Question "Ryzen Burnout? AMD Board Power Cheats May Shorten CPU Lifespan" - Tom's

[UsandThem]

https://www.tomshardware.com/news/ryzen-burnout-amd-board-power-cheats-may-shorten-cpu-lifespan

Unbeknownst to you, your motherboard may be silently killing your Ryzen processor faster than expected. HWinfo introduced a new feature today that the vendor says exposes that some X570 motherboard vendors are clandestinely misreporting key measurements to AMD's Ryzen processors, thus boosting performance. Unfortunately, this tactic is similar to overclocking. It results in higher power draw and more heat production, thus potentially killing Ryzen chips sooner than expected – but all without the user's knowledge.

"I'd like to stress that despite this exploit is essentially made possible by something AMD has included in the specification, the use of this exploit is not something AMD condones with, let alone promotes. Instead they have rather actively put pressure on the motherboard manufacturers, who have been caught using this exploit," The Stilt added.

At least there is a free HWinfo tool (linked to at the end of the article) that lets people see if their motherboard behaves this way.

 

[DrMrLordX]

I was going to say, weren't they caught pulling this stuff with Intel CPUs a few years back? Which that's why there's more leniency with AMD, because we know who's to blame for it because of the previous situation.

1). Some mobo vendors were caught doing this with Z490
2). Intel is thought to be somewhat complicit, though obviously there is no smoking gun. Intel's power management makes it very easy for mobo OEMs to fudge values. It may be more accurate to say that Intel hasn't done anything concrete to stop mobo OEMs from "cheating".
3). AMD's power management makes it harder (though obviously not impossible) for OEMs to fudge power delivery. Mobo OEMs have had to go to a great deal more effort to screw things up in order to (maybe) win some review benchmarks.

Whether or not AMD deserves leniency really depends on whom you believe to be responsible for motherboards feeding extra power @ stock. The Tom's article shows a screenie of a board feeding twice the reported power to the CPU running CBR20, which is actually quite ridiculous, and it's a wonder how anyone wouldn't notice that from thermals or power readings taken at the wall. I guess there are some people out there that don't have their rig hooked up to a Kill-a-Watt all the time.

 

[piokos]

Intel's power management makes it very easy for mobo OEMs to fudge values. It may be more accurate to say that Intel hasn't done anything concrete to stop mobo OEMs from "cheating".

"STOP"?!
Intel CPUs are built around the idea of being easy to tune for OEM PCs. They're made mostly for OEMs.

Has anyone here made wall power observations (via a tool like Kill-a-Watt or otherwise) to try and observe the effect noted by HWiNFO64?

Reviews showing whole system power consumption usually show
Zen with smaller advantage than those that those relying on sensor reporting (say: HWMonitor). The usual comment is that the chipset pulls more. Or RAM. Or something.

 

[DrMrLordX]

"STOP"?!
Intel CPUs are built around the idea of being easy to tune for OEM PCs. They're made mostly for OEMs.

As you can see, Intel has left wiggle room for the OEMs to pull as much power as they want for as long as they want, as long as they don't exceed 4.9 GHz all-core. That leads to boards that make a CPU burn ~230W in perpetuity while running Blender (for example). That is way out of spec. "Normal" behavior for a 10900k beyond 56s is to drop to 125W, which should be 4.3 GHz but winds up being different speeds depending on the motherboard (which is actually kind of weird).

Intel is absolutely supposed to stop that behavior. Don't give me that "Intel CPUs are made for OEMs" garbage. It's totally irrelevant. OEM prebuilt systems will not be specced with massive (and expensive) cooling solutions to aid-and-abet the blatant violation of power standards observed in motherboards designed for the DiY crowd. OEMs will use custom motherboards that will likely follow Intel power specs because it will lower their BoM and increase their profits.

Reviews showing whole system power consumption usually show
Zen with smaller advantage than those that those relying on sensor reporting (say: HWMonitor). The usual comment is that the chipset pulls more. Or RAM. Or something.

Smaller advantage than what? Do you understand what I'm saying here? I'm talking about a given 3900x being compared against itself in different motherboards.

 

[Zucker2k]

3). AMD's power management makes it harder (though obviously not impossible) for OEMs to fudge power delivery.

How so? Please explain.

