AMD Ryzen 7000X3D Power Consumption Spiking Beyond 100 W in Idle Mode

[AsRock]

How did they test this? I haven't seen any such spikes on my 7700X with HWinfo. Not that a momentary spike of 130 W (which is well within TDP limit) should burn your CPU anyway.

You need some thing like a oscilloscope to see them.

 

[MarsM4N]

What I find more interesting is the breadcrumbs in the comments section at igorsLAB. Apparently there is/was a S3 zero temp Hibernation bug in the BIOS, also noted in a BIOS update from MSI as "Fix CPU temp shows 0 degree after resume from S3". igorsLAB reported on this already on the 24th April, he even could replicate it:

Quote: "In this context it is surely also interesting, and this information comes from Igor, that e.g. MSI had hastily added a beta BIOS on 04/14/2023, which is supposed to fix problems with the temperature monitoring, which can always occur when the PC “wakes up” from the so-called S3 sleep mode. It also happened to him that the temperature feedbacks for the CPU all remained without a value, i.e. returned “zero”. The consequence was that the fans of the installed AiO (the mainboard regulates this via the CPU temperature) stopped.

Normally, when the CPU overheats, the self-protection mechanism kicks in, which takes a number of measures to prevent the CPU from burning out and being destroyed. Igor only noticed the problem because the CPU became extremely slow and the screen almost froze. However, we can only speculate whether these sporadic dropouts are related to the damage mentioned on Reddit. However, one can only advise every user to flash an up-to-date BIOS that corrects this error. It is absolutely not the time to spread panic, but the circumstances have to be investigated and evaluated first."

https://www.igorslab.de/en/mehrere-nutzer-haben-gemeldet-dass-die-ryzen-7000x3d-chips-abkratzen/
This could explain why only a few CPU's went up in ashes. You need to use hibernation, it needs to wake up from it while unattended & keep cooking itself for some time. And you need to have a CPU cooler that is fan controlled by the BIOS. It does also explain why temp control didn't kick in. So if these where the issues we can check the checkbox for /fixed. What it doesn't explain are the idle power spikes, so that's something that still needs to be investigated.

Not just the peaks. The average power consumption readings are surprisingly high.

Igor/Aris:
7950X3D Idle AVG - 50.531 W
7950X3D Idle AVG, CO -15 - 45.952 W

W1zzard (from review):
7950X3D 1T - 37W
7950X3D 1T PBO+UV - 39W

Why are the idle readings higher than when loaded by an MP3 encoder?

It's even more out of whack for the 7900x. Aris reports idle is 63.8W, W1zzard reports 1T is 36-42W (depending on the review). Unfortunately W1zzard no longer records idle power for his reviews so I can't compare more directly.

OT: I would be pissed if I upgraded to a 7900, nominally 65W TDP, to later learn that it idled at 47W.

It's very common that power consumption varies hugely depending on mainboard. All these supa dupa overclocker boards take a fair share extra watts compared to boards with less "features".

Looks like the affected CPU's were rushed to the market for whatever reason and now these are the results, or Microsoft just prefers Intel?
Does the same thing happen on Linux (near-idle power spikes)?

That's a good point. They should re-run the tests with Linux & Win10. Just to rule out that it's not one of the many Win11 bugs.

I couldn't find a info about what BIOS version they used, but I guess they used the latest available. Might be worth to re-test with a updated BIOS.

 

[Dr. Dro]

Likely with socket level voltage probes. The reviewer in question was TPUs PSU reviewer, and does not mess around.

Agreed. These oscillations are too fast for hardware sensors to pick up. Since at least Zen 3, Ryzen CPUs can change their power state in milliseconds, which greatly reduces sensor accuracy. Still, those seem to be quite extreme, but if within spec - and looks like they are, it's odd behavior at best.

A lot of kinks to iron out here, it seems... and whoever greenlit the last AGESA release needs to be fired.

