Updated - Dr. Ian Cutress weighs in - HWInfo's Power Reporting Deviation Sensor Reveals CPU Lifespan-Reducing AMD Motherboard Enhancements

[jagdtigger]

1 hour ago, porina said:

That's probably unrelated. The high voltage is needed to support the higher boost clocks during lightly threaded workloads. If all cores are in use, the clocks are lower to control total power, and thus a lower voltage is seen.

It was getting it even at idle, i could always hear the AIO fans,  lowered it to factory manually with set multiplier, now i cant even hear that its on when idling/light load.

Edited June 9, 2020 by jagdtigger

[porina]

Next test system: 3700X on Asus Prime X370-Pro bios 5220. Prime95 8x128k FFT stabilised around 91%. Cinebench R20 varied between 92-94%, but unlike the earlier system this varied through the run, and wasn't a gradual rise.

 

The thought occurs. If the mobo is misrepresenting a value to deceive the CPU, how does hwinfo64 then detect that has happened? It would need some separate indicator to show that has happened. 

 

Anyway, on my two systems, it seems pretty consistent, with indicated values in the ball park of 90%. Perhaps a bit further from 100% than would be ideal, but not so far out to call it outrageous cheating. Especially given it seems to vary with load, I'm inclined to give the two mobos I have the benefit of doubt. The value may be close enough to ideal, given the uncertainty I have right now on how this is calculated.

 

It will be interesting to see if anyone has a system with a value so far out that is more likely to be intentional cheating.

 

15 minutes ago, jagdtigger said:

It was getting it even at idle, i could always hear the AIO fans,  lowered it to factory manually with set multiplier and now i cant even hear that its on when idling/light load.

Even when you are doing nothing, the system will often be doing other things. Even running the monitoring tool is giving the CPU work to do. The high voltage is still normal for stock. If you changed either voltage or clock (multiplier), you're over/underclocking it.

[NCC1701JR]

My system shows a 60-70% deviation with R5 1600 and Asus prime B350 board.

But temps stay at 65-70 C during repeated Cinebench R20

[D13H4RD]

Ian Cutress just released this piece

 

https://www.anandtech.com/show/15839/electromigration-amd-ryzen-current-boosting-wont-kill-your-cpu

 

TL;DR - Unless your CPU is really running hot, it's very unlikely to meet an early death over its useful life. But would also be handy to have a toggled option to disable it.

[matchrocket]

I played around with this today on my ASRock Tachi x470  +3900x with the latest bios and latest AMD Chip-set drivers

 

Full load I I get ~80% consistently with a -0.050 voltage offset.  I set it to default which runs ~1.5v in BIOS and 1.4+ hanging around the desktop with the latest balanced Ryzen power plan and my offset "reduced" to 88% or so in a 300s C20 run. 

 

Seems neg voltage offsets increase the reporting dependency.  They only thing I gain from the offset is lower temps under idle/load.  I see no impact in R20 scores.   BIOS Settings are "default" expect fan curves and XMP profile for RAM  and neg voltage offset.  PBO is enabled as it was out of the box. 

 

Very interesting.  Not sure what it means in the long run, if anything.  Thermals are well under control (<80c) and full load voltages are typically ~1.25 - 1.3 @ 4.1 / 4.05ghz doing extended handbrake runs or if I am doing stress tests with a neg 0.050 offset.  Voltage offsets just move the dependency around. 

 

P.S.  I don't recommend this board.  Its twitchy and likes to clear CMOS randomly.   Also sometimes I get very strange PPT/TDP/EDC max vales reported  E.G. 650a EDC and TDP of 400w.  Have to shutdown to get it to report right.  Restart doesn't help.   really considering getting a ASUS x570 when they come back in stock and prices come back down to earth. 

[Doobeedoo]

So I've test on X570 Aorus Elite with 3800X with CB full load and it's like ~100% so seems fine.

[NumLock21]

Does this only apply to AMD systems. I tried it on my Intel machine and I don't see "Power Reporting Deviation".

[DildorTheDecent]

3950X (PBO values 300/230/230) + Crosshair VIII Impact reports 96%-97% under Cinebeans. Minimum of 95.3%.

[porina]

1 hour ago, NumLock21 said:

Does this only apply to AMD systems. I tried it on my Intel machine and I don't see "Power Reporting Deviation".

It is only applicable to AMD systems, possibly Ryzen only although I'm not sure of the exact scope. Certainly later Ryzen anyway.

 

42 minutes ago, Benji said:

Well, I don't really know what to make of it and what this value is actually saying anyway.

Asus ROG Strix B450-F Gaming

Ryzen 7 1700X

DDR4-3000MHz

During idle it fluctuates between ~54-55% and ~120%. Under CineBench R20 it stays pinned between 55.6% and 55.8% with its clockspeed pinned at the maximum all-core turbo of 3.5 GHz. Everything stock except the enabled D.O.C.P. profile.