It's amazing how bias you can be towards AMD! You've got the nerve to suggest Intel is complicit but AMD is not, when AMD has made zero effort to indicate the existence and manipulation of this feature (why is it even a feature on the package?) when you yourself have admitted it was reported on day one by The_Stilt. Maybe because it helps inflate benchmark numbers for AMD's benefit?

Reviews showing whole system power consumption usually show
Zen with smaller advantage than those that those relying on sensor reporting (say: HWMonitor). The usual comment is that the chipset pulls more. Or RAM. Or something.

This.
Package Power, performance per watt for AMD chips need to be thrown out the window because this has to be the biggest cheating feature, deliberate or non-deliberate, in recent memory. At least with Intel, PL2 and TAU are readily observable. To deliberately misreport/hide actual power consumption? That's just dishonest, and frankly, criminal.

 

[Zucker2k]

I've observed an unusually power delta/discrepancy between package power and system power that was difficult to explain for purely cpu benchmarks. Now, it all makes sense.

 

[darkswordsman17]

1). Some mobo vendors were caught doing this with Z490
2). Intel is thought to be somewhat complicit, though obviously there is no smoking gun. Intel's power management makes it very easy for mobo OEMs to fudge values. It may be more accurate to say that Intel hasn't done anything concrete to stop mobo OEMs from "cheating".
3). AMD's power management makes it harder (though obviously not impossible) for OEMs to fudge power delivery. Mobo OEMs have had to go to a great deal more effort to screw things up in order to (maybe) win some review benchmarks.

Whether or not AMD deserves leniency really depends on whom you believe to be responsible for motherboards feeding extra power @ stock. The Tom's article shows a screenie of a board feeding twice the reported power to the CPU running CBR20, which is actually quite ridiculous, and it's a wonder how anyone wouldn't notice that from thermals or power readings taken at the wall. I guess there are some people out there that don't have their rig hooked up to a Kill-a-Watt all the time.

Yeah, that's absurd to be pushing that much extra through. Its like they wanted to get caught. And for probably a pretty miniscule gain?

Yet another reason I wish AMD would've had the gall to do something crazy like sell a complete engineered system and sell it to OEMs which would then just be tasked with coming up with the enclosure (and possible accessories but in a modular sort of way similar to that one HP one from awhile back, or how companies developed prodcuts to stack with the Mac Mini). Which, was there ever much investigation into the pricing of the X570 boards? I still can't get over the pricing on them, although I'm sure some more reasonable ones have been out, and certainly the delay in 550 boards is partly the issue there, but its one of the reasons why I've held off on Zen 2 was waiting for more reasonable boards prices with 550.

 

[DrMrLordX]

How so? Please explain.

It's pretty simple. AMD uses a standard firmware blob (AGESA) that all motherboards must implement for their CPUs to work. You can read a bit more about it here:

AGESA - Wikipedia

en.wikipedia.org

It's been around for awhile, though as the article notes, it's relevance has come to the fore on AM4. AMD uses it to control boost behavior and power usage. It is explicitly mandated in the AGESA firmware that CPUs will have a maximum PPT based on the SKU. Motherboard OEMs are not permitted to change that spec at all. In order for mobo OEMs to cheat, they had to deliberately misreport sensor readings to the firmware to make the firmware violate its own power limits.

It's amazing how bias you can be towards AMD! You've got the nerve to suggest Intel is complicit but AMD is not

When I'm correct, I'll be as nerve-y as I please. See above video from Gamer's Nexus. Intel has left it up to the motherboard OEM how they wish to set PL values and tau limits. As long as that 10900k stays at 4.9 GHz all-core or lower, they just don't care. AMD seized control of their own platform and forbade OEMs from doing anything like that.

when AMD has made zero effort to indicate the existence and manipulation of this feature

You don't know that. Now I'll grant you, they could probably have listened to The_Stilt some more (personally I think they should listen to him more often; I think AMD had started ignoring him about three years ago), but you have no idea what they're doing or what they're saying to motherboard OEMs in the background.

The fact that it took a guy as dedicated as The_Stilt and whoever it is that's helping him a year to expose this to the public tells us it's not something that's particularly easy to diagnose or describe. It very well could be that AMD had no idea this was happening. It might explain @VirtualLarry 's crazy-arsed 3600 though.

when you yourself have admitted it was reported on day one by The_Stilt.

