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

I've got a 3600 and I started with the stock Stealth cooler, I found out within minutes of installation that cooler is a worthless pile of junk! Running Folding@home (full load to all cores) it went into the high 80s & was thermal throttling! I knew looking at the cooler it wouldn't be much cop (it's tiny & is all Alu), but it is actually incapable of cooling the CPU properly! :mad:Note this is with stock Bios settings, and the power deviation for that is ~99% (as I posted previously), and no sign of over-volting.
As a temporary fix I lowered CPU PPT from default 'auto' (max 88w) to 65w, this dropped loaded temps to about the mid 70s. It also significantly cut power usage with only a modest lost in all core loaded clock speed of 3-4% (I don't know about single core clock boost). See here for more numbers on power, temps & clock speed effects vs CPU PPT.

Anyway, I have now finally ditched that POS cooler & have fitted my ancient (but still good) Thermalright Ultima90 on it (using the TR True bolt through kit which is available again atm), that has dropped CPU temp's by about 15+C! :)
Of course I could've gone for a bigger cooler, but I had the Ultima90 lying around since I retired my Core 2 gear from my 2nd rig, and I wanted to make good use of it, also I figured it would be good enough as I won't be overlocking the 3600, and it is :).

I have an TR Ultima 90 as well, it was a beast of a cooler for the price! What temps are you getting under full load? It is currently keeping an ancient and over locked Athlon64 x3 435 Raina core CPU from burning up! I wanted to use it on new 2600X but I couldn’t get mounting hardware to do it.
 
Well, under the same conditions as the stock cooler it dropped temps by about 15C, if you click on the AnandTech thread of mine I linked above you can see the temp's for various settings.
But at the same 65w PPT with 22C ambient temps, it was going from 59-63C with all cores fully loaded running Rosetta/LHC.

You can get the AM4 adapter bracket for it now again :).
 
Last edited:
sorry " The FIT controller regulates frequency and voltage within a few msecs in very short but rapid bursts to keep performance high and silicon stress low, that no static frequency or voltage can match." this is 100% BS.

"playing" with pbo settings, prime B450M-A / r5 -3600 / eventually just copied the PPT-TDC-EDC values from my FIXED all core 4.2 ghz 1.1v oc and put those MANUALLY on bios ( PBO) (60) PPT (36) TDC (80) EDC > ( also did 45- 35- 70) GOOD. lower cpu tempS on auto! with boost on, running cinebench 20, BAD! SHIT lower overall boost not even reaching 4.2 ghz . 3.8, GHZ. NOW. in games? all cores DID go to 4.2 ghz with the capped pbo settings. BUT... im STILL SEEING CRAZY TEMPS SWINGS from 32c to 47 C vs STABLE 100% locked 32c temps with fixed 4.2 oc

so .... i give up? i just have -insanely- better EVERITHING . temps / clocks and power - from wall- with a simple static multi + "static" low voltage + (btw the later not technically true, ryzen master shows that all cores ALSO go to SLEEP, and vcore goes WAY down to NOTHING with a " static" voltage and multi ( 4.2 ghz) i really like to run things stock ... but tried everything and 1 ) my motherboards regulators ARE OFF 2) ON AUTO CORRECTED DEFAULT AND EVEN WITH LOWER PPT TDC EDC VALUES with precision boost 2 are hot garbage
 
Last edited:
This is a thread, one of those that you tired reading again and again the same questions, as if the answers are not on the OP (by The Stilt) and/or Martin's posts, the author of HWiNFO.
Read carefully please... this thread is overinflated by unessesary questions and more...

1. Power Reporting Deviation ONLY counts when CPU is on stock settings. Boost is on, with or without PBO (Disabled/Auto/Enabled). No static multis, no static voltages.
___Using any manual setting on voltage/multi disables any regulator (false or true) that the motherboard introduces and it even disables the ultimate protection of the CPU = The FIT controller that cannot be bypassed by any board regulator (false or true), ONLY IF the CPU is on auto/stock settings.

