IMPORTANT Explaining the AMD Ryzen "Power Reporting Deviation" -metric in HWiNFO

OK, I set negative offset voltage 0,0125 V (lowest I can get) with PBO enabled - now Power Reporting Deviation is 85% during Cinebench. :D

Unfortunately with PBO my score in Cinebench is like ~100 points lower...

OK, and how did you determine the negative offset voltage 0,0125 V is the lowest you can get? It crashes? Or doesn't boot?
Seem a quite low offset voltage... Normally on Ryzen 3000 about 0.05V should be easily feasible. And thats where you get the headroom with PBO...
 
MOBO: Asus Tuf Gaming X570-Plus, BIOS Version 1407 2020/04/10
CPU: Ryzen 7 3700X

When I start PC it gets various spike numbers between 95-120%
After 10-15 minutes in idle it's usually between 120-160%
When I run Cinebench R20 test, it gets 84%

1591893425215.png

When I run Blender Cycles Render it get straight to 74%

1591893715943.png

BIOS settings:
- DOCP enabled
- CPU voltage offset -0.1

Side note: This is my new CPU which I got from RMA using my warranty because my previous Ryzen 7 3700X died. Now I don't know if MOBO caused this, or maybe previous CPU was faulty from start.
Should I be worried about something?
 
Anyone concerned about this, there are a few things you can do... First, you can change settings in HWInfo, to make it give you more accurate results. I'll use my numbers as an example. Running Prime95 with small FFTs, I'm getting a deviation reading of 59%. So, I right clicked on "CPU Package Power" then "Customize Values", and set a multiplier of 1.7. I determined that value by dividing 59 into 100. 100/59 ~ 1.7. This forces HWInfo to multiply the reported power usage value by 1.7 when displaying it. Then, I did the same thing for the "Power Reporting Deviation" setting. Now both values are reporting much closer to what they should be. What was being reported as 65 watts is now being reported as 110 watts - 65 is about 59% of 110. And, the deviation number stays close to 100% under load now.

The next thing you can do is apply the same logic to the PPT setting in your BIOS. In my case, it's being under-reported, at 59% of what it actually is. The AMD stock value setting for PPT is 88 watts for my 3700x. 59% of 88 is about 52. So, I set the PPT if my BIOS to 52 watts, leaving the TDC and EDC at Auto. When the motherboard thinks it's limiting the processor to 52 watts, it's actually limiting it to 88 watts, which is the standard for PPT, for my 3700x, according to AMD. And yes, before you ask, performance does take a hit. Full load benchmarks are down about five percent, but maximum temperatures are down more than 10°C. Power now maxes out at about 90 watts - close enough to 88 for me, and within my CPU's specifications. Before, I was seeing power hit 120 watts and more under heavy loads, with temperatures in the mid-eighties.

Just do the math, based on your own deviation reading, and PPT for your processor. Set to stock, Ryzen Master will tell you what the PPT setting is.

Hope this helps someone...
 
Ryzen 9 3950X with Gigabyte X470 Aorus Wifi 7
BIOS F51d.

so far the temperature never exceed 80 degress, and I am using Noctua NH-D15.
Any Idea to fix the incorrrect Power Reporting Deviation?
 

Attachments

  • Test R9 3950x Hwinfo 6.27-4190.png
    Test R9 3950x Hwinfo 6.27-4190.png
    1.6 MB · Views: 21
  • Test R9 3950x Hwinfo 6.27-4190-2.png
    Test R9 3950x Hwinfo 6.27-4190-2.png
    829.6 KB · Views: 21
Last edited:
Ryzen 3950X with Gigabyte X470 Aorus Wifi 7
BIOS F51d.

so far the temperature never exceed 80 degress, and I am using Noctua NH-D15.
Any Idea to fix the incorrrect Power Reporting Deviation?
Is this at stock core voltage settings or did you undervolt?
 
Hi Martin, any view on my question I also asked on the Power Reporting Deviation figure while undervolting?
Can this affect the Power Reporting Deviation ???

Yes, it does. We stated several times that it needs to be run at stock settings.
 
