Reports of Ryzen 3000 High Idle Voltage Exaggerated, a Case of the "Observer Effect"

[btarunr]

With AMD's 3rd generation Ryzen processors finally falling into the hands of PC enthusiasts, many early-adopters are taking to tech communities such as ours, to share their experiences with others. A trend appears to be emerging of users reporting higher-than-usual voltages for these processors when idling. AMD investigated this phenomenon, and declared this to be a non-issue. Apparently, most modern CPU monitoring utilities cause what is known as "the observer effect:" the process of measuring the processor's load itself causes load on the processor.

In case of the Ryzen "Matisse" processors, monitoring software appear to be polling each processor core for load by sending it instruction at a high rate of speed - sending them a workload of 20 ms every 200 ms. This causes the processor's embedded firmware to think that the cores are being subjected to workload, and it responds by increasing the clock-speeds, and proportionately voltages of all CPU cores. Monitoring software poll each CPU core, and so core voltages are raised across the chip.



"We have determined that many popular monitoring tools are quite aggressive in how they monitor the behavior of a core. Some of them wake every core in the system for 20 ms, and do this as often as every 200 ms. From the perspective of the processor firmware, this is interpreted as a workload that's asking for sustained performance from the core(s). The firmware is designed to respond to such a pattern by boosting: higher clocks, higher voltages," stated Robert Hallock, AMD's head of technical marketing for processors. "So, if you're sitting there staring at your monitoring tool, the tool is constantly instructing all the cores to wake up and boost. This will keep the clock-speeds high, and the corresponding voltages will be elevated to support that boost. This is a classic case of observer effect: you're expecting the tool to give valid data, but it's actually producing invalid data by virtue of how it's measuring," he added.

Hallock recommended CPU-Z to be most accurate at measuring CPU voltages without causing the observer effect. In a screenshot shared on Reddit, Hallock showed that when twiddling its thumbs, a Ryzen 9 3900X can drop its voltages well below 0.4 V. To demonstrate his use-case, Hallock configured his machine with the latest Windows 10 May 2019 Update (version 1903), which has greater awareness of AMD "Zen" processor multi-core topology; the latest BIOS on his ASUS Crosshair VIII Hero motherboard, and AMD Chipset drivers 1.07.07, which include the latest version of the "Ryzen Balanced" Windows power-plan.

AMD highly recommends Ryzen 3000 users to use the latest version of Chipset drivers, and enable the Ryzen Balanced power-plan, which adjusts the rate at which the processor and the OS talk to each other on performance-output from the processor. With Ryzen Balanced, this is set at 1 ms, whereas the default "Balanced" power-plan provided by Microsoft polls the processor only once in 15 ms, giving users the illusion of processor voltages having "settled down." This is a sub-optimal scenario for Ryzen processors, which like to tweak their clock-speeds every 1 ms, responding to workloads better.

Hallock also prescribed a few tips to measure voltages correctly: 1, to not run multiple monitoring utilities simultaneously, which amplifies the observer effect; 2. close apps such as your motherboard's "command center" utility, Corsair iCue, NZXT CAM, etc., which too are monitoring tools; 3. Set BIOS voltages to their default or Auto values, except those voltage domains that are adjusted by your memory's XMP profile; 4. keep your chipset software, Windows version (1903 recommended), and motherboard BIOS version up to date; and 5. don't worry if you don't see <0.5 V values, <1 V is the desired idle range.

View at TechPowerUp Main Site

 

[Crackong]

Where is the cat ?

 

[blobster21]

Where is the cat ?

Who can tell ? he's half here, half always

 

[PaddieMayne]

Ive noticed this "observer effect" on my Ryzen 2600X when using the NZXT CAM software, in so much as that the CPU will not idle or downclock itself at all with CAM software running unless i choose power saving mode. However if i disable the CAM software and use CPU-Z then the CPU will idle itself in balanced mode.

 

[ratirt]

Ive noticed this "observer effect" on my Ryzen 2600X when using the NZXT CAM software, in so much as that the CPU will not idle or downclock itself at all with CAM software running unless i choose power saving mode. However if i disable the CAM software and use CPU-Z then the CPU will idle itself in balanced mode.

Does this apply for 2nd gen Ryzen? I need to double check what I get.

 

[_Flare]

"schoedingers cat" not only illustrates the observer effect, it also describes the uncertainty of the predictability.

for example the radioactive decay is a phenomenon not exactly predictable in a small timescale,
you have to contol it to be sure
and by the fact that it´s a quantumeffect, the permanent observation will destroy the inherent uncertainty of the unobserved process.