He talked about it on OCN, and then went largely silent. One of the problems is probably that different UEFI revs may or may not have sensor cheats built-in. I know he fingered Asus back then. The other problem is that almost nobody picked up on it in reviews, or even running their own hardware. It's pretty obvious why I didn't, because apparently my board behaves itself.

Maybe because it helps inflate benchmark numbers for AMD's benefit?

Doubtful. If you look at most of the 3900x reviews, the default scores it gets are below mine, and my board doesn't cheat. For example, a casual run of CBR15 on my machine produces a 3298, while Anandtech's 3900x on an MSI x570 Meg Axe produces only 3102. I don't think I've found one review where a 3900x @ stock can beat mine on CBR15 or CBR20.

Package Power, performance per watt for AMD chips need to be thrown out the window

You can throw anything you want out the window. Don't expect anyone to pay attention.

Yet another reason I wish AMD would've had the gall to do something crazy like sell a complete engineered system and sell it to OEMs

Well they kind of did? AMD's AGESA doesn't permit PPT limits to be altered at all. The problem is that AMD doesn't control sensors external to the CPU package. Except for the chipset but that's irrelevant to the topic at hand.

 

[coercitiv]

Package Power, performance per watt for AMD chips need to be thrown out the window because this has to be the biggest cheating feature, deliberate or non-deliberate, in recent memory. At least with Intel, PL2 and TAU are readily observable. To deliberately misreport/hide actual power consumption? That's just dishonest, and frankly, criminal.

I warned you the day might come when you will consider it proper to criticize AMD for the way they handle their TDP ratings (maybe rightfully so), and that will require you you to choose a wider, brand agnostic stance on power related issues.

Here's a sample of what happens when you start throwing stones in a glass house.

VCCIN - Haswell Input Voltage. Different motherboard manufacturers have slightly different names for this voltage but it is basically the source voltage from which all other parts of the CPU get their voltage from. On desktop CPUs that are significantly overclocked, I have heard that setting the input voltage about 0.6V higher than the CPU core voltage is appropriate. 1.9072V seems like a reasonable and safe value for Haswell VCCIN. Default for my 4700MQ is somewhere around 1.80V. I get a BSOD at about 1.775V.

What I found during testing is that there seems to be a bug within these CPUs. Setting VCCIN a little on the high side tricks the CPU. Intel CPUs do not measure actual power consumption. They calculate an approximate power consumption value internally and then they use this data to control the Turbo Boost feature. I found that a high VCCIN setting results in less reported power consumption. If the CPU thinks it is using less power, more turbo boost will be available so you will get less turbo throttling. A neat trick but it would be wise not to get too carried away. The maximum allowed setting for my 4700MQ is 2.30V but I would not want to be running VCCIN that high 24/7.

Friendly reminder that although Skylake no longer uses an IVR, Skylake-X does use IVR and VCCIN is very much alive. AFAIK the same is true for the Cove cores. Thread lightly.

 

[DrMrLordX]

By the way, for those who are still confused about what's going on, you are encouraged to read this post by The_Stilt:

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

Ryzen CPUs for AM4 platform rely on external, motherboard sourced telemetry to determine their power consumption. The voltage, current and power telemetry is provided to the processor by the motherboard VRM controller through the AMD SVI2 interface. This information is consumed by the processors...

www.hwinfo.com

(linked from the Tom's article)

AMD doesn't control the sensors external to the SVI2 interface, so the mobo OEM can feed false data to the CPU to subvert power limits. If you look at the MSI Meg Ace example where the MSI board actually allows the user to tweak the extent to which the board "cheat"s in the UEFI, cheating allows at most an 80 MHz overclock on a 3700x due to limitations set in the CPU's FIT tables. It certainly looks like some of the OEMs actively tried to sneak around PPT limits and ran into FIT voltage limits instead, which probably explains why these cheats didn't show up in reviews much.

Also, according to The_Stilt, AMD is actively pressuring mobo OEMs to stop doing this, but obviously that hasn't yet worked.

 

[lobz]

Wait, so is it now official on this forum?
- High power doesn't kill CPUs.
- OEMs pushing CPUs over default specification is fine.

Isn't these among the most popular Comet Lake-S criticism points?