2. Power Reporting Deviation ONLY counts when CPU is on full 100% load. Any other situation is uncountable and/or irrelevant.

3. Examples:

_a. You have a R5 3600 Ryzen rated at 65W TDP with 88W max power draw (PPT). You run R20 multi benchmark and you see a 90.9% PRD and a PPT of 86W during the test. This means that the 86W you see during that test is the 90.9% of the true/actual power the CPU draws. Hence the true/actual power draw (PPT) is the difference between 90.9 and 100 =(+10%). So the true power draw is 86 + 10% = 94.6W

_b. You have a R5 3600 Ryzen rated at 65W TDP with 88W max power draw (PPT). You run R20 multi benchmark and you see a 110% PRD and a PPT of 88W during the test. This means that the 88W you see during that test is the 110% of the true/actual power the CPU draws. Hence the true/actual power draw (PPT) is the difference between 100 and 110 =(-10%). So the true power draw is 88 - 10% = 79.2W

4. When the CPU is on auto/stock settings and you see a PRD under 100% when you FULL load your CPU you dont have to fear anything. BECAUSE you keep it on auto/stock setting the internal regulator of the CPU (the silicon FITness controller) is still on and protects the chip by regulating voltage/frequency/current and eventually temp. Seeing a <100% PRD just means that the CPU is operating outside its rated PPT that AMD has set for product segmentation purposes. It is still within the silicon limits ruled by FIT. Side effect is higher temps, again within limits.

5a. ZEN2 CPUs have a throttle limiting temp of 95°C. Although the FIT controller starts the reduction of frequency (boosting) over 50°C. You must understand first that ZEN2 is a high dynamic chip, like no other, regarding boost frequency. Keep reducing temp by better cooling and it will keep boosting more at a more sustained manner. If this is worth by investing a large amount of money for cooling is a different issue and out of the scope of this discussion. We are just explaining the mechanics.

5b. ZEN2 is a CPU with no match regarding frequency and voltage regulation. What you know from previous generations or Intels must be flushed down the drain. The FIT controller regulates frequency and voltage within a few msecs in very short but rapid bursts to keep performance high and silicon stress low, that no static frequency or voltage can match. More or less there is little to no room for more performance without better cooling, without stressing the silicon out of limits. Even if metrics tells you otherwise. I cannot explain it better, please forgive me!
 
edit. "playing" with pbo settings, ( goal ? stable / rational temps / v-core when boosting, or like intel does) prime B450M-A / r5 -3600 / eventually just copied the PPT-TDC-EDC values from my fixed all core 4.2 ghz 1.1v oc and put those manually on bios ( PBO)
(60) PPT (36) TDC (80) EDC > ( also did 45- 35- 70) finally? even lower cpu temp on auto with boost on, running cinebench 20, but also lower overall boost not even reaching 4.2 ghz , BUT in games all cores DID go to 4.2 ghz with both test AND AGAIN BUT... im STILL SEEING CRAZY TEMPS SWINGS from 32c to 46 C vs 100% locked 32c temps with fixed 4.2 oc IN GAMES... so .... WHY????? i give up, AMD precision boost 2 / and PBO are garbage, maybe this works on laptops that anyway run hot, but i just have insanely better temps / clocks and power - from wall- with a simple static multi + "static" low voltage (btw the later not technically true, ryzen master shows that all cores ALSO go to SLEEP, and vcore goes WAY down to NOTHING with a " static" voltage and multi ( 4.2 ghz) i really like to run things stock ... but tried everything and stock amd is BS, volatile, ineffective CRAP.

EDIT. "Using any manual setting on voltage/multi disables any regulator (false or true) that the motherboard introduces " DINg- ding- ding- and thats why things finnaly WORK AS THEY SHOULD.
ZEN2 works as it should if mainboard vendors let it work as it should. Dont just quote a section of my post. It cancels the value and the meaning of it, one that you have miss. What you call "as they should" in reality is the way you want and not as should, or AMD intended.
And you keep comparing ZEN2 with Intel CPUs. Apples and oranges... What you call crazy temp swings exist on all chips (CPU/GPU, AMD/Intel/nVidia). The difference is that AMD is letting you see it with ZEN2 and RDNA GPUs. That is called the "hotspot" of the chip. Just because all others dont show it, doesnt mean it doesnt exist.
The main problem with ZEN2 is the mindset of users to previous generations and/or Intel/nVidia CPUs/GPUs.

