Power manipulation by some motherboards

[mackerel]

https://www.hwinfo.com/forum/thread...er-reporting-deviation-metric-in-hwinfo.6456/

See above link for the details. The claim is essentially that power calculation in Ryzen depends on a value fed to it by the mobo. If that value is adjusted you can cause the CPU to use more or less power than AMD intended. hwinfo64 since beta 6.27 has implemented some way to detect and report if this is happening.

I've now done some initial testing on my two Ryzen systems.

3700X on Asus Prime X370-Pro bios 5220. Prime95 8x128k FFT stabilised around 91%. Cinebench R20 varied between 92-94%.

3600 on Asrock B450 Gaming-ITX/ac bios P3.70. Prime95 6x128k FFT 88%. Cinebench R20 started around 91% and crept up to 94% during the run.

100% is ideal. A value less than 100% means the current/power is under-presented to the CPU so it could use more as it thinks it has more headroom. These two boards perform quite similarly. While a bit lower reported value than ideal, it isn't outrageously low so I wonder if there are limitations in the measurement method, and given some tolerance it isn't that bad. I'd be interested if anyone has a mobo reporting a much lower value.

 

[EarthDog]

Saw this.. interesting. Just feels like what ready happens with intel on each board....most dont follow intel spec in the first place.

Edit: also, this......

https://twitter.com/x/status/1270245205090934784

Feels much ado about nothing?

 

[mackerel]

The situation on Intel is very different. Intel do not enforce the power limit, and encourage system builders to set it appropriate for the whole system design (power delivery, cooling). It's more like a serving suggestion. You are welcome to adjust the power limit as much as you like and it is not warranty affecting. The big name box shifters will tend to stick to Intel recommendations (not a specification) because they cheap out on the cooling, and likely the power delivery also. Enthusiast builds with more overkill designs can take the higher power, so it is more appropriate to use it.

AMD's position is they do have a PPT value which is warranty affecting if you choose to bypass it. In this case, with the mobo manipulation, it is more of a mess. The CPU thinks it is within limit at stock settings, but actually isn't. What happens then?

That leads to the next point, I wonder how many motherboards are significantly affected by this. My two, as I said, are ball park of 10% under. I don't know how reliable the detection mechanism is, or what tolerance is considered reasonable for that value in the first place. It doesn't seem excessive.

 

[EarthDog]

End result is similar though the method is quite different, indeed. My point wasn't that the method was the same. They certainly are not.

Make sure you read my link to Ian's Twitter (6n specific says what I'm saying about intel boards ppwer manipulation)

 

[mackerel]

Saw it earlier, even replied to it.

His beef there is more about how Tom's is scaremongering. I'm more interested in the impact this trick might have on actual behaviour and reporting.

 

[EarthDog]

No idea on how many boards are affected. I'm reviewing b550 now. So ill keep an eye out. Outside of calling out toms, the information he provided is quite helpful in explaining things.

For those who may be concerned about the sensationalist Tom's article....

or AMD users, it's a nothingburger. If you don't already do some form of manual clock tweaking, it's some extra frequency for the sake of some extra thermals.

Much ado about nothing.

 

[mackerel]

While I didn't state it, I'm not worried about this even if I were to have a system significantly affected by it, which I'm not sure I do. Power limit relaxation is relatively safe as other limiters (particularly thermal) can take over if needed. I'm assuming the CPU will still be on its voltage/frequency curve so it is fine. Maybe less efficient.

However I'd still like to understand this more than we currently have details on, as it can impact testing and comparisons, especially across different CPUs. I'm sure many will have the opinion already that CPU reported power is of limited accuracy. I don't have the capability to make better measurements than that, so it is the best I have. I'm now also wondering if there is additional deviation depending on loading.

 

[Woomack]

I highly doubt there is any "gaming" or "overclocking" motherboard which is even close to mentioned 100% value. Pretty much every manufacturer is using something to adjust power or frequency which is tied to some other settings, and barely anyone sticks to AMD or Intel specs. It's not even because they don't want to. It's more like products are not always well-designed to meet power delivery requirements. On the other hand, barely anyone cares about that as gaming/overclocking products have to perform better than standard so often have "auto-OC" or "bclk boost" or forcing max boost clock higher than standard or anything like that. On the other hand who cares about that in a standard home/office PC? Cheaper motherboards are not even reviewed. So we can comment on the way how manufacturers are designing their products but it doesn't really matter on the mass market. No matter how bad a product can be, marketing will make it sell well.

 

[mackerel]

Is it just me then? I'm interested in the mechanism at play here, and the implications resulting from it. I agree, to the typical user it makes little to no difference. This only just surfaced, and Zen 2 has been out for almost a year, and it may have been happening before that anyway.

 

[EarthDog]

I'm curious, sure, but to what end when it doesn't really seem to matter?

I'd be more worried about Intel honestly... considering how they work TDP with PLs and then boards take liberties on top of that... more so for cooling though, than processor life.

Anyway, I'll keep an eye out during B550 testing and see if there is anything wonky with any of the boards I'm going through this week.

 

[mackerel]

The Intel approach is open and encouraged, within the limits of people's ongoing inability to understand what TDP actually means. This AMD mobo power thing is not obvious. It could have implications on past testing, especially when comparing across Intel and AMD power usage and performance.

 

[EarthDog]

The Intel approach is open and encouraged,

Is it? You've said that twice now, but I don't recall hearing about board partners being encouraged by intel to do such things. That said, they are not required to stick to Intel parms, but encouraged to go past, I don't think that motivation came from intel so much as it is marketing so their boards are 'faster' than others.

What implications do you believe this may have outside of testing/performance?




