Power Reporting Deviation on Ryzen: Exploration & Tentative Fix

00:08:03
https://www.youtube.com/watch?v=0yeBf6fOKxY

Summary

TLDRThe video explains issues with certain motherboards on AMD's AM4 platform, which misreport CPU power consumption due to discrepancies in the VRM Controller's reporting to the processor. This misreporting can lead to performance inefficiencies as the CPU either unnecessarily limits itself or uses excessive power beyond specifications. This video aims to answer why motherboards lie, how much they lie, and how to correct this. Manufacturers may intentionally misreport to deceive benchmarks and improve competitive performance. To determine the extent of the issue, users should check the 'Power Reporting Deviation' metric with HWiNFO while running a Cinebench R20 test in standard conditions (i.e., with stock CPU settings and no overclocking). Ideal deviation values should be between 90% and 110%; outside this range indicates potential deception. The video provides a troubleshooting process, including updating the BIOS and adjusting the PPT (Power Package Tracking) values. The goal is to have accurate power tracking that complies with design specifications, allowing efficient CPU operation without temperature surges or performance losses.

Takeaways

  • 🛠️ Motherboards can misreport power usage to CPUs, affecting performance.
  • 🔍 Use HWiNFO to detect power reporting deviations.
  • 📝 Run Cinebench R20 for accurate measurement.
  • ⚙️ Reset BIOS to ensure no overclocking for tests.
  • 📉 PRD between 90%-110% is typically acceptable.
  • ⚠️ Significant deviations suggest deliberate misreporting.
  • 🔄 Update BIOS to rectify potential inaccuracies.
  • 🔧 Adjust PPT in BIOS using Ryzen Master settings.
  • 💡 Accurate power tracking ensures optimal CPU performance.
  • 🔥 Misreporting can lead to unnecessary temperature rise or performance reduction.

Timeline

  • 00:00:00 - 00:08:03

    On the AMD AM4 platform, the motherboard's VRM controller communicates CPU voltage, current, and power consumption to the CPU. If the motherboard inaccurately reports power draw, the CPU may adjust performance incorrectly. Manufacturers might exploit this for competitive advantage. Power Reporting Deviation in HWiNFO measures reporting accuracy, valid under specific conditions (stock CPU, Cinebench R20 workload). A deviation within 90-110% is reasonable, but outside indicates intentional inaccuracy. To counteract, update BIOS or manually adjust PPT values using proportions. Motherboards may still misreport, but adjustments can reduce negative impacts. Re-testing performance and temperature differences is advised.

Mind Map

Video Q&A

  • Why do motherboards lie about power consumption?

    Manufacturers may underreport power consumption to gain a performance edge, although it could also be a genuine mistake.

  • How can I detect if my motherboard is lying about power consumption?

    Use the 'Power Reporting Deviation' metric in HWiNFO during a Cinebench R20 test to check for inaccuracies.

  • What should the Power Reporting Deviation (PRD) percentage be?

    A PRD between 90% and 110% is reasonable. Below 90% or above 110% indicates significant misreporting.

  • How can I fix my motherboard's power reporting inaccuracy?

    Update the BIOS and adjust the PPT value in BIOS settings to match specifications using Ryzen Master.

  • What does resetting BIOS settings do in this context?

    It ensures that overclocking is disabled, allowing for accurate testing of power reporting.

  • How does misreporting affect CPU performance?

    The CPU might either restrict itself unnecessarily or allow excessive power consumption, leading to suboptimal operation.

  • What are the steps to accurately test power consumption?

    Disable overclocking, use Cinebench R20 for full-load testing, reset sensors, and check values in HWiNFO.

  • What is the significance of the Cinebench R20 test?

    It provides a consistent, almost full-load state for accurate 'Power Reporting Deviation' readings.

  • How can misreporting affect power limits?

    Underreporting may cause the CPU to use more power than specified, while overreporting may restrict CPU power unnecessarily.

  • What should be done after detecting misreporting?

    Adjust the BIOS settings according to the corrected PPT values to align with the CPU's intended power limits.