Ignore idle results. The values under Cinebench seem low, but I wonder if because you have 1st gen Ryzen, they really didn't have advanced a boost mechanism then and could that lead to a result that isn't meaningful.

[Tristerin]

I will have to read the article and understand the values better so I can bench and see - but with the voltages I apply, Id have killed my 1700 long ago if the board was giving it more juice over the last year.

[DrMacintosh]

Debunked

[D13H4RD]

58 minutes ago, DildorTheDecent said:

3950X (PBO values 300/230/230) + Crosshair VIII Impact reports 96%-97% under Cinebeans. Minimum of 95.3%.

Sounds like most ASUS X570 boards are adhering closely to AMD's specifications. 

 

The Prime, TUF Gaming Plus and the Crosshair VIII Impact all seem to be around the 100% mark with only a max deviation of -6% in one case based on what I've managed to gather. 

[Mihle]

My Ryzen 3600 Is reaching 104% at the top in CB R20. Average 107%
Most here are sub 100% it seems but I am above.
HWInfo is reporting a SVI2 TFN average of 1,35v if I let it run long enough, at the start its above that.
(AMD OC in auto, whatever it was called)

 

MSI B450m Mortar Max btw.

[BabaGanuche]

My 3900x on the Aorus Master is averaging almost exactly 100% under R20 and 104% under p95 small fft

[StDragon]

93.5% with an R7 3800x on an ASRock X570 Taichi

 

BTW, I think we should dump these values into some google spreadsheet online.

 

MB Make/Model, BIOS rev, CPU. Something like that for the fields.

[leadeater]

 

[D13H4RD]

2 minutes ago, leadeater said:

 

Watched this too. Basically, for the end user, it’s nothing to worry about. Your CPU won’t fry itself and it likely won’t degrade noticeably throughout its useful life. You may want better cooling but it’s not like a huge deal.

 

If anything, seems like they caught AsRock being a naughty boi with how they did their review BIOS. Tsk tsk tsk...

[leadeater]

10 minutes ago, D13H4RD said:

Watched this too. Basically, for the end user, it’s nothing to worry about. Your CPU won’t fry itself and it likely won’t degrade noticeably throughout its useful life. You may want better cooling but it’s not like a huge deal.

That and it's just reporting, the CPU may not actually draw any more power even when intentionally under reported. That's the thing about CPU power draw, remove all limits and it won't actually draw infinite power 

[D13H4RD]

26 minutes ago, leadeater said:

That and it's just reporting, the CPU may not actually draw any more power even when intentionally under reported. That's the thing about CPU power draw, remove all limits and it won't actually draw infinite power 

I think it's hard-coded somewhere where the CPU has a hardline limit?

[mr moose]

So storm in a teacup?  that's a relief.  Though it did prompt me to get of my ass and finally push my 3600. 

[leadeater]

5 minutes ago, D13H4RD said:

I think it's hard-coded somewhere where the CPU has a hardline limit?

Only the PPT etc values, it's more that components only draw as much current as they need. You don't push power in to something it demands it so it'll only ever use as much as it needs to do the task it is doing. If you try and run the CPU above stability it'll just crash.

 

CPU is the demander and the VRM is the supplier, unless the CPU wants it the VRM won't give it no matter how you fiddle with reported figures.

[mr moose]

5 minutes ago, leadeater said:

Only the PPT etc values, it's more that components only draw as much current as they need. You don't push power in to something it demands it so it'll only ever use as much as it needs to do the task it is doing. If you try and run the CPU above stability it'll just crash.

 

CPU is the demander and the VRM is the supplier, unless the CPU wants it the VRM won't give it no matter how you fiddle with reported figures.

Does this mean the CPU dictates the voltage supply exclusive of the motherboard? 

[porina]

CPUs will have a built in "stock" frequency/voltage curve. It will go up and down that curve according to the operating conditions and load. For performance, it will try to go up as far as it can, up to the power or current limits. The curve doesn't go forever either, there will be an upper frequency or voltage. Increasing the power limit only helps if the power limit is being hit.

[porina]

2 minutes ago, mr moose said:

Does this mean the CPU dictates the voltage supply exclusive of the motherboard? 

To the best of my understanding, the CPU will request the voltage it wants at a given time. It is up to the mobo to provide the correct voltage. This is made more complicated by losses between the output of the VRMs and it reaching the CPU, and we get into load lines.

[leadeater]

19 minutes ago, porina said:

To the best of my understanding, the CPU will request the voltage it wants at a given time. It is up to the mobo to provide the correct voltage. This is made more complicated by losses between the output of the VRMs and it reaching the CPU, and we get into load lines.

But you can configure these voltages, when it comes to voltage it's like a two way agreement but the motherboard does have the overriding power in regards to voltage scale since you can manually configure a voltage that is too high and kill the CPU and it won't prevent that. CPU will request a voltage within the configured and agreed values as it needs during operation but this is basically scaled to the agreed range or rather the vcore you set which is the maximum and everything is scaled off that.