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

[therealsummoner]

Just fired up HWInfo 6.28, right now running Prime95, reseted values after beginning of each test:
Asus Crosshair VIII Hero Wifi (latest BIOS v.1302)
Ryzen 9 3950x (PBO Enabled)

Prime95 Blend test:
Min: 73.6%
Max: 83.0%

Prime95 Small FFTs:
Min: 95.5%
Max: 95.9%

Prime95 Large FFTs:
Min: 72.4%
Max: 112.3%

I think however, the Small FFT test is the most representative, because of the real-deal-CPU-stress it does there, i.e. in idle situations i get also ~80% for Min and ~120% for Max readings.

 

[groundzeroday]

81.7% Min on Cinebench on ASRock x570 creator w/ 3950x, stock everything except infinity fabric clocked to 1800 for my RAM.
Getting as low as 77% on other benchmarks.
Do I just email ASRock and complain and demand a BIOS update?

 

[zlymporyllin]

That's a good point. Is there anything we can/should do to try to stop motherboard manufacturers from doing this? Is emailing them and asking them to fix it even worth the electrons it's written on? Should we try contacting AMD to see if they'll indicate their displeasure to the manufacturers? Or is it just a BOHICA moment and we should just learn to live with it?

 

[YeaSomebody]

Apologize in advance if this has been mentioned several times including maybe even the original post.....but is a red value for minimum in this category anything serious? showing me a 67.4% minimum rating for power reporting deviation.

 

[therealsummoner]

As it happens, Asus just popped options for CPU and SoC Current Telemetry, i believe this is what we look for:

 

[jawajuice]

As it happens, Asus just popped options for CPU and SoC Current Telemetry, i believe this is what we look for:

View attachment 4838

This is with the new BIOS update?

 

[YeaSomebody]

Just updated the BIOS...but man my values are way higher then what I've seen in this thread and elsewhere...most people are throwing numbers like 99 and 120 max whereas my current rating was 140 and max was showing as over 300 in hwinfo. Only number in red is my minimum though at 67...*shrugs*

 

[therealsummoner]

This is with the new BIOS update?

Yep, Asus Crosshair VIII Hero Wifi Bios 2103

 

[zlymporyllin]

Apologize in advance if this has been mentioned several times including maybe even the original post.....but is a red value for minimum in this category anything serious? showing me a 67.4% minimum rating for power reporting deviation.

The red was mentioned in one of the beta announcements:

"...
- The tolerance threshold has been increased from 5% to 10%.
- Values below 90% will be shown in red color to denote under-reporting.
..."

HWiNFO v6.27-4195 Beta released

HWiNFO v6.27-4195 Beta available. Changes: Fixed and improved monitoring of VR VCC sensor on Intel SVID VR systems. Added per-rail voltage and power monitoring on NVIDIA GPUs. Enhanced sensor monitoring on ASUS ROG STRIX B460 series. Fixed a possible crash on some systems with down-cored CPUs...

www.hwinfo.com

 

[zlymporyllin]

As it happens, Asus just popped options for CPU and SoC Current Telemetry, i believe this is what we look for:

View attachment 4838

I've got the non-Wifi version of this board and was holding off a bit on the 2103 BIOS. Maybe I'll install it and see what happens if I play with those options. I've always wanted to turn off all the Pre-Overclocking Asus automatically turns on in the BIOS, but was unsure how to do it. I guess I'll see how the system behaves with those two "telemetry" options turned to something other than Auto and then do the same with the "Force OC Mode Disable."

 

[YeaSomebody]

I'm still not quite understanding this concept...when I run a test in ryzen master it shows the value at current at 95% and 87% minimum....basically is the minimum not really the minimum usage? Also why is it reporting some ridiculously high value when idling (currently its at 160% with nothing but discord and monitoring apps open)

 

[alxns]

The weakness of this method is, that the telemetry essentially uses an undefined scale for the current (and hence power) measurements. This means that the motherboard VRM controller will send an integer between 0 - 255 to the CPU, and based the reference value known by the co-processor firmwares, this integer is converted to a figure, that represents a physical current drawn by the CPU. Based on the accurately known current flow and the voltage, it is possible to calculate to CPU power draw in Watts (V * I).