 

[phubar]

You do realize that 200ms is laughably slow compared to how they're testing, right? lol

If the spikes are so short that 200ms measuring time is "laughably slow" and are only during idle (which seems to be the case as far as I can tell going by the story) then no one should care about them in the least.

They will have litterally no effect and you will never notice them.

You're getting worked up into a lather over what is pretty much a minor bug somewhere.

One that should've been fixed before launch of course. But that is true of all bugs.

 

[Mussels]

Wasnt a big part of this problem that the older BIOSes were allowing EXPO to set high SoC voltages, and boards were sending too-high voltages on top of that, with some BIOSes?

LLC going too high, with VRM's too slow to smooth out the transitions between low and high loads, much like the SoC related issues that plagued AM4 with high RAM speeds (but in reverse, where the voltages dropped and we had USB/PCI-E disconnects)

If the spikes are so short that 200ms measuring time is "laughably slow"

200ms is very slow for this sort of thing when ryzen CPU's update their power states every 1ms, since Zen2


Whatever these two agree on, carries series weight as to what's happening
Igor Wallossek / Aristeidis Bitziopoulos

Even if i can't pronounce their names.


Edit: Isn't this something that the PBO values like EDC would prevent? It's there to prevent high amperage spikes at low loads (with TDC for sustained load and PPT for max wattages)

Does this issue only exist with PBO enabled and maximum values set? (even if they default to those values)

 

[Arkz]

Zens power consumption when idle sucks regardless. From my experience with Zen 2 and 3, chips drawing around 30w when barely doing anything is pretty unacceptable.

 

[phubar]

200ms is very slow for this sort of thing when ryzen CPU's update their power states every 1ms, since Zen2

If its happening for 1ms every now and again at idle I'd say that only reinforces my earlier comment about not caring* about it because you litterally won't notice it.

Now if its happening during some sort've load like gaming or whatever (which would probably cause micro stuttering) that is a different story but that doesn't seem to be the case.

*not caring about it in practical sense. Its a bug it should get fixed. But there are already dozens of bugs all through out any given system or OS even with the latest updates and hardware. Freaking out about them all the time is pointless and exhausting if they're effectively unnoticeable, don't effect performance, stability, lifespan, or security.

 

[Mussels]

Not just the peaks. The average power consumption readings are surprisingly high.

Igor/Aris:
7950X3D Idle AVG - 50.531 W
7950X3D Idle AVG, CO -15 - 45.952 W

W1zzard (from review):
7950X3D 1T - 37W
7950X3D 1T PBO+UV - 39W

Why are the idle readings higher than when loaded by an MP3 encoder?

It's even more out of whack for the 7900x. Aris reports idle is 63.8W, W1zzard reports 1T is 36-42W (depending on the review). Unfortunately W1zzard no longer records idle power for his reviews so I can't compare more directly.

OT: I would be pissed if I upgraded to a 7900, nominally 65W TDP, to later learn that it idled at 47W.

RAM Speed and their required DRAM and SoC voltages affects those wattages a lot - I'm using about 15W more on my setup going to 64GB at 3800 from the SoC (and then undervolt the CPU, to recover greater than that)

The observer effect was in regards to hwinfo

HWinfo has an "AMD snapshot polling" Feature to fix that problem, and we can hope that these two experts used it

Zens power consumption when idle sucks regardless. From my experience with Zen 2 and 3, chips drawing around 30w when barely doing anything is pretty unacceptable.

Mines not like that, roughly 15W CPU and 15W SoC (because of the RAM OC)

With edge open (15 tabs) discord, steam, afterburner, etc.

And my 3700x

Looking at external sources since TPU reviews don't include CPU idle wattages - outside of HEDT setups, they all seem pretty even? Unsure what your complaint is there

AMD include a few more things in the CPU than the motherboard (Like a SATA controller) that move some power consumption off the motherboard and to the CPU, so the total system wattages need to be considered here, not just CPU alone

I know if you ran an all-core OC the "intel way" you could disable power saving and waste juice at idle, per-CCX overclocking was the correct method there that not everyone used
Even the windows power plan chosen can screw with that, especially on Zen2 as you needed the chipset drivers power plan for them.