ZEN2 is the most advanced CPU ever created whether you like it or not, understand it or not.

Screw the board regulator... By static voltage and/or frequency you cancel and negate the internal protection of the chip that no board can bypass. If your board's regulator tell big lies to the CPU about power draw then blame the vendor and adjust PPT accordingly to have the PRD as close to 100% as posible if you want nominal true values. But if your board causing a PRD under 100% (when on full load) then adjudting PPT and make it close to 100% means that you will loose performance. You cant have cake and eat it too...
 
i dint mean to "miss quote you", personally i just dont see the real benefit ,BECOUSE THERE IS NONE ? . forget about , motherboard regulators, boosting algorithms. the later is all marketing bs to me, my killawatt meter DOES NOT LIE ? PERIDOd. DISABLING BOOST? / PBO > lower power consumption. manual oc? hi clocks + low power consumption. AUTO? CRAP POWER CONSUMPTION. CRAP CLOCKS. CORRECTED PPT VALUES? STILL CRAP !!! ... SO MAYBE THIS ZEN2 IS TOO ADVANCED , or i need a better motherboard, i dont mind the temps swings, ( much) i do mind the 35- 50W increased wattage usage from WALL .... ( to do what? one or 2 cores at 4.2 ghz?)... coss OCD i will try AGAIN!

LOOKING AT MY posted -stock- values here... I HAVE A ( rounded up) 96 PPT ( 88 stock) - 66 TDC (60 stock ) - 98 EDC (90 ctock) . lowering this values does nothing so far .... setting x number in PBO also feeds bs numbers?
 
Last edited:
ZEN2 works as it should if mainboard vendors let it work as it should. Dont just quote a section of my post. It cancels the value and the meaning of it, one that you have miss. What you call "as they should" in reality is the way you want and not as should, or AMD intended.
[...]
Screw the board regulator... By static voltage and/or frequency you cancel and negate the internal protection of the chip that no board can bypass. If your board's regulator tell big lies to the CPU about power draw then blame the vendor and adjust PPT accordingly to have the PRD as close to 100% as posible if you want nominal true values. But if your board causing a PRD under 100% (when on full load) then adjudting PPT and make it close to 100% means that you will loose performance. You cant have cake and eat it too...
You can't adjust the PPT limit to change PRD, if that's what you meant.

About running static Vcore/clocks... I agree but I wouldn't generalize.

There is a way to run a static voltage at least in a way that AMD has intended and that is through Ryzen Master. Some boards will also let you use that way directly in the BIOS (e.g. the "AMD Overclocking" voltage setting on MSI boards). Now, what protections are disabled if you use AMD's way to set a fixed voltage, I have no idea. At least the AVFS/FIT is disabled, that's for sure.

Using that way, the CPU *SHOULD* still be able to model its power consumption (PPT/TDC) as if it were using stock settings. This does not mean it's accurate by any means. If you didn't have a 100% Reporting Deviation in the first place, it'll still be wrong. But the Deviation should remain the same. I say "should", because I'm sure some boards/controllers will still alter the load line ¯\_(ツ)_/¯