I don't understand the leniency towards AMD in latest comments from Cutress.
Also, I love how he categorized PC users as:
- doing manual tweaking,
- DIY crowd buying excessive cooling,
- DIY crowd doing SFF and monitoring thermals.

Stop seeing only what your strange desires tell you to see, but most importantly stop lying, because you're not talking with goldfish here.
1, it's not Ian who's leaning anywhere, it's Tom's that is many times a disgrace of tech journalism these days and this article is reaching the cringe evels of The Verge's how to build a PC video. The only way it's not intentional click bait and sensationalism is heavy drug usage, where the editor hallucinates his CPU, the untold hero crying on his dying bed, begging him to tell his story to the world, no matter the forces that are trying to hide the truth.
2: I have never seen an article titled 'the new, cheating Z490 boards might kill your computer before you know it!'
3: even in this very topic, I see everyone agreeing on this practice being not cool, and trying to figure out how their mobo-CPU is handling this.
4: there's a big difference between being SOMETIMES AN EXTRA 100% OR MORE out of spec without OCing and what is discussed here. Even then I've never seen anyone saying that Z490 boards will kill your CPU @ stock, so what are you talking about??
5: OEMs pushing out of spec is fine? Again, WHO SAID THAT???

 

[lobz]

I have yet to decide on the long term cooling for my 3900x I guess I'll sport eco mode till then.

Just let him die a hero!!!

 

[TheGiant]

I posted it many times, never got an answer
My 3900X@default while handbraking consumes ~240W wall power, measured many times.
I expected less.
I expected something like pic below +20W , so around 215, because
since 142W is in HW info for package power and some losses of VRM and PSU
125W is the package power of the 10900K and everything else in the system is the same
we can see here the result of 10900K
all calculated with my NapKInMath(TM)

I asked several times where are the missing 25W, never got and aswer here
Never got an answer in the review articles
So is this it? can someone answer me please with the AnandtechForum(R) IfICanTakeYouSeriously(TM) approach ?
thanks
/off

 

[piokos]

As you can see, Intel has left wiggle room for the OEMs to pull as much power as they want for as long as they want, as long as they don't exceed 4.9 GHz all-core. That leads to boards that make a CPU burn ~230W in perpetuity while running Blender (for example). That is way out of spec. "Normal" behavior for a 10900k beyond 56s is to drop to 125W, which should be 4.3 GHz but winds up being different speeds depending on the motherboard (which is actually kind of weird).

Once again: PL/tau configurability is made *FOR OEMS*.

Intel is absolutely supposed to stop that behavior.

Give away one of the key advantages of the platform?
Everyone noticed you cheer AMD. No need to be that explicit.
And you're probably confident that I cheer Intel.
So let's call it a tie and focus on your lack of understanding of OEM market. It's so much more fun!

OEM prebuilt systems will not be specced with massive (and expensive) cooling solutions to aid-and-abet the blatant violation of power standards observed in motherboards designed for the DiY crowd.
OEMs will use custom motherboards that will likely follow Intel power specs because it will lower their BoM and increase their profits.

So first of all, I think you're old enough that someone should tell you this: not every PC comes with a top-of-the-range CPU for enthusiasts, like a 10900K.
And as we're at it: Santa Claus doesn't exist.
PL/tau mechanism works on all CPUs that have boost functionality and is absolutely crucial for SFF and laptops.

And about OEMs following default specs: just imagine the humongous air cooler they would need for an 10700 (PL2 = 224W, tau = 28s).
Are you imaging it?

OptiPlex 7780 27-Inch All-in-One PC with Dell Optimizer | Dell USA

You probably know a lot about building enthusiast DIY PCs, overclocking and so on. Why not focus on that?
Why tackle a topic you have no interest in?

 

[piokos]

2: I have never seen an article titled 'the new, cheating Z490 boards might kill your computer before you know it!'

And you have every right to write one if you believe that to be true (or you earn from clicks).

4: there's a big difference between being SOMETIMES AN EXTRA 100% OR MORE out of spec without OCing and what is discussed here. Even then I've never seen anyone saying that Z490 boards will kill your CPU @ stock, so what are you talking about??

No, it's not. At worst case scenario, what supposedly happens here will lead to sustained higher load.

And you know what?
Maybe after all Zen2 CPUs aren't getting hot because the node is dense. Maybe it's just that the reported power consumption is incorrect.