 

[AusWolf]

Hmmmm I wonder if it was done with integrated graphics and no dedicated gpu??

I use integrated graphics myself (for the second HDMI output), and my idle power is between 20-25 Watts.

As we've learned from other reports, one should watch out for unnecessarily high SoC voltages when EXPO is enabled.

 

[Jism]

If i'm correct that cache cant turn into an idle state. So it's constantly running. Same as with Threadripper; some CPU's to be seen with 100 ~ 150W idle power consumption, which is considered normal.

 

[Denver]

Click Bait Bs. Great.

 

[JustBenching]

It's likely motherboard related. I had a 7700X and now have a 7800X3D and didn't see any idle spikes when keeping HWInfo open for hours. I'd get between 28w - 32w on idle.

28 to 32w on idle is insanely high actually, but that's not the point. Hwinfo doesn't poll fast enough to detect spikes, that's common sense, you need dedicated hardware

 

[R0H1T]

~30W is not that high & the system isn't actually idle given the CPU usage reported, it should be 1-2% on a typical 16c/32t system. There were programs running actively in the background ~ that's also common sense.

 

[AusWolf]

So then... what if I don't have specialist tools that can detect such small spikes? My 7700X has been running fine since November. Is there anything I should know?

 

[R0H1T]

You're fine, sorry to say this but a lot of hardware review(er)s neglect the OS part of the equation! The first thing I do when I install a fresh windows, whatever version, is to disable or set to manual any system services/drivers that aren't required to boot. Check your process explorer/hacker for the number of such entries which are set to boot or auto start ~ I bet you can disable/set to manual at least 20-40% of them. Then you go on complain about some random spikes in CPU/power usage

Ok Windows is not perfect, neither is AMD but you can lower your system power consumption & CPU usage by being a bit more informed, easily! If you're a software developer do you also neglect some rando process/thread running on that (company) server?

 

[AusWolf]

You're fine, sorry to say this but a lot of hardware review(er)s neglect the OS part of the equation! The first thing I do when I install a fresh windows, whatever version, is to disable or set to manual any system services/drivers that aren't required to boot. Check your process explorer/hacker for the number of such entries which are set to boot or auto start ~ I bet you can disable/set to manual at least 20-40% of them. Then you go on complain about some random spikes in CPU/power usage

Ok Windows is not perfect, neither is AMD but you can lower your system power consumption & CPU usage by being a bit more informed, easily! If you're a software developer do you also neglect some rando process/thread running on that (company) server?

That is true, although I never messed with those settings, and I also have Steam, GOG Galaxy and the Wargaming.net launcher starting automatically and running in the background. Like I said, no issues so far.

I never checked the underside of the CPU after I first installed it back in November, but I guess I would have noticed it during use if it had burned itself in a power spike.

Edit: One more thing... a single power spike of 130 W is well within the CPU's TDP (PPT) limit, so it shouldn't be able to cause any damage.

 

[phubar]

By the time your processor shows signs of damage on the bottom of the package its already dead.

So there is no point in pulling out the processor to inspect it from time to time.

This issue woldn't cause any damage. So far as anyone can tell it causes some increased power usage at idle. Which isn't ideal at all, it needs fixing, but it won't damage anything or cause instability or effect performance or security.

Currently it seems the damaged processors that have shown up are due to EXPO or XMP being enabled and the BIOS (its not clear to me if its the AGESA or the mobo OEM who sets this) volts for the SOC being set WAY too high automatically. To something like 1.4-1.5v. Apparently 1.3v or less is considered safe going by others comments who know more than me.

Steven from GN sounds like he'll have a good video on the subject coming out in a few days so we'll know more then.

 

[Suspecto]

it is way worse than we thought

 

[Why_Me]

it is way worse than we thought

Beat me to it. I was just about to post that vid.

 

[Mussels]

Beat me to it. I was just about to post that vid.

Theres three seperate values/power settings to read and a total so people confuse them regularly

You could read that as 5.