Anyone concerned about this, there are a few things you can do... First, you can change settings in HWInfo, to make it give you more accurate results. I'll use my numbers as an example. Running Prime95 with small FFTs, I'm getting a deviation reading of 59%. So, I right clicked on "CPU Package Power" then "Customize Values", and set a multiplier of 1.7. I determined that value by dividing 59 into 100. 100/59 ~ 1.7. This forces HWInfo to multiply the reported power usage value by 1.7 when displaying it. Then, I did the same thing for the "Power Reporting Deviation" setting. Now both values are reporting much closer to what they should be. What was being reported as 65 watts is now being reported as 110 watts - 65 is about 59% of 110. And, the deviation number stays close to 100% under load now.

The next thing you can do is apply the same logic to the PPT setting in your BIOS. In my case, it's being under-reported, at 59% of what it actually is. The AMD stock value setting for PPT is 88 watts for my 3700x. 59% of 88 is about 52. So, I set the PPT if my BIOS to 52 watts, leaving the TDC and EDC at Auto. When the motherboard thinks it's limiting the processor to 52 watts, it's actually limiting it to 88 watts, which is the standard for PPT, for my 3700x, according to AMD. And yes, before you ask, performance does take a hit. Full load benchmarks are down about five percent, but maximum temperatures are down more than 10°C. Power now maxes out at about 90 watts - close enough to 88 for me, and within my CPU's specifications. Before, I was seeing power hit 120 watts and more under heavy loads, with temperatures in the mid-eighties.

Just do the math, based on your own deviation reading, and PPT for your processor. Set to stock, Ryzen Master will tell you what the PPT setting is.

Hope this helps someone...

What's the point if faking the results ?!
 
Hello! Is this indicator really worked out correctly? I have a 2600x in full stock 1.38-1.40v (I have a very bad sample).
Until now, I thought that I should be guided by the SVI 2 TFN indicators. This is not true?
Безымянный.png
 
OK, and how did you determine the negative offset voltage 0,0125 V is the lowest you can get? It crashes? Or doesn't boot?
Seem a quite low offset voltage... Normally on Ryzen 3000 about 0.05V should be easily feasible. And thats where you get the headroom with PBO...
When I tried put a number like 0,0100 V my bios changed it to "Auto", so I could put 0,0125 V min. or higher number in neg. offset volt. (0,3000 V is max)

By 0,0125 V I have 85% in Power Reporting Dev. I tried 0,05V but I had 80% in PRD during Cinebench.

I should have 100% in Cinebench if I good understood... So I need to put lower number like 0,0100V but that my bios doesn't allow me to do (0,0125 V is the lowest).

By 0,3V it doesn't boot up... ;) (my specs: R7 3700X, MSI MPG X570 Gaming Edge)
 
According to the latest beta (v 6.27-4190), the Power Reporting Deviation for my Asus ROG Crosshair VIII Hero (X570) with my AMD Ryzen 7 3700x is 82.4% under full multi-core load in Cinebench R20. I'm on the latest BIOS (v 1302).
How much variance should there be in these results? I did the same test on my wife's identical system and got a minimum of 77.1% and average of 78.4% (multicore full load in Cinebench R20). Re-doing my system, I got a minimum of 80.0% and average of 81.5%.
 
I did the test on my 3700X with an Asrock X570 Phantom Gaming 4. Getting somewhere between 82% and 85% during the Cinebench R20 load. I haven't done any changes to how CPU runs in the BIOS, the only thing I did is overclock RAM.
Either way, that's still well above 5% of deviation...
deviation.jpg
 
I get between 129% and 318% at idle (about 220% average), currently running Cinebench R20 and I get 128-131%.
Is it because I overclocked my CPU ?

CPU : Ryzen 5 3600, OC @ 3.8 GHz, 1.1V vCore
MB : Asus X570 TUF Gaming PLUS, bios v1407 (04/01/2020)
Temp : 38,6°C (average, mostly idle) ; 59,8°C (max = running CInebench R20)
Install the latest BIOS. That's what I had to do. My ASUS board with 3900X was doing the exact same thing.
 
When I tried put a number like 0,0100 V my bios changed it to "Auto", so I could put 0,0125 V min. or higher number in neg. offset volt. (0,3000 V is max)

By 0,0125 V I have 85% in Power Reporting Dev. I tried 0,05V but I had 80% in PRD during Cinebench.

I should have 100% in Cinebench if I good understood... So I need to put lower number like 0,0100V but that my bios doesn't allow me to do (0,0125 V is the lowest).

By 0,3V it doesn't boot up... ;) (my specs: R7 3700X, MSI MPG X570 Gaming Edge)

Ok, now I understand what you were trying to say...