No, just kidding. Dense node is the main reason. This thing we're discussing would only make it a little worse.

 

[amrnuke]

Maybe after all Zen2 CPUs aren't getting hot because the node is dense. Maybe it's just that the reported power consumption is incorrect.

You do realize that a substantial number of reviewers tested actual power draw and not sensor-reported power draw, don't you?

 

[lobz]

And you have every right to write one if you believe that to be true (or you earn from clicks).

Again, WHO said anything even resembling that being illegal? Do you get excited off of provoking people?
As for the rest of your comment, thanks for not denying that you're intentionally lying and making false accusations towards the members of this forum. Not to mention putting those smiley faces at the end of those accusations, expressing some heavy superiority complex or whatever it is you're feeling.

Maybe I'm at fault and I should just report those posts instead of calling you out on them - but then what is a forum for?

 

[thesmokingman]

WTF? Why is this suddenly about pointing fingers?

Seriously, instead of considering the fact that we now have another tool to judge boards instead being ignorant on this often neglected area, you got one poster pointing fingers... That's some gall coming from a side that hid a 1kw chiller...

 

[Iron Woode]

WTF? Why is this suddenly about pointing fingers?

Seriously, instead of considering the fact that we now have another tool to judge boards instead being ignorant on this often neglected area, you got one poster pointing fingers... That's some gall coming from a side that hid a 1kw chiller...

why should that surprise you? The only thing that surprised me was how long it took.

 

[misuspita]

Electromigration: Why AMD Ryzen Current Boosting Won't Kill Your CPU

www.anandtech.com

Looks like Ian made an article about this.

 

[aigomorla]

current kills everything, even a solid state capacitor if you feed it high enough current.

Ive lost so many chips back in my active days from over voltage, except the Q6600 (Kentsfield) which i could not kill for the life of me even at 1.5V

But i lost count at how many Wolfdales, Yorktowns, and even Westmeres (Nehalems were fairly durable but not on the equal ground as a kentsfield.)

But Intel also had this problem, and no one blew it up.
It was known that if you set overclocking on AUTO, you were ASKING FOR IT, big time.

 

[tamz_msc]

Once again: PL/tau configurability is made *FOR OEMS*.

This is blatantly false. Just read Anandtech's interview with Guy Therien.

Ian Cutress: One of the things we’ve seen with the parts that we review is that we’re taking consumer or workstation level motherboards from the likes of ASUS, ASRock, and such, and they are implementing their own values for that PL2 limit and also the turbo window – they might be pushing these values up until the maximum they can go, such as a (maximum) limit of 999 W for 4096 seconds. From your opinion, does this distort how we do reviews because it necessarily means that they are running out of Intel defined spec?

Guy Therien: Even with those values, you're not running out of spec, I want to make very clear – you’re running in spec, but you are getting higher turbo duration.

We’re going to be very crisp in our definition of what the difference between in-spec and out-of-spec is. There is an overclocking 'bit'/flag on our processors. Any change that requires you to set that overclocking bit to enable overclocking is considered out-of-spec operation. So if the motherboard manufacturer leaves a processor with its regular turbo values, but states that the power limit is 999W, that does not require a change in the overclocking bit, so it is in-spec.

 

[RetroZombie]

Looks like Ian made an article about this.

And a very good one, 'refuting' all the Toms and Stils finds.

 

[Markfw]

THIS IS CRAP imo
"but states that the power limit is 999W, that does not require a change in the overclocking bit, so it is in-spec. "
So its "in-spec" to run 380 watt for example, indefinitely, on custom water @ 95c. That is what that says to me, and that crap, that should be out of spec.

 

[RetroZombie]

This is blatantly false. Just read Anandtech's interview with Guy Therien.

That's weird no OC but 999W is ok?

Let's imagine a very bad scenario where the cpu would be hit by some stupid 'power virus' that someone will make in the future, does that mean the cpu is left unprotected from the motherboard?

 

[tamz_msc]

THIS IS CRAP imo
"but states that the power limit is 999W, that does not require a change in the overclocking bit, so it is in-spec. "
So its "in-spec" to run 380 watt for example, indefinitely, on custom water @ 95c. That is what that says to me, and that crap, that should be out of spec.

Not according to Intel, as long as you're not exceeding the values specified in the turbo table.