Based on that and the Gamers Nexus video on Power Reporting Deviation, we can know better understand what the doodleedoo is going on, so thanks! I never understood what the "Full Scale Current" parameter was in my BIOS because of the very convenient 8-bit range (0-255) while being described as a Current in Amps at the same time.
EDIT I've been fooled by the BIOS help box. It actually ranges from 1 to 1000.



From the GN video sharing AMD documentation info, Full Scale Current = TDC * [255 / dec_value(TDC)]. Which is TDC and what is dec_value(TDC), not sure yet, it can be figured out. The point is, it allows to directly adjust reported TDC and thus the power reporting deviation.

Anyway, I wanted to test it and came up with a chart to illustrate. Only the CPU VDD current scale was changed, SoC scale was left on auto.



Nothing really new here, but we can see the following

• At stock CPU settings and default Full Scale Current, power reporting is at 95%. TDC and SVI2 TFN Core Current are approximately the same.
• With PBO enabled and default Full Scale Current, again, TDC and SVI2 TFN Core Current are close.
• Below a certain value, reported PPT will start to fall below its 88W limit and thus, the effective limit will be raised (without enabling PBO). So PBO performance and cheating performance will start to "match". From there, everything is perfectly linear so I've removed the 0-105 range from the initial chart.
• Above a certain value, reported TDC will start to become the limiting factor, and not PPT anymore.

Notes

• Generally speaking, stock vs PBO values are not to be directly compared since I didn't do it all at once. That's why max temperatures are different. It's mostly so we can see the trends.
• beta BIOS A71 with AGESA 1.0.0.6 and HWinfo 6.28 were used
• How is HWinfo accurately reporting deviation? Mystery... how many reference values did you guys collect??

Hopefully this can help understand better. Or bring even more confusion. Bear in mind I'm no expert and that's data from just one system. I hope it's accurate. Nice to see that ASUS is now offering to tweak the parameter as well.

 

[Martin]

It looks like ASRock is giving a very unusual (unofficial) response to this story with a new BIOS update: https://www.hwinfo.com/forum/threads/agesa-combo-am4-v2-1-0-0-2.6531/

 

[alxns]

If ASRock is doing this intentionally, it's not super smart.

While with MSI (and now ASUS apparently), you can impress all your friends with your highly efficient CPU!
Look at how my 65W TDP R5 3600 is converting 700W of raw power into computing mightiness while barely getting warm

 

[groundzeroday]

It looks like ASRock is giving a very unusual (unofficial) response to this story with a new BIOS update: https://www.hwinfo.com/forum/threads/agesa-combo-am4-v2-1-0-0-2.6531/

WOW... I emailed ASRock support 2 days ago in detail, asking them for an explanation and how they intended to address this issue, including a link to this thread and some quotes from the initial post.
I didn't get a response yet, but then I see this...
I don't know how long it takes to push out a firmware update, but am I paranoid to think that I'm one of few (if any) people to directly contact them and call them out on this, that maybe led them to try and bury the issue by blocking these readings?
Also I'm on x570 Creator and see a new BIOS update as of yesterday...

 

[LookItsRain]

Brand new ASRock B550M Steel Legend. Ryzen 5 3600, Bios Version 1.00, AMD AGESA Combo-AM4 v2 1.0.0.1, Cinebench R20 Multicore Power reporting Deviation of 71.7%, consistent scores in the upper 3600 to low 3700 range.

 

[hortiSquash]

R5 3600 on a Gigabyte x470 Gaming7 here, BIOS f50 AGESA 1.0.0.4B
im getting 72W reported with a power deviation of 73%, thats 95W total
i'd like to fix that thing since my cpu runs way too hot for my taste

 

[alxns]

@hortiSquash you can limit your PPT to something like 68-70W and that should give you about the same limit as if it were stock

 

[hortiSquash]

@alxns will that affect the performance?

 

[alxns]

Compared to how your Gigabyte board was -incorrectly- set out of the box, yes Compared to how it *SHOULD* have been set, no it shouldn't.
But you should check for BIOS updates before, perhaps the setting has been changed.