity of the 7 nm process that Zen 2 relies on. According to TSMC their 6 nm (really 7 nm+) is a bit more durable. Reports of rapid degradation with PBO on Zen 2 CPUs and AMD being forced to tweak their firmware only confirmed my suspicions. Only time will tell how long those 7 and 5 nm CPUs with ultra aggressive boost behavior will last but I doubt it will be as long as, say, 32 nm and 22 nm CPUs and I for one am not going to be a guinea pig. With a three year warranty there certainly is zero incentive for AMD to make them last 10 years or more; planned obsolescence and squeezing the highest performance possible out of a piece of silicon is much more profitable.


I believe that is the consensus. Its clocks match those of the original 1600. I could probably OC it to match the 2600 so I can say in my system specs that I have a real 2600 but I have no real reason to do so. I care about stability and efficiency rather than maximum performance.

We're exceptionally late into the 7 nm process, these not only are matured by now but they're old enough to be tried and true as well. You sound like you're just running a standard config for a basic Linux box, so there's no reason why would you distrust it, the "rapid degradation" are smarties on reddit setting a static voltage to their chip and ignoring that they're designed to rapidly oscillate their v/f curve as required. So combine a raised current limit from PBO + static voltage = you're murdering your chip. It can be 7, 12, 14, 28... any nm; if you're doing voltage + current + heat you will damage your processor.

I use PBO myself, at responsible settings of course... this Ryzen 5950X I have is probably going to outlive me if I just leave it in my rig, and I'm not even 30 yet!

ps. the 3900XT I bought a few years ago is running as great as the day I got it on my brother's rig these days. No sweats

 

[mplayerMuPDF]

We're exceptionally late into the 7 nm process, these not only are matured by now but they're old enough to be tried and true as well. You sound like you're just running a standard config for a basic Linux box, so there's no reason why would you distrust it, the "rapid degradation" are smarties on reddit setting a static voltage to their chip and ignoring that they're designed to rapidly oscillate their v/f curve as required. So combine a raised current limit from PBO + static voltage = you're murdering your chip. It can be 7, 12, 14, 28... any nm; if you're doing voltage + current + heat you will damage your processor.

I use PBO myself, at responsible settings of course... this Ryzen 5950X I have is probably going to outlive me if I just leave it in my rig, and I'm not even 30 yet!

ps. the 3900XT I bought a few years ago is running as great as the day I got it on my brother's rig these days. No sweats

I am glad to hear you have had good experiences with your 7 nm CPUs. Back in 2020H1 7 nm was still pretty new though and that is when I bought my 1600 AF. I don't have any plans to upgrade in the near future due to lack of funds and doubt I will invest any more money in AM4 as I am still experiencing USB issues on this system (as I have documented in another thread on TPU) with the latest AGESA (May 2022) and I have read there were USB issues with X570 as well. So for me the future will be ARM/RISC-V or otherwise a (probably second hand) Intel platform (LGA 1700 or whatever succeeds it). I would take the opportunity to rebuild this entire system as I am unsatisfied with this case too and would rather get an ATX case with 5.25" bays. If the economic/financial situation gets really dire than my Richland laptop could become my main system (it is mostly a backup computer right now) connected to my old Iiyama and keyboard/mouse, but let's hope it doesn't come to that.

 

[Dr. Dro]

I've been hit with the USB issues, they were fixed with AGESA 1.2.0.3c and newer, I'm on B550 (specifically ROG Strix B550-E), that is quite odd. Isn't 1.2.0.7 available for your motherboard yet?

 

[mplayerMuPDF]

I've been hit with the USB issues, they were fixed with AGESA 1.2.0.3c and newer, I'm on B550 (specifically ROG Strix B550-E), that is quite odd. Isn't 1.2.0.7 available for your motherboard yet?

I am on 1.2.0.7 and have been since June. Biostar has been nothing but great with their support. It is AMD and in particular their lousy contractor ASMedia (aka ASUS) that was responsible for the USB part of the X470 that are responsible for this. USB issues are not as bad as they used to be in 2020 and for some reason it seems only my ALFA USB 2.0 WiFi adapter is still affected by them. I at least am not experiencing any more issues with my keyboard and mouse now, so there's that. It has occurred with all USB ports: front (header), rear header (Startech USB 2.0 expansion card) and all rear USB 3.0/3.2 ports. It (at least now) only occurs after boot and IIRC resuming from suspend; replugging to another port fixes it. I don't recall having ever experienced any USB issues with any other computer I have ever used and that includes a Sandy Bridge laptop (ASUS consumer) and desktop (Lenovo ThinkCentre), HP EliteBooks (Kaveri and Carrizo), a Dell K8 laptop and the laptops in my sig (as well as some older stuff pre-2011ish).

 

[Ruru]

I should also update my BIOS, a new version came last month. Sucks that I always need to put the settings again tho.

Can you undervolt your 1600AF? My 3600 can take few negative steps offset stable. And I truly understand your system specs, I have the same case and it's a total toaster with its solid front panel.

