Explained: AMD Ryzen CPU “Burn-Out” & Power Reporting Deviation Benchmarks (HWINFO)

by birtanpublished on August 21, 2020

Popular monitoring utility hardware info 64 has jolted life into a new discussion on Andy risin parts specifically talking about power consumption motherboard manufacturers cheating results as we've discussed about before and CPUs allegedly burning out on the am for a

Platform in this content today we're going to be looking into the new metric that's entitled power reporting deviation and hardware info it'll be addressed in three key points number one we'll be talking about Hardware infos a

New feature and explaining how it works and what it does and what it's actually showing you two we'll be talking about the headlining fear of burnouts as it's been reported of AMD CPUs as discussed by another outlet and our general

Disagreement with the premise of that and then three will be exploring the testing data we've collected and a set of benchmarks looking at this new metric in hardware info on a couple of different motherboards including one

Specifically that does deviate quite far from baseline so this will help show you which boards are juicing up and these CPUs to try and score higher in benchmarks before that this video is brought to you by the Corsair dark core

RGB Pro wireless mouse which claims sub 1 millisecond Wireless response low latency Bluetooth an 18,000 dpi sensor a 2000 Hertz polling rate and interchangeable side grips Corsairs mouse can charge on the MM 1000 now

Spoken or it can be used wired learn more at the link in the description below this piece is really interesting because in the process of researching for this piece one of the things that we discovered was that an initial reviewer

BIOS launched with a particular motherboard had a significantly higher power budget afforded to the CPU by tricking the CPU into thinking of storing less power than it is all of this stems from the ideas posed by the

Stilt and his post on hardware infos forums and we've just sort of explored it a bit more with our own approach to testing so it's interesting because in reference to that one particular motherboard you end up with the scenario

Where the later released publicly available versions of BIOS that you all would use if you bought the product are actually significantly lower in power budget than the reviewer release what and that poses a problem but because we

Test with the settings from the manufacturer that be AMD Rand tell manually enforced and we make sure that we follow the official guidance from AMD or Intel and don't let the motherboard to do whatever we want it

Doesn't really affect our test results but it has potential to affect test results or to mislead consumers or similar things like that and it's like that seee discussion all over again except with AMD this time so we'll start

With an explainer of what's going on and how all of this relates to Hardware infos new line item which is the power reporting deviation item that is measured in a percentage against 100% baseline so if you have a rise in system

You install Hardware info you can scroll down with the new beta version of the software and you'll see things like CPU package power and then you'll also see this new power deviation item and that will be represented as 100% if it's

Under for example a full all core load and it's running at the expected power budget whereas if you see something like 50% or 120% there's a deviation that is maybe undesirable or it could just be error which we'll talk about both of

Those here today this whole topic looks at a similar concept to shunt resistor mods on GPUs if you've seen our content on that it's the same basic idea except instead of modifying a shunt resistor or physically modifying a board this is all

Happening through software through BIOS on the motherboard manufacturer side of things but ultimately it's the same concept where with a shunt resistor mod you're tricking something like a Titan r-tx so think that it's drawing maybe

180 watts instead of 250 watts and by tricking it to that degree you then effectively have additional power budget that you can spend on increasing the performance of the part because you're no longer hitting as far as the GPU is

Concerned the power restrictions that have been imposed by v bios by the manufacturer it's the same type of concept here just a little bit different execution this also means that anyone using the CPU package power metric to

Report cv power consumption would be exposed to potentially bad data we take our CPU power measurements using hardware and we can pop some of those charts up on the screen from our recent CPU reviews these measure directly at

The EPS 12-volt cables for our testing so fortunately the issues that have been posed by the stilt and hardware info do not affect us and it's a good reminder as to we never really trust login software to

Begin with something we discussed at length in our CV of methods video any data using total system power measurements EPS 12-volt kill measurements or hardware info readings of boards that don't cheat will remain

Accurate but those that do cheat the results it'll actually be underreported potentially versus reality as we recently reported yet again with z4 90s launch manufacturers of Intel motherboards have a habit of playing

Fast and loose with turbo guidelines to give themselves an edge in benchmarks according to an explanation of hardware infos new metric by technical analysts of the stilt a related issue occurs with AMD boards the stilts post starts off by