Anyway, I think anyone with a Matisse CPU can verify this, it at least works for me:
  • Set stock settings in BIOS (to make sure the board does not alter AMD's spec load line etc.)
  • Open both Ryzen Master and latest HWinfo
  • Enable PBO in Ryzen Master
  • Start Cinebench R20 and run nT passes on a loop (change the minimum run time in preferences)
  • Once stable, note down the all-core clock and PPT/TDC from ryzen master
  • Also note down the stable Core VID (effective), TFN values, and Power Reporting Deviation % from HWinfo. Core VID (effective) is the SVI2 VID instruction the CPU sends to the VR controller.
  • Now use Ryzen Master to manually set all cores to the stable clock you noted (if it happens to exactly be a selectable ratio, perfect), and manually set Core Voltage to the stable VID
  • By this point the AVFS/FIT has been disabled, yet Ryzen Master reported PPT/TDC, and HWinfo's PRD % remain the same or close to the same, if HWinfo's telemetry (SVI2 TFN) values are the same (If you did not close and reopen HWinfo after enabling the manual OC-mode in Ryzen Master, PRD % will still be shown. Sorry @Martin I didn't mean to trick your program! :) )
  • You can now even try an all-core overclock using Ryzen Master, and reported deviation should still stay the same.
(Source)

Now, if you manually set the voltage in the BIOS, the CPU's SVI2 instructions will be bypassed (there are exceptions as said above), the load line will be different, and VID (effective) will read something like 1.10V (at least for the R5 3600). This means the CPU doesn't model its power consumption as if it were at stock settings anymore, so PPT/TDC/deviation will be different.
 
Last edited:
How do i fix this ????.CB RUN (ENDING) STOCKIMG_20200711_023510_0.jpg auto.NO PBO. (Vs 47c same fullCBoad but with an all core 4.2 ghz @1.131v / 41c with no multitrd) IMG_20200711_031053_0.jpg
 
Last edited:
Hello. If I understand correctly, my stone works with underload. Is there any need to worry about what needs to be done to rectify the situation?
 

Attachments

  • Untitled-1.jpg
    Untitled-1.jpg
    419.5 KB · Views: 33
Hello. If I understand correctly, my stone works with underload. Is there any need to worry about what needs to be done to rectify the situation?
Your board understates (by about 15~16%) the power consumption of the CPU. This means that the 142W (PPT) is actually 142 + 16% = ~165W.

As many said in this thread, even if any board does that, it cannot bypass the protection of the chip if everything (CPU related) is on auto. Thats the silicon FITness controller. So even if this over-consumption leads to higher temps its still within silicon limits, and I'm not talking about only temp limits. FIT does monitor and regulates many aspects of the chip like current(A), voltage, speed etc, and keep all of them under silicon limits.

First check if you have installed latest BIOS of your board (F20). I have the same board and my PRD(Accuracy) is around 93%. So my board understates consumption only by 7~8%.
 
Your board understates (by about 15~16%) the power consumption of the CPU. This means that the 142W (PPT) is actually 142 + 16% = ~165W.

As many said in this thread, even if any board does that, it cannot bypass the protection of the chip if everything (CPU related) is on auto. Thats the silicon FITness controller. So even if this over-consumption leads to higher temps its still within silicon limits, and I'm not talking about only temp limits. FIT does monitor and regulates many aspects of the chip like current(A), voltage, speed etc, and keep all of them under silicon limits.

First check if you have installed latest BIOS of your board (F20). I have the same board and my PRD(Accuracy) is around 93%. So my board understates consumption only by 7~8%.
Thank you for the answer. Yes, I have the last BIOS (F20) installed, no overclocking, all by default. I don't want to overclock anything. It's important to me that the iron works optimally.
 
Thank you for the answer. Yes, I have the last BIOS (F20) installed, no overclocking, all by default. I don't want to overclock anything. It's important to me that the iron works optimally.
Pretty much at this point we are relaying on vendor's BIOS releases. We cant do anything to make PRD match or come close to 100%, as end-users I mean.
Make sure to install also the latest chipset drivers (June 3rd 2020) from AMD directly and not Gigabyte.

What we can do is to limit further PPT through BIOS PBO settings. This will not make PRD come close to 100% but it will reduce the actual power draw of the CPU. Temps will drop but this indicates a performance loss. By how much no one really can tell. Reducing power draw limit by lets say 10% does not mean that performance will drop 10%. It will be much less than 10%.