 

[mplayerMuPDF]

I should also update my BIOS, a new version came last month. Sucks that I always need to put the settings again tho.

Can you undervolt your 1600AF? My 3600 can take few negative steps offset stable. And I truly understand your system specs, I have the same case and it's a total toaster with its solid front panel.

It's even worse than that. One of the case fans is horribly noisy (and sometimes it momentarily gets even louder than usual) and one of the motherboard holes does not line up, so my X470GTQ is only secured by three screws... Needless to say, I am incredibly dissatisfied with this relatively expensive case.

 

[Mussels]

Zen cores go the sleep and don't report clock speeds - they report the last clock speed before they 'slept'
for software to report a clock speed they have to wake up the core, or use a core already awake - and in a shocking twist being woken up and given a task makes them report a higher speed.


When Zen 2 came out, there was a big deal that Zen could update clock speeds every 1ms (I've forgotten what intel updated at) - so they could be polled 1000 times a second for temperature and clock speed readings causing high idle temps, and wasteful usage.
iCue 3.x was terrible at this as an example, you had to delete the CPU-ID exe files every update to fix the problem.


All my cores are locked at 4.6GHz, look how different HWinfo reports them vs Ryzen master

You could oversimplify it to

1. Some programs spam the CPU asking for updates, which causes higher clocks and power consumption to report more often than needed
2. Some programs make assumptions to prevent that (HWinfo is showing my static all core clock, but ignoring the downclock and sleep states, shown in a different area)

HWinfo shows my effective clocks, which could be viewed as "If 4.6Ghz but asleep 50% of the time, effective clock is 2.3GHz"

It's even worse than that. One of the case fans is horribly noisy (and sometimes it momentarily gets even louder than usual) and one of the motherboard holes does not line up, so my X470GTQ is only secured by three screws... Needless to say, I am incredibly dissatisfied with this relatively expensive case.

The case not lining up is usually an issue when the mobo was installed - if you tear the build apart make sure you've got standoffs in the correct places and insert the screws finger tight - mobos have 1-2mm wiggle room and the first screw you fully tighten is what locks it all the way in.

Personally i've found screw-less standoffs in a central location (standard on fractal cases, for example) really help ensure things line up properly

Remember that mATX has screws in places that would push into an ATX board and vice versa - often you need to take all but four out, and move them around as needed.

The amount of build's i've seen with excessive GPU sag purely because the mobo was never installed straight...

 

[Ruru]

Zen cores go the sleep and don't report clock speeds - they report the last clock speed before they 'slept'
for software to report a clock speed they have to wake up the core, or use a core already awake - and in a shocking twist being woken up and given a task makes them report a higher speed.


When Zen 2 came out, there was a big deal that Zen could update clock speeds every 1ms (I've forgotten what intel updated at) - so they could be polled 1000 times a second for temperature and clock speed readings causing high idle temps, and wasteful usage.
iCue 3.x was terrible at this as an example, you had to delete the CPU-ID exe files every update to fix the problem.


All my cores are locked at 4.6GHz, look how different HWinfo reports them vs Ryzen master
View attachment 261436
View attachment 261435


You could oversimplify it to

1. Some programs spam the CPU asking for updates, which causes higher clocks and power consumption to report more often than needed
2. Some programs make assumptions to prevent that (HWinfo is showing my static all core clock, but ignoring the downclock and sleep states, shown in a different area)

HWinfo shows my effective clocks, which could be viewed as "If 4.6Ghz but asleep 50% of the time, effective clock is 2.3GHz"
View attachment 261437


The case not lining up is usually an issue when the mobo was installed - if you tear the build apart make sure you've got standoffs in the correct places and insert the screws finger tight - mobos have 1-2mm wiggle room and the first screw you fully tighten is what locks it all the way in.

Personally i've found screw-less standoffs in a central location (standard on fractal cases, for example) really help ensure things line up properly
View attachment 261438


Remember that mATX has screws in places that would push into an ATX board and vice versa - often you need to take all but four out, and move them around as needed.

The amount of build's i've seen with excessive GPU sag purely because the mobo was never installed straight...

Bolded the most important one; mine has exactly a standoff like that. At least my Define Mini C has all the standoff locations where they should be.

 

[mplayerMuPDF]

Bolded the most important one; mine has exactly a standoff like that. At least my Define Mini C has all the standoff locations where they should be.

This was my first build, so I will check it some time to see if I did anything wrong. But anyway, it seems like the frequency monitoring on Linux is not accurate for Zen. I also wonder whether whatever icewm uses may be keeping cores awake on my system.

 

[Zach_01]

This was my first build, so I will check it some time to see if I did anything wrong. But anyway, it seems like the frequency monitoring on Linux is not accurate for Zen. I also wonder whether whatever icewm uses may be keeping cores awake on my system.

I do not have a clue how Linux works, but HWiNFO64 has a portable version for DOS, if you can use this...

Free Download HWiNFO Sofware | Installer & Portable for Windows, DOS

Start to analyze your hardware right now! HWiNFO has available as an Installer and Portable version for Windows (32/64-bit) and Portable version for DOS.

www.hwinfo.com