Explaining that for a m4 Rison CPUs quote the voltage current and power telemetry is provided to the processor by the motherboard vrm controller through the am the spi to interface the CPU uses the numbers passed to it by the

Motherboard to keep itself within the socket and for power limits that be ppt TDC and EDC the vrm controller represents electrical current to the CPU as an integer between 0 and 255 but this integer isn't literally a

Number of amps to convert it to the real current measurement the CPU multiplies this number by a static reference number that's supplied by the motherboard this number must be calculated for each motherboard model by the manufacturer

And baked into firmware the number is referred to as the telemetry reference current the problem is that motherboard manufacturers can define this telemetry reference current as whatever they want and the CPU has to take that number at

Face value so in the unlikely event as an example that a manufacturer sets the value too high the CPU will actually think that it's drawing more power than it actually is and in that instance what happens is potentially a down clock or

Something to conform with the pvt limit and you can read our earlier piece or watch it on precision boost overdrive vs. PB 2 and how those work if you want more information and a better understanding of the various limits like

PBT and EDC 4 and the Rison processors will link that below if you want more on that but for now the important number is PBT or specifically the package our tracking so quote AMD this is the allowed socket power consumption

Permitted across the voltage rails supplying the socket applications with high thread counts and or heavy threads can encounter ppt limits end quote so the manufacturer sets the telemetry reference current value to low the CPU

Will think much like a shunt mod would do it's consuming less power than it is bypassing the PV T limit and potentially hitting higher clocks and temperatures than it should the new line and hardware info is power reporting deviation

Calculated by dividing the reference value reported by the board by the correct value as reported by the stilt quote this metric is only valid during a relatively stable near full load condition and he also suggests using a

Reporting interval of 1000 milliseconds or less and hardware info to cope with the quote highly advanced and fast power management to unwise and CPUs under these conditions then again a reading of 100% is exactly correct readings of 95%

May indicate in the Stiltz opinion intentional fudging of the results by allowing a bit more power than is intended originally readings over 100% under load all this is under full load would indicate that the CPU is being

Sort of tricked into thinking it's drawing more power than reality and therefore this instance is rare note that a single core load as an example is not reliable neither is idle and shouldn't be used

For checking this percentage number for power deviation you need to use an all core load to really do this accurately so we did check it for good measure but just for reference idle consumption you might see a number like two hundred

Seventy percent or we've seen up to five hundred sixty percent for example neither of these mean that you're drawing two hundred seventy percent or five hundred sixty percent more power than the expected baseline or that you

Have a limit that is changed by that percentage against baseline instead a higher value indicates theoretically lower power allowance but also it doesn't mean that you're drawing that much so let's just say that you have a

Lower value like fifty percent this means that you can push more power into the CPU it doesn't mean that you're going to get a doubling of the power in reality as measured at the EPS 12-volt cables it just means that that's

A distinct possibility but not that it's happening so idle and low load scenarios are not sufficient for using this metric in measurement it's not going to be accurate or at least not accurate enough as reported by hardware info and the

Still–the we do however further suspect that hardware info will rework this metric and it's naming in the future if deviation is in fact the word that they want to use to describe this it would be a lot less confusing for the users to be

Offset from zero percent rather than 100 percent which will and already has caused confusion based on people think that 100% means it's a doubling of what it should be from a an assumed baseline of zero percent so it's going to confuse

People the way it's representing now but either way 100 percent is actually equal to a zero percent deviation from baseline that's something you need to know as we get into the testing this still claims that at least two

Motherboard manufacturers are deviating as low as 50 percent meaning that their software a CPU package power reading and hardware info has potential to be half of reality you might see 60 watts instead of 120 if measured properly as

Far as we can tell the correct values are tested and calibrated per motherboard model and stored in a lookup table by hardware info meaning that the power reporting deviation relies on the accuracy of hardware infos internal

Testing of each individual motherboard model to be clear power reporting deviation is a metric invented calculated and named by hardware info not pulled from the board at least that's our understanding of it right now