If your board's VRM report back to CPU that it draws 142W but it actually draws 165W, then if you limit PPT to 128W it will probably draw around 148W in reality. To do this kind of reduction from 165W to 148W it will allow less frequency and less voltage. This will have great impact on temperature. As many know when ZEN2 operates cooler and cooler it keeps frequency and voltage higher during boosting. Thats why the 10% reduction on power draw will translate into much less reduction on performance.

I hope this makes sense to you.

Also, are you using the stock cooler? Is case airflow high enough? Room(ambient) temp?
 
Last edited:
Pretty much at this point we are relaying on vendor's BIOS releases. We cant do anything to make PRD match or come close to 100%, as end-users I mean.
Make sure to install also the latest chipset drivers (June 3rd 2020) from AMD directly and not Gigabyte.

What we can do is to limit further PPT through BIOS PBO settings. This will not make PRD come close to 100% but it will reduce the actual power draw of the CPU. Temps will drop but this indicates a performance loss. By how much no one really can tell. Reducing power draw limit by lets say 10% does not mean that performance will drop 10%. It will be much less than 10%.

If your board's VRM report back to CPU that it draws 142W but it actually draws 165W, then if you limit PPT to 128W it will probably draw around 148W in reality. To do this kind of reduction from 165W to 148W it will allow less frequency and less voltage. This will have great impact on temperature. As many know when ZEN2 operates cooler and cooler it keeps frequency and voltage higher during boosting. Thats why the 10% reduction on power draw will translate into much less reduction on performance.

I hope this makes sense to you.

Also, are you using the stock cooler? Is case airflow high enough? Room(ambient) temp?

I've been able to adjust the PRD to close to 100%. On my MSI board the stock VDD full scale current is set to 170amps. It gives me a PRD of around 110%. Lowering the Full scale current to 150amps gives me 98-100% PRD

I don't know what that setting will be on other Manufactures boards.

I've also contacted MSI to see if this is an "oversight" that needs ammending.

With the PRD closer to 100% I get a slightly increased all core in Cinebench R20 with a 5c temp increase. I've reverted it back to stock for now until MSI respond.
 
I've been able to adjust the PRD to close to 100%. On my MSI board the stock VDD full scale current is set to 170amps. It gives me a PRD of around 110%. Lowering the Full scale current to 150amps gives me 98-100% PRD

I don't know what that setting will be on other Manufactures boards.

I've also contacted MSI to see if this is an "oversight" that needs ammending.

With the PRD closer to 100% I get a slightly increased all core in Cinebench R20 with a 5c temp increase. I've reverted it back to stock for now until MSI respond.
What is your CPU? Even 3900X/3950X have a max rated EDC of 140Amps.

I believe what you did made CPU power draw even higher and now the board's report understates it. Hence the 110% PRD. With that setting of 170A what HWiNFO reports for PPT/EDC/TDC, during R20 multi bench?
 
Pretty much at this point we are relaying on vendor's BIOS releases. We cant do anything to make PRD match or come close to 100%, as end-users I mean.
Make sure to install also the latest chipset drivers (June 3rd 2020) from AMD directly and not Gigabyte.

What we can do is to limit further PPT through BIOS PBO settings. This will not make PRD come close to 100% but it will reduce the actual power draw of the CPU. Temps will drop but this indicates a performance loss. By how much no one really can tell. Reducing power draw limit by lets say 10% does not mean that performance will drop 10%. It will be much less than 10%.

If your board's VRM report back to CPU that it draws 142W but it actually draws 165W, then if you limit PPT to 128W it will probably draw around 148W in reality. To do this kind of reduction from 165W to 148W it will allow less frequency and less voltage. This will have great impact on temperature. As many know when ZEN2 operates cooler and cooler it keeps frequency and voltage higher during boosting. Thats why the 10% reduction on power draw will translate into much less reduction on performance.

I hope this makes sense to you.

