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

thesmokingman

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
Saw this in my firefox news feed and was posting on it at TPU.

Some boards are reporting false power draws to Ryzen cpus.

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

Get your beta here.

https://www.fosshub.com/HWiNFO.html


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 power management co-processor, that is responsible for adjusting the operating parameters of the CPU and ensuring, that neither the CPU SKU, platform or infrastructure specific limits are being violated.

The weakness of this method is, that the telemetry essentially uses an undefined scale for the current (and hence power) measurements. This means that the motherboard VRM controller will send an integer between 0 - 255 to the CPU, and based the reference value known by the co-processor firmwares, this integer is converted to a figure, that represents a physical current drawn by the CPU. Based on the accurately known current flow and the voltage, it is possible to calculate to CPU power draw in Watts (V * I).
Click to expand...

Get the latest beta and fire up Cinebench and test your board out. Don't forget whatever power draw limits you raised already and undo them.

HWiNFO will display "Power Reporting Deviation" metric under the CPUs enhanced sensors. The displayed figure is a percentage, with 100.0% being the completely unbiased baseline. When the motherboard manufacturer has both properly calibrated and declared the reference value, the reported figure should be pretty close to 100% under a stable, near-full-load scenario. A ballpark for a threshold, where the readings become suspicious is around ±5%. So, if you see an average value that is significantly lower than ~ 95% there is most likely intentional biasing going on. Obviously, the figure can be greater than 100%, but for the obvious reasons it rarely is
Click to expand...

Also the deviation reporting only applies to full load.

As stated before, this metric is only valid during a relatively stable near-full-load condition
Click to expand...
 
Last edited:
I have been experimenting a little with this since I read the post on HWiNFO's forum.

On my system with my current settings, where I run with a -0,1V offset to the core voltage, HWiNFO64 reports a Power Report Deviation (PRD) of about 75%. Which seems rather bad (lower percentage is supposed to mean the CPU is using more power than its specified to do).

Going into the BIOS and setting core voltage to auto, makes PRD report about a 95% deviation which should be a better result. Though in this state the CPU is clearly using a lot more power and running hotter.

Both of these was running CB R20.

So if you are playing around with core voltage offsets, the numbers don't seem to add up?

Anybody else have any input?
 
X570 Aorus Ultra F12e bios with vcore as "normal". (3950x)

RandomX mining load:
New Bitmap Image.png

Edit: If folks want, i can take a pic of the applicable bios settings.

Edit2: AT just posted this.
 
Last edited:
Anandtech's take:
To answer the big question right off the bat then, no, your CPU is not at risk. For regular users with enough cooling running at stock frequency, there is no issue to any degree that will matter within the expected lifetime of the product.
Click to expand...
 
Why does this seem like a Intel paid sort of thing, whether it is or is not? The timing seems entirely suspect. Basically, this was reported shortly after Intel said Benchmarks do not matter any longer, or something to that effect.
 
ManofGod said:
Why does this seem like a Intel paid sort of thing, whether it is or is not? The timing seems entirely suspect. Basically, this was reported shortly after Intel said Benchmarks do not matter any longer, or something to that effect.
Click to expand...

it's not. but toms hardware's editors should be slapped for allowing a article title like the one they used and taking what Stilt said out of context to push fear mongering click bait.

either way this just further proves that both AMD and intel need to put their foot down and put the motherboard manufactures back in their place.. they've gotten away with far to much b.s. on the intel side and even with how strict AMD is compared to intel they've found a gray area to get away with the same crap.

the reality is what's happening isn't going to make a massive difference and isn't going to kill your cpu over night but what it does is give users more insight into the games board manufactures are playing to make their boards look better than their competitors by cheating.

also recommend watching GN's latest video, goes into a lot of detail about it.
 
My Asus Crosshair VIII Hero keeps it right around 100% running CB20. Starts around 96% and moves up as the benchmark runs, getting right at 100%
 
Should be taken with a massive grain of salt. This is a subtle form of fear mongering and is misleading noobs or the uninformed. Tom's Hardware used to be great, but now it's the Kardashians of the tech world.
 
kirbyrj said:
Anandtech's take:
Click to expand...

I swear ian missed the forest from the trees. This is a serious issue in the context that if you didn't know about it and think your cpu is doing what it's supposed to be doing. And then you go to overclock like the majority of users who don't read or research anbd they end up jacking PBO up or running random manual voltages. That's a recipe for issues. And Ian's also wrong about degradation as it can happen at mild voltages under typical cooling. Stilt's shown this and he's been hunting for that missing power envelope for a year now.

Threads like this lead ppl down a bad ending and it only gets worse if the board at the outset is lying to the cpu.

https://www.overclock.net/forum/13-amd-general/1741052-edc-1-pbo-turbo-boost.html
 
thesmokingman said:
I swear ian missed the forest from the trees. This is a serious issue in the context that if you didn't know about it and think your cpu is doing what it's supposed to be doing. And then you go to overclock like the majority of users who don't read or research anbd they end up jacking PBO up or running random manual voltages. That's a recipe for issues. And Ian's also wrong about degradation as it can happen at mild voltages under typical cooling. Stilt's shown this and he's been hunting for that missing power envelope for a year now.