But you cannot change the way the motherboard misreport the power by playing with the offset voltage...
The Power Reporting Deviation value needs to be measure at full stock settings, so no manual voltage, no offset voltage, no fixed clocks,... just everything on auto...

What I was trying to explain, was that there is a better way to sqeeuze out more processing power from your 3700x then manually putting x42 and a locked voltage.
So actually nothign to do with the reported 'Power Reporting Deviation', but your 3700x will pull less power and produce less heat if you apply the negative offset...

So from your reply I see that you were able to put 0.05V als negative offset value?

What you should do to find the optimum negative offset value is the following:

1. Take a few base readings with Cinebench R20 with everything on auto. And note down the R20 score, that is your reference...

2. - Go into the bios and put 0.05V as negative offset value (should be a good start) and see if you can find the 'CPU Loadline Calibration Control' and put it on the least agressive setting. With MSI the higher levels are less aggressive, so think you can choose level 8? If your not sure, just leave the CPU Loadline Calibration Control on Auto.
- Boot back into windows, again take a few base readings with Cinebench R20. And note down the R20 score, it should already be better than in 1. with auto settings.

3. Go into the bios and with the +/- sign got a step higher with the negative offset (believe this will then be put 0.0625V?)
- Boot back into windows, again take a few base readings with Cinebench R20. And note down the R20 score, and see if its better than the previous run...

And so on...

The trick is that from a certain (too high) negative offsett voltage your Cinebench R20 results will begin to drop. Then you know you have gone a bit too far with the undervolt.
Go back in the bios and reduced the offset a step or two and your done.

You should then have a much cooler running 3700x, but with high Cinebench scores as it now can boost higher without hitting the thermal and power thresholds...
You can then still further improve your scores by enabling PBO and allowing high number PPT numbers if you desire...
 
Edit. ASUS PRIME B450M-A . Using auto bias and auto load line? More heat and power cons from wall. Using regular load line? Worst. No bias and auto line somehow finaly worked. In have lower temps than ocn with low vcore!!!! Cinebench goes from 3845 to 3282, but i dont care...

oc.png
 
Last edited:
Per Tomshardware update to original article about this:

"AMD issued a statement today to Tom's Hardware regarding a feature from software vendor HWinfo that exposes that motherboard vendors have developed firmwares that misreport key power telemetry data to Ryzen processors. As covered in the article below, the developers stated this could have an impact on processor longevity. Here's AMD's statement:

"We are aware of the reports claiming that select motherboards may be under-reporting certain power telemetry data that could alter the performance and/or behavior of AMD Ryzen processors under certain conditions. We are looking into the accuracy of these reports.

"We want to be clear with our customers: AMD Ryzen processors contain a diverse array of internal safeguards that operate independently of external data sources. These safeguards enforce the safety and reliability of the processor during stock operation. Based on our initial assessment, we do not believe that altering external telemetry in the manner described by those public reports would have a material impact on the longevity or safety of a user's processor."

It's good to hear that AMD is investigating the matter and that based on its initial assessment, the company doesn't believe the misreported power values will cause the processor to work in a way that will have a material impact on longevity. AMD's statement doesn't entirely rule out the possibility of a reduced lifespan due to the adjustments, but given the company's engineering teams have obviously studied the matter to some extent, it's obvious they haven't yet seen any adjustments that could result in significant degradation during the warranty period that users should worry about.

The statement seemingly confirms that AMD wasn't aware of the manipulations. It will be interesting to see if motherboard vendors end the practice, or if AMD finds that the adjustments don't adversely impact longevity, the company allows it to continue. According to our testing, the condition does exist on a few motherboards we have in the lab. Stay tuned for our full report.

YOU DON'T NEED TO WORRY ABOUT YOUR RYZEN DYING!
 
What worries me, is the shit temps i get , even with insane cooling on auto. never had this problem with intel... edit. any imput from pics provided??? max confusion. i prefer to run things stock but its seems this is an alien concept for amd / asus
 
big ass aircooler. but notice the 10C increase from auto stock vs fixed oc, its more running prime / blend small ftts 72c
An overclock that raises the temperature 10 degrees for a very minimal increase of performance is not worth it and will only give you a very small fps increase for a lot more heat. You're used to overclocking Intel, and Ryzen doesn't behave the same. I'm running my R5 3600 completely stock, not even PBO, and I get very high fps in every game. Third gen Ryzens are already pushed close to the limit out of the box. That's the difference between Intel and Ryzen third gen.
 
Back
Top