We've spoken with board manufacturers and they said this isn't a value they've heard of before so it appears to be a new name coined to describe this real behavior here's our first chart tested on the gigabyte X 570 master motherboard

That we use in our CPU reviews again you can ignore the deviation between each of the four Cinebench our 20 benchmark passes because the number is only valid when it's under full load at 100% CPU utilization during runs of sentiment our

20 the reported power deviation percents dropped as low as 94% under reporting amperage by 6% isn't ideal but in the still tone piece he found that the MSI x5 70 godlike under reported by 7.14 percent and described it as a quote

Minor infraction that quote deserves the benefit of a out in any case this board isn't one of the major offenders so gigabyte may be in the clear we obviously haven't tested their entire catalog and we started with

One of the most relevant to our reviews but this one is in the clear the lines on this chart don't line up perfectly because harbour info has difficulty maintaining a constant log in interval under heavy all core loads even if you

Set it to a thousand milliseconds so without adjusting process priority or without tweaking the charts each of the lines ends up about like this the four blue Peaks corresponds to each of the four red Peaks by measuring current at

The EPS 12 volt connection and multiplying it by the power supply voltage or simply twelve here we can get an accurate measurement of packaged power and compare it against the number reported in hardware info there's one

Point where we do have some losses in the measurement and that's the vrm efficiency because we're between the CPU and the wall but that does have a VRM next to that so across four passes the actual CPU power started at 150 watts

And dropped to one forty seven point six while the software reported CPU power is starting at one forty five point five watts and remaining constant at worst that comes to a deviation of ninety seven percent better than the percentage

Reported by hardware info but fairly close theoretically the reported versus actual wattage percent should be directly equal to power deviation percent but again there are other losses and other variables that complicate

Matters such as post cable pre CPU vrm efficiency losses and also instrumentation losses it should be easier to see the correlation with a board that breaks the rules more heavy-handed Li average score for these

Cinebench are twenty multi-threaded runs was seventy one ten point five but that was with hardware info active mind you so it's not comparable to a strict are twenty only run there's a chance that the exercise 70 Master is cheating a

Little bit but it's close enough to correct that we can still use it as a baseline for performance and boost behavior as measured against the next few boards we tested as Rox X 570 Tai Chi twice first with the reviewer BIOS

That it originally shipped with P 120 and then again with the latest publicly available bios with the original BIOS power reporting deviation hit a spectacularly low 31 to 32 percent under load this is worse than the worst

Numbers by the still–the so we've done it we've found something distinctly and uniquely bad and it happens to be on a reviewer BIOS that the still may not have had access to note that we actually never

Even took this motherboard out of the box sorry as rock so it was still strapped in and it was literally exactly in the same state that it was sent to us at the launch of the 3900 acts last year but we never used it because we trusted

The exercise seventy mass or more trust well placed it seems for as rocks externally measured power consumption for the CPU across four passes of Cinebench are twenty multi-threaded we started out about 180 watts or 181 watts

At the beginning of each pass and then settled to about 170 to 171 watts by the end meanwhile the numbers reported by hardware info started at 61 to 62 watts and dropped to 59 to 60 watts by the end this comes out to a deviation of about

33 to 35 percent again a little better than the hardware in phone number but close hopefully a reviewer would realize that the 3900 X does not in fact run at 60 watts under load but we're beginning to understand why the stilt was so

Frustrated outside of the seasoned reviewer community non reviewer users might end up posting these results online and causing confusion despite the numbers being clearly inaccurate to a veteran hardware tester that's not going

To be the case for a new user also note that power consumption is affected by the auto v core applied by the board and that this will be variable board to board and have a tertiary impact on results that were not

Controlling for today to drive the point home here's the percentage of the CPUs PB T limit the 3900 X thinks it's running at under the old BIOS version versus the new one all the normally behaved tests we did for this piece

Showed the CPU at or near 100% of the ppt limit and that's what as rocks version 3.0 BIOS shows as well while the older review BIOS has the board telling the CPU that's only using half of its PPT budget although the reported power

Numbers are hugely inaccurate and the power draw for the Taichi is a good bit higher than the masters the difference in performance between the two boards is minor an average all core frequency that said the Taichi and it's old BIOS boost