Threads like this lead ppl down a bad ending and it only gets worse if the board at the outset is lying to the cpu.

https://www.overclock.net/forum/13-amd-general/1741052-edc-1-pbo-turbo-boost.html
Click to expand...

I disagree. If you're setting a voltage manually, then you are overriding the PBO, etc. You might actually be running less voltage if you set something in bios vs. allowing PBO to do its thing.

I don't think it's something to be worried about. At least not in the average lifespan of the CPUs. I haven't seen anyone complaining about degradation with 1st gen Ryzens and they've been out for 4 years now. So it lasts 8 years instead of 15. It's outside the normal lifespan of the CPU anyway.
 
Well, I downloaded the beta and I'm reporting 115% power deviation on my Asus board and 3600x (RandomX mining), so evidently, my understanding is that I'm actually over-reporting and I'm actually drawing less power?

I did a more standard "stress" test with AVX2 and it showed 103%.

So my understanding of this number based on the Tech Jesus video and looking back in the actual thread over at the HWInfo forums is that the reported power draw is 103% of the actual power draw meaning that I'm actually drawing some 3% less than reported (which is a confusing way to phrase what they are after). When they showed an early ASRock bios (in the Tech Jesus video) and the number was somewhere in the 80% range, it's saying that the reported power draw was only 80% of the actual power draw meaning that it's drawing some 25% more than it should.
 
Last edited:
kirbyrj said:
Well, I downloaded the beta and I'm reporting 115% power deviation on my Asus board and 3600x (RandomX mining), so evidently, my understanding is that I'm actually over-reporting and I'm actually drawing less power?

I did a more standard "stress" test with AVX2 and it showed 103%.

So my understanding of this number based on the Tech Jesus video and looking back in the actual thread over at the HWInfo forums is that the reported power draw is 103% of the actual power draw meaning that I'm actually drawing some 3% less than reported (which is a confusing way to phrase what they are after). When they showed an early ASRock bios (in the Tech Jesus video) and the number was somewhere in the 80% range, it's saying that the reported power draw was only 80% of the actual power draw meaning that it's drawing some 25% more than it should.
Click to expand...

103% just means the bios is reporting a slightly higher voltage than the cpu VID voltage which means you're potentially losing some boost performance since the cpu thinks it's using more power than it really is but doubt it makes a real world difference.
 
Late to this party. I did the Cinebench test and it is reporting 91% during the test. Sitting idle hwinfo reports 50%. Should I worry about this? Can it be fixed? MOBO is an asrock x570 taichi.
 
Smoked Brisket said:
Late to this party. I did the Cinebench test and it is reporting 91% during the test. Sitting idle hwinfo reports 50%. Should I worry about this? Can it be fixed? MOBO is an asrock x570 taichi.
Click to expand...

The deviation test is only valid during an allcore load.
 
Well, my Crosshair VI Extreme with my 3900x is showing me 101% under Cinebench R20. It errs a little on the conservative side, it seems...
 
So My temps have been going up to 95 under load while benching with cinebench r20 or prime95. They get there in under a minute. My idle is around 40, and under load my power deviation is showing 93% as you can see in the screenshot. Any way to fix this or make my chip run actually at stock? I'm using the stock cooler and I really don't want to have to upgrade at the moment. The second screen is idle after benching, but you can see the maxed out cores etc.

Clipboard01.jpg

Clipboard02.jpg
 
My power deviation was so bad (nearly 50%) that I set my voltage to 1.0 and my offset to -.1v and everything is stable and I lost around 15'c all around. So I'd also suggest manually lowering it.
 
Hmm. thanks for the advice. I'm a little nervous since I've never tweaked voltage before. What is offset? I have the 1.10 BIOS on my asrock phantom gaming 4 b550 a/c. In the manual it seems to have voltage / vcore op;tions in multiple sections.

I thought i read in that long post by the AMD guy on reddit not to undervolt your processor since it will disable it being able to 'sleep' or something.

This is why this sucks, I run my stuff at stock so I don't have to learn how to tweak this junk :p Won't buy ASRock again unless they stop this. Even if it isn't dangerous it is underhanded.
 
roz1281 said:
Hmm. thanks for the advice. I'm a little nervous since I've never tweaked voltage before. What is offset? I have the 1.10 BIOS on my asrock phantom gaming 4 b550 a/c. In the manual it seems to have voltage / vcore op;tions in multiple sections.

I thought i read in that long post by the AMD guy on reddit not to undervolt your processor since it will disable it being able to 'sleep' or something.

This is why this sucks, I run my stuff at stock so I don't have to learn how to tweak this junk :p Won't buy ASRock again unless they stop this. Even if it isn't dangerous it is underhanded.
Click to expand...

From what I have seen, ASRock is particularly egregious. On my ASUS board, it was pretty much spot on at 100%.
 
You must log in or register to reply here.
Back
Top