EDIT:

At some point, you’re left asking who’s really at fault. Maybe Intel isn’t being strict enough, as most of the turbo policies are “guidance,” not enforced specification or requirements. Maybe the motherboard manufacturers are stepping out of bounds. We have heard rumor that Intel will crack-down harder on board manufacturers for violating these specs this generation, and we hope that’s true...

...By and large, despite motherboard manufacturers’ propensity to ignore specifications,

https://www.gamersnexus.net/guides/3389-intel-tdp-investigation-9900k-violating-turbo-duration-z390


The way Steve wrote that article, it doesn't seem like there is encouragement from Intel at all... this is all board partners (sans SuperMicro who sticks to it, lol). If they are rumored to begin cracking down (which, they didn't seem to on Z490 either), it sounds like Intel isn't doing any encouraging to break their 'guidance'... unless silence is encouragement?

 

[mackerel]

I'll have to dig it out later. I guess my primary reference is an interview Ian Cutress did with the guy at Intel who came up with the scheme we have today, somewhere on Anandtech. Maybe I'm over-stating it with the word encouraged, but it is explicitly allowed by Intel.

 

[EarthDog]

Allowed, sure (they are guidelines that don't seem to be enforced)! Encouraged...I'll have to see/hear it to believe that.

Anyway, what implications outside of performance do you believe will come of this? I'm wondering which rabbit hole you want to go down.

 

[mackerel]

Both PL2 and Tau are configurable by the OEM and ODM, and are ‘in spec’. So you have the ability, of course, if the power delivery supports a higher power limit, and also if the system has the thermal capability (because it makes no sense if you’re just going to thermally throttle), to adjust these values. So motherboard manufacturers and ODMs are investing in their power delivery and thermal solutions to allow them to maximize performance or get a certain about of turbo duration without throttling – to the maximum extent possible. As a result of this you can invest different amounts of money into the power delivery, the thermal solution, the thinness of the system and so on – it’s the ability to design something that is differentiated for the audience, both in terms of form factor and performance.

https://www.anandtech.com/show/1458...ng-an-interview-with-intel-fellow-guy-therien

Above is the interview I was thinking of. I'm tempted to re-read the whole thing, but I think the above section reaffirms that you can set PL2 to whatever you like and not go out of spec. It also states that how you build this into the system can be used to differentiate vs others. The word encouraged isn't used, but certainly it is well within expectations of what companies are expected to do.

I don't expect this thing to change anything at a practical level for anyone currently owning an AMD system. It'll work today just as it did yesterday before this was pushed into the open. Maybe mobo manufactures will change their policy in newer bios, or at least newer mobos.

My interest in this is primarily from a technical level. Are the numbers reported for CPU power consumption affected by this? If so, it should be taken into consideration. Especially when doing cross platform comparisons. I guess if people test "at the wall" power, that would not be affected by this, but it is more convenient to use software approaches where available. I'm actually a bit surprised by the responses I'm seeing in this thread. I had thought of this forum has having more technically knowledgeable types, so thought there would be more engagement in understanding this together. I don't exactly have a lot of data points, with only two systems myself. It may well be this is not such a common phenomenon at all.

 

[EarthDog]

Weird... multiple 'specs' are "in-spec"?!

Noted on the implications...ty.

The problem with software is that it can be inaccurate. This is why many test with a Clamp on meter and use it or at least as a sanity check against software.

I don't have a lot of time on my hands... but happy to help where I can.

 

[mackerel]

I'm watching on another forum where AMD owners are posting their results. Only one report has been well below 100%, and I'm wondering if that is related to them having a Zen CPU. I think that's about all I'm motivated to do for now. Might have blown over by tomorrow anyway.

That reminds me, I did get a NZXT PSU which has power measurements separated on different cables, so the ATX connector is separate from EPS, is separate from PCIe... wonder what system that is in?

 

[Zerileous]

Interesting stuff. Using CBR20 I get a steady 80% for the reporting deviation value. This is while running full stock and the CPU is reported as mostly sitting at 100% PPT and occasionally hits 100% EDC.

I really haven't taken the time to dig into this CPU yet. I've tried effectively removing all power limits with PBO and noticed about 75mhz all core difference. Really was not worth the temperatures though, as my heavy load (F@H/CBR20/prime95) temp went from 80C to 90C. I was also able to manually do a 4.3 GHz all core with just 1.275v, bench stable. According to reddit at least at least 7nm is much less voltage / current tolerant then 12nm, so I haven't really decided how far I'm going to take it. 4.4GHz at 1.3v was a no-go, so it seems like 4.3GHz is near the voltage wall for this chip anyway. I've been considering trying to use PBO to make it behave more like a 3800x rather than just remove all limits, but this already seems to be the way Asrock is handling it lol. I'll still try it at some point, but I have to reach a point where the curiosity outweighs other activities.

 

[RJARRRPCGP]

It's more like products are not always well-designed to meet power delivery requirements.

This reminds me of the terrible mess of the AM3+ era! Almost only two motherboard models can meet minimum required specification for stock frequencies.
Too many AMD motherboards in the AM3+ and FM2/FM2+ era had FET issues. And that's why the "motherboard police" existed.

I remember that period. For AM3+, I got scared of using anything other than a Crosshair or a Sabertooth 990FX! Thus, mostly down to only two motherboard models with AM3+!
And even the Phenom II era was trouble, too, because I have an early-2010s' AM3 motherboard with no MOSFET heatsink, but there's clearly mount holes for one!

 

[Woomack]

AM3 and FX remind me of Gigabyte and their UD5 mobo. Instead of fixing issues, they were saying that the mobo is fine but every 1-2 months we could see a new PCB revision and corrected power section