Also, are you using the stock cooler? Is case airflow high enough? Room(ambient) temp?
The Noctua NH-D15 fan is installed. Bios (F20). Default settings. Installed driver from AMD with your prompting.
Temperature on CPU under load not exceeding 74°C. The temperature on the Chipset, about 64°C, is this normal? The rest of the temperatures are normal. Temperature at rest, 40-45°C.
When rendering in After Effects, PPD(HWiNFO64) rotates around 100%.
 

Attachments

  • screen_2.jpg
    screen_2.jpg
    357.3 KB · Views: 13
What is your CPU? Even 3900X/3950X have a max rated EDC of 140Amps.

I believe what you did made CPU power draw even higher and now the board's report understates it. Hence the 110% PRD. With that setting of 170A what HWiNFO reports for PPT/EDC/TDC, during R20 multi bench?
I'm running a R5 1600 AF on a MSI Pro-VDH MAX. I think you misunderstood what I wrote. The STOCK VDD full scale current is set to 170amps giving the 110% PRD. This causes the board to tell the CPU its using more power than it is (According to HWInfo PRD). By lowering the VDD full scale current I can bring the PRD value to 100%. Its the VDD full scale current that is being used to calculate the power usage. Of course lowering the full scale value makes the cpu use more power as the motherboard is telling it its using less.
 
Hello everyone. I am still unable to properly understand power reporting deviation telemetry. Can someone please explain it to me using my example?
I am using ryzen 5 3600 on asus prime b450m-a motherboard (bios 2203 - the latest version) with 1 X 16GB ram stick of 3000mhz overclocked manually to 3200mhz.
While encoding video using handbrake at 100% cpu usage power reporting deviation is recorded 87.0% (red) as minimum and 119.3% as average.
While running prime95 (latest version) small fft test for maximum power and heat for 15 minutes the record is 92.9% as minimum and 119.2% average.
While running cinebench r20 the record was 85.0%(red) as minimum and 92.2% average.
Kindly please explain how big is the risk for my system at these records.
(no overclocking done on processor)
(only ram is overclocked from 3000mhz to 3200mhz)


Thank you for caring to reply.
 
Last edited:
Hello everyone. I am still unable to properly understand power reporting deviation telemetry. Can someone please explain it to me using my example?
I am using ryzen 5 3600 on asus prime b450m-a motherboard (bios 2203 - the latest version) with 1 X 16GB ram stick of 3000mhz overclocked manually to 3200mhz.
While encoding video using handbrake at 100% cpu usage power reporting deviation is recorded 87.0% (red) as minimum and 119.3% as average.
While running prime95 (latest version) small fft test for maximum power and heat for 15 minutes the record is 92.9% as minimum and 119.2% average.
While running cinebench r20 the record was 85.0%(red) as minimum and 92.2% average.
Kindly please explain how big is the risk for my system at these records.
(no overclocking done on processor)
(only ram is overclocked from 3000mhz to 3200mhz)


Thank you for caring to reply.
You can read post #285. I have examples there. And to be able to use your numbers we have to know the reported PPT value during the 100% load. And we can only discuss the 100% load/usage. All other situations are not to be accounted, and are insignificant.

A question for you...
Is there a reason why you use only 1 stick of RAM? You have cut the system's memory performance in half by not using 2 sticks and utilize dual channel.
 
You can read post #285. I have examples there. And to be able to use your numbers we have to know the reported PPT value during the 100% load. And we can only discuss the 100% load/usage. All other situations are not to be accounted, and are insignificant.

A question for you...
Is there a reason why you use only 1 stick of RAM? You have cut the system's memory performance in half by not using 2 sticks and utilize dual channel.
Thank you so much for replying. I'll get to post #285 right away.
And yes, to answer your question the reason for using 1 x 16GB stick is that my friend is about to upgrade his pc to complete rgb rig and his current rig has the same identical ram stick as mine that he will be giving to me at a much lower price and that too on installments. So I planned to go with 32GB in dual channel.
Was quite a tempting offer for me even though I might not have use for that extra ram at present. But might just come in handy in future. Hope i'm not sounding stupid.
 
Back
Top