 

[ZoneDymo]

So when Volkswagens are being tested, their software limits emissions to paint a positive picture
And when Ryzen is being tested, the software increases cpu usage to paint a negative picture.

 

[Frick]

So when Volkswagens are being tested, their software limits emissions to paint a positive picture
And when Ryzen is being tested, the software increases cpu usage to paint a negative picture.

What?

 

[Spencer LeBlanc]

It's funny, because i use CPU-Z. When i overclock leaving voltage at stock on a R5 2600 and OC to 4Ghz i see voltages jump all the way up to 1.46V..
Idles between 3.6-3.9Ghz jumps anywhere from 1.0V to 1.46V

 

[ZoneDymo]

What?

Volkswagen? the car company, big scandal about using software to limit emissions when the car detects its being tested to emissions?

Ryzen, the cpu AMD makes what seemingly uses a lot of power when idle, but in fact its just the firmware reacting to the way the testing software works?

Hope that clears it up, I cant really do much with the question just comprising of "what?".

 

[Octopuss]

Where is the cat ?

What cat?

 

[Xuper]

I use Manual OC ( 3850 at 1.337 v) , Hwinfo64 with 100ms pooling period.

 

[Imsochobo]

Volkswagen? the car company, big scandal about using software to limit emissions when the car detects its being tested to emissions?

Ryzen, the cpu AMD makes what seemingly uses a lot of power when idle, but in fact its just the firmware reacting to the way the testing software works?

Hope that clears it up, I cant really do much with the question just comprising of "what?".

it's kinda true but opposite, amd cpu's look worse when checking their temps and what not and better when not looking.
VW cars were opposite, worse when not looking and better when looking

 

[Deleted member 50521]

So HWINFO or AIDA64 may have potential future update to address this?

 

[Frick]

Volkswagen? the car company, big scandal about using software to limit emissions when the car detects its being tested to emissions?

Ryzen, the cpu AMD makes what seemingly uses a lot of power when idle, but in fact its just the firmware reacting to the way the testing software works?

Hope that clears it up, I cant really do much with the question just comprising of "what?".

Yeah, but ... I mean it's ... unrelated.

 

[ratirt]

Schrodingers cat That's funny but I dont think it applies here.
About that voltage, I'm sure it will be fixed at some point.

 

[Kohl Baas]

Yeah, but ... I mean it's ... unrelated.

It was an analogy and in that a question. Does the CPU uses more power because the monitoring sowtware or it detects the software to use less power?
I vote for the first occasion, since we know that Zen is a low-power architecture manufactured on a technology meant for very hight power efficiency. That was the problem with the first generation. The technology on which it was manufactured pretty much limited it's clock speed, making it voltage-hungry above about 3GHz and almost impossible to break 4GHz without winning on the silicone lottery.

So yeah...

The only way to be sure is going after it thermally. 99% of electicity going in a CPU will came out as heat, so measure that externally and you will se the result.

 

[TheGuruStud]

People are just dumb. Mine are at .9v 2.2 ghz in every monitoring app until they sleep. If you're too dumb to know that polling can't keep up with wakes/sleeps/core hopping, then you need to leave PC building/testing to someone else.

 

[ZoneDymo]

it's kinda true but opposite, amd cpu's look worse when checking their temps and what not and better when not looking.
VW cars were opposite, worse when not looking and better when looking

which is exactly what I said, for volkswagen it paints a positive picture and for AMD a negative one.

Yeah, but ... I mean it's ... unrelated.

Not really, this is about how software makes the product looks bad and recently there was software that made a product look good.

 

[Frick]

Not really, this is about how software makes the product looks bad and recently there was software that made a product look good.

Ah yes, Big CPU bought monitoring software devs ages ago. A stripclub was blown up a few weeks ago, but up to that point it wasn't blown up. It's all connected man.

 

[Slizzo]

What do they have to say about this:

Looks like their partners making motherboards aren't helping them at all (ASUS in this particular case commanding WAYYYY too much voltage, stock.)

 

[TheGuruStud]

What do they have to say about this:

Looks like their partners making motherboards aren't helping them at all (ASUS in this particular case commanding WAYYYY too much voltage, stock.)

I have C6Hs and they're fine until you enable PBO. So....

 

[Eskimonster]

Sounds like AMD is giving blame for their own product to software issues.

 

[TheGuruStud]

Sounds like AMD is giving blame for their own product to software issues.

Sounds like Corsair makes shit software. iCue wakes the cores nonstop.

 

[Eskimonster]

Sounds like Corsair makes shit software. iCue wakes the cores nonstop.

I dont know, just the first time i hear of such issues.