View more video summaries

Get instant access to free YouTube video summaries powered by AI!
Subtitles
en
Auto Scroll:
  • 00:00:06
    On the AMD Platform AM4, the Motherboard VRM  Controller is responsible for communicating
  • 00:00:13
    the actual CPU voltage, current (and thus  power consumption) to the processor itself.
  • 00:00:20
    The CPU then employs this information to tweak  its operating variables so as not to violate
  • 00:00:27
    default limits / design specifications. Alas, some Motherboards lie to their CPU.
  • 00:00:36
    If the former declares a power draw higher than  the real one, the latter believes it consumes
  • 00:00:44
    more Watts than it truly does, hence the  processor will wrongly restrain itself.
  • 00:00:51
    Conversely, if the former declares a power draw  lower than the real one, the latter believes
  • 00:00:58
    it consumes less Watts than it truly does,  hence the processor will wrongly free itself.
  • 00:01:07
    Now we’re going to answer three questions.
  • 00:01:11
    One: why do they lie? Two: how much do they  lie? Three: how can we counteract their lying?
  • 00:01:23
    The response to the first query is simple: whereas  an overstatement and understatement of power
  • 00:01:30
    could both be a genuine mistake,  the second is more likely a sly
  • 00:01:34
    exploit used by the motherboard manufacturer to  achieve a performance edge over the competition.
  • 00:01:42
    The entity of this benefit depends on how  much they make their motherboards lie.
  • 00:01:48
    To answer question number two, you should learn  how to read the “Power Reporting Deviation” metric
  • 00:01:54
    in HWiNFO. As you can see, the reported value  is all over the place with a CPU in idle state,
  • 00:02:01
    and so must be disregarded; in fact, it’s  only valid with a stock CPU in a specific
  • 00:02:08
    and almost full-load state. Pay attention now,  because what I’m about to say is paramount.
  • 00:02:16
    By “stock CPU”, I mean you must disable any  kind of overclocking; the quickest way to
  • 00:02:22
    do so is reset your BIOS settings.
  • 00:02:26
    By “specific and almost full-load state”,  I mean you must use the Cinebench R20
  • 00:02:33
    multithreaded workload and nothing else; no  R23, Prime95 or any Linpack-based stress test.
  • 00:02:43
    Therefore, open Release 20 beside HWiNFO  with its polling rate set to 100ms.
  • 00:02:50
    Now click “Run”… and the “Reset Sensors”  button as soon as the rendering really starts.
  • 00:02:59
    Wait until the CPU renders the last vertical line  and press WIN + PRINTSCREEN keys halfway through…
  • 00:03:22
    Now! Next, open the screenshot
  • 00:03:36
    and write down these five values: average  Core Clocks, CPU (Tctl/Tdie), CPU PPT
  • 00:03:49
    and Power Reporting Deviation, obviously.  Oh, don’t forget the Cinebench score!
  • 00:03:57
    Finally, repeat this test at least nine more times  and compute the mean of those ten deviations.
  • 00:04:06
    In case your average Power Reporting Deviation  is 100%, well… there’s no bias at all,
  • 00:04:12
    so you can stop watching this video. If it  sits between 90 and 110%, that’s regarded as
  • 00:04:20
    a reasonable error, negative below 100 (namely,  an underestimation) and positive above 100
  • 00:04:28
    (namely, an overestimation): you could stop  watching too… or not. Lastly, if it’s less than 90
  • 00:04:36
    or more than 110, that’s absolutely an intentional  inaccuracy and you should continue watching. In
  • 00:04:44
    my particular case, my Mobo is reporting a CPU  power consumption that’s only 78% of the real one.
  • 00:04:52
    To counteract this inaccuracy, the  first thing you need to do is update
  • 00:04:57
    the BIOS to the latest version, since the  manufacturer might have already made amends
  • 00:05:03
    for their deception. In case they did not,  you’ll have to remedy it yourself, as follows.
  • 00:05:11
    Open Ryzen Master and take note of  the default PPT value for your CPU:
  • 00:05:16
    it represents the maximum power  consumption allowed per specification.
  • 00:05:21
    Now, we’re going to use your PRD and PPT  values as terms in two simple proportions:
  • 00:05:28
    the first to expose the true max CPU power  consumption allowed by our dishonest motherboard;
  • 00:05:34
    and the second to pinpoint the PPT value that  indeed realises the expected power consumption.
  • 00:05:43
    Proportion one: PRD 78% is to PPT 88W as 100% is  to X; thus, X is 112W. Basically, my Mobo lets my
  • 00:05:57
    CPU to consume a maximum of 112W instead of the  88W, as per spec… that's a difference of 24W! Please
  • 00:06:08
    note this doesn’t mean that my CPU will surely  consume 112W, because other limiters may kick in
  • 00:06:15
    before reaching that power draw… but still. Proportion two: 100% is to PPT 88W as PRD 78% is to X;
  • 00:06:27
    thus, X is 69W. In other words, I must set a PPT  of 69W in BIOS in order for my CPU to consume a
  • 00:06:38
    maximum of 88W, as per spec. To accomplish this:  enter BIOS and find the “AMD Overclocking” entry,
  • 00:06:47
    then select Precision Boost Overdrive:  Advanced > PBO Limits > PPT 69.
  • 00:06:53
    As an alternative: find the “AMD CBS” entry  and select XFR Enhancements > Precision Boost
  • 00:07:00
    Overdrive: Manual > PBO Limits > PPT 69.  In both menus, if you want, you can also
  • 00:07:07
    specify the default TDC and EDC values, which  Ryzen Master state next to SOC Power, here.
  • 00:07:17
    As you might have guessed from my answer to the  third question, yes: all in all, the Motherboard
  • 00:07:23
    is still lying to the CPU… but at least it’s  an untruth that doesn’t benefit or hurt anyone.
  • 00:07:30
    To conclude on a high note anyway,  re-run Cinebench R20 ten times
  • 00:07:36
    and compare before and after conditions: at the  cost of a few benchmark points, we have seriously
  • 00:07:43
    reduced the CPU temperature. But you may ask:  “is this difference in performance significant?”.
  • 00:07:51
    A good question, for another time. Farewell for now.
Tags
  • AMD AM4
  • Motherboard
  • VRM Controller
  • CPU Power Consumption
  • HWiNFO
  • Cinebench R20
  • Power Reporting Deviation
  • BIOS update
  • Ryzen Master
  • PPT value