To a significantly higher peak of 41 80 megahertz at at least one point max single core boost clocks with the 3900 X installed in ex-wife Sony master under all core load was 40 25 megahertz with spikes to 40 50

While the Tai Chi's it was 40 50 megahertz average that's only the maximum clocks average core clock under load for the master fluctuated around 40 20 versus around 40 45 for the Tai Chi average Cinebench score for the tai chi

With the old bios was actually slightly lower than the masters with the newest bios at 70 100 8.3 vs 71 10.5 first of all this is a hundred percent within variance Cinebench has huge variance and this is part of it so there's no real

Difference here but also a large part of the reason for the Tai Chi's lower-than-expected performance is because it's from launched a BIOS so older micro codes and plays all the rise and optimization implementations had not

Been added yet to this version will see this boost with newer biases and we'll look at that soon but we don't doubt that abusing power number variables can have an effect on performance but the more significant impacts we've observed

Are inaccurate power numbers and software and boost clocks that don't necessarily reflect the stock behavior of the CPU updating the Tai Chi to the latest publicly available BIOS version 3.0 changed behavior dramatically power

Deviation percentage stayed above 95% for the most part under load even better than the master physically measuring the power draw under load showed a near constant power consumption of 156 watts while the number reported by software

Rose from approximately 137 watts to 138 watts over the course of each past 138 is eighty eight point five percent of 156 which isn't quite the 95 percent we'd expect based on the power deviation percentage especially given that this is

The first test where we've seen a worse number calculated from physical measurements than from software the power deviation percentage is a brand new beta feature so perhaps there are some kinks to work out but we're more

Suspicious of as rockier than anything else there are a couple of ways to view the behavior we've seen from the Tai Chi on the one hand as ROC has clearly patched its bios since this time to correct the issue on the other they

Mailed us a motherboard to be reviewed with a BIOS version that lies about power consumption their BIOS support page also has a giant red caution sign with a subheading we do not recommend users to update the BIOS if their system

Is already running normally we can't know whether this wasn't but MSI has gotten in trouble for this sort of thing in the past shipping review or early model GPUs that clocked higher than the mainline it's a

Way of slightly improving first impressions and then returning to stability and reducing your RMA is by giving the customer something worse than was reviewed we dislike this for the same reasons we dislike MCE it might not

Be officially breaking the rules but it's not a fair way to represent a product especially if you're going to change it as soon as the reviewers are done with it we weren't expecting emphasize tomahawk

To have any problems given that the stilt used an msi godlike as his example for a board with the mostly correct settings it's unlikely that msi is one of the two blatant rule breakers that he alludes to and obviously as rock is at

Least half of those harbour and first power reporting deviation statistic was 99% or better under load and as seen in the chart on the screen the real life power draw of the tomahawk is accordingly close to the software

Reported number hardware infos number was actually higher than ours for once at about one forty five point five Watts under look while the logs 143 to 144 Watts externally was the results via the 12

Volt cables kudos to MSI for reporting the most accurate numbers thus far the ACS crosshair seven hero was a little looser with its numbers hitting approximately 94 percent deviation during the Cinebench test runs we saw an

Extremely constant softer important package power 140 5.2 watts with an even more constant externally measured value of 150 3.6 watts the lack of variance here maybe down to different boosting behavior on the older

X4 70 platform and this is the widest gap between software and hardware reported numbers receive you package power that we've seen yet but it's still an accident or minor tweak territory so that's all the testing data then one

Point that we're working on fully confirming is the Stiltz claim that quote despite that this exploit is essentially made possible by something and the included in the specification the use of this exploit is not something

Andy condones let alone promotes he says further saying instead they have rather actively put pressure on motherboard manufacturers who have been caught using this exploit now we can at least confirm

That Andy provides documentation to the motherboard manufacturers on how to calculate a quote gain factor and stresses it's important and we've acquired such documentation as well here's a quote from Andy's

Documentation which we can't share the document but we can share some info from it quote in some cases Andy may use this information to enable power management features therefore it is critical that this information be correct on all AMD

Based platforms later in that paragraph it says quote in order to effectively utilize current telemetry data it is necessary that the full-scale current value be validated by the platform designer and it is also true that the

Motherboard manufacturer is responsible for putting the correct reference multiplier on the board quote again using the full scale current a gain factor can be calculated this value can be programmed into the system bios by

The firmware designer and quote there are step by step instructions for calculating the full-scale current and the correct gain based on it so in short there's really no excuse for intentionally screwing up a number the

Formula is full scale current and amps equals TDC times 255 divided by the value and TDC the gain equals full-scale current and amps divided by 255 in some ways this is a big deal and in some ways it's really not it just depends on what

Angle we're talking about it's easy to miss a really key important point here which is that screwing with the power reporting does not inherently mean that the CPU is actually drawing more power so it probably will to some degree but

Not to the degree that we're seeing in the charts we've shown you like the asrock instance where you see a 32 percent number but it's not core Latorre to the actual power consumption that we're measuring it is higher but it's

Not that crazy this is where we disagree with the Tom's Hardware piece on meaningful long-term damage and it's messing with the reported CPU power it should have the same end result as basically raising PPT enabling PBO and

We've been unimpressed with the effects of PBO in the past already so there are plenty of other limits to keep the CPU in check we explore them in that previous piece and 50% power reporting deviation does not again mean that the

CPUs power consumption is instantly doubled versus stock it just doesn't work that way messing with the reported CPU power does though mean that the PBT and package power

Metrics reported in software are potentially very inaccurate hence our disclaimer earlier about taking physical power measurements the way we've always done it so many years ago we did it at the wall that's fine now we do it at the

EPS 12-volt cables and that's fine as well and all this the way we test is because even when the manufacturers aren't actively why an intentionally line the accuracy of the power metric and software still relies on their valid

Testing or at least honesty to some degree the stoats conclusion is that the distinction between this whole mass and PBO is that PBO is manually enabled by the users it's a user controlled action whereas the telemetry reference current

Is defined secretly and without user control except for on some msi boards one quote from the Stiltz piece that was absolutely blown way out of proportion by an increasingly notorious Tom's Hardware article is quote it also makes

The CPU to run hotter and potentially negatively affects its lifespan that's what was still said and that wasn't his original post this is a single speculative sentence used to emphasize what the stilt sees as a dishonest

Practice by motherboard manufacturers we've used extremely similar language discussing mze and we do agree with that that sort of one sentence almost throwaway line that is still made based on this though Tom's Hardware ran the

Headline quote Rison burnout and the board power cheats may shorten CPU lifespan with the leading line quote unbeknownst to you your motherboard may be silently killing your Rison processor faster than expected while technically

True electromigration is a minimal risk with CPUs and none of this is really meaningful further we've kind of already voiced that we feel about question marks and headlines period it's not something that I

Personally allow at our organization except for when it's literally a question asked by the audience ie asked GN so I mean a question mark and a headline is just designed to remove the responsibility from yourself and put it

On either the audience or someone else and then you're saying hey I didn't say that the Seabees are gonna burn out I was just asking if maybe they will I didn't say that this is the worst product ever I said is this the worst

Product ever normally the answer to that by the way is no but anyway the thing here is that the CPUs in question are unlikely to run far beyond their correct values anyway their stock limits and more on that in a moment the Tom's

Hardware sister site actually an intact published its own article a rebuttal that addresses the rise in burnout topic and further an auntie contact we spoke with off record expressed the opinion that quote overall Andy is so

Conservative on lifetime that it won't ever bring at electromigration into play even with PBO enabled and our contact further speculated that the absolute worst case scenario would be in three plus years you might have a twenty five

Megahertz hit to your performance that sort of the infrequent intervals like one millisecond every every now and then so if you're talking 25 megahertz maybe after several years of constant uptime and years at that point it's debatable

Whether a user or a real user could even measure it anyway you need to be in a lab environment and especially at that frequency so horizons boost clocks are already so touchy and variable that this would be difficult to detect you're more

Likely to have a change from permeation in your loop then you are from electromigration here you're more likely to have a change from your room ambient temperature going up a little bit you're more likely to have a change from moving

Across the country and how your CPU performs so the best AV comparisons for power consumption that we have with the x5 70 Taichi before and after BIOS update are the ones that really show the potential gravity of this issue the old

BIOS and the fudge tower numbers ran our 3900 acts at about 180 watts under load and with the BIOS update and the closer to correct behavior the number was something more along the lines of 156 watt so that has significant

Implications for thermals but that's mostly where the issue stops this is especially in the context of a CPU review where when you're looking at thermals what you're really doing you're saying here's kind of a reasonable

Cooler to select and if you're testing on a board at 156 Watts versus 180 that is a potential or change you might be upgrading your recommendation as a reviewer and that's what matters the most so it's not really

Dangerous and it's not going to fry a CPU it's not even going to meaningfully degrade it within the usable lifespan of the system which is a critical point but it will make you potentially run your fans at a louder rpm and there's a human

Factors impact there more so than an electrical impact you might buy a better cooler there's a financial impact there then an electrical impact or a lifespan impact to the CPU so that's it for this one thanks for watching hopefully this

Helped you we turn this all around in a day so the testing data is pretty scarce but we're happy with the results we have so far obviously there are dozens of motherboards out there if not nearing or over a hundred we're not gonna test them

All but if you have something specific you want us to look at related to this topic leave it below we'll try and find time to do it but this is going to be an overtime day for everyone involved so thanks for watching subscribe for more

Go to Stuart I Karen XS net to support this type of content specifically this is the type of research content that we try to turn around as quickly as we can and the mouse mats the mod mats or the shirts are a great way to support it or

You can go to patreon.com/scishow and access subscribe for more thanks for watching we'll see you all next time

Related Videos

because the Atari VCS is definitely something that's coming out we need to do a teardown on one to get some perspective and Atari if you don't know it&...
Hey how's it going guys Jack I'm out here with the toasty rose and today we're going to be doing a four hundred and fifty dollar PC that any of you ...
The biggest rule in testing coolers is to never trust anything don't trust the numbers don't trust the software don't trust the firmware and definit...
Hey what's up guys Jack and Matt here with the toaster brothers and today we're gonna be finding out if this 350 dollar laptop can game this is gonna be...
Everyone welcome back to another episode of ask GN sorry I'm not doing the intro with snowflake for this one we're flying to Taiwan and we need to get t...
Everyone welcome back to another hardware news recap for the week we are in Taipei right now working on a factory tour series and we actually have a point for y...
Hey how's it going guys Jack and Matt here with the toaster bros and today we're gonna be doing our first 20 20 thousand dollar gaming PC but guess what...
Coursers a 500 has been an embarrassing show of performance for the company's first revisit to air cooler isn't a long time but the upside is that it ma...
When I received the thread Ripper 3990 X on loan from a yet unnamed youtuber we quickly threw together an extreme overclocking stream and got the CPU to about f...
Every one so we're in Taiwan right now we're doing our factory tour series and while we were out here we stopped by Lian Lee's offices and they want...
Hey what is up guys Jack and Matt here with the wait what do you do it on sex you back your desk back to your desk sorry about that okay hi Jack and Matt here w...
We thought we'd found the limits of viewer interest when he posted our tour of a screw factory last year as poor Freight it was more of a screw shop than a ...
Everyone welcome back to another hardware news recap for the week we're in Taiwan where we venture outside to film things that's always a nice change of...
Hey how's it going guys Jack and Matt here with the toasty rose and today we're gonna be doing a 350 dollar gaming PC build it's gonna be really awe...
out of all the factory tours we've done until today we'd never gotten an opportunity to see how water blocks and open-loop cooling components like fitti...
Paints is one of the chief processes to dial in for product manufacturing and today's tour will bring us to a taiwanese factory replete with a mix of automa...
hey what's up guys Jack I'm Matt here with the tasty bros and today we're gonna be showing you our 2020 CES recap live from our CES Airbnb but befor...
except you can't wipe your ass with it well I mean we're going to take a gamble on the YouTube D monetization system and talk about the impact of corona...
everyone we're closing out our trip in Taipei for our factory tourists we have a lot of factory footage live already and we have more coming to the the chan...
hey what's up guys Jack and Matt here with the Thai stereos and today we're gonna be finding out is a six hundred dollar laptop worth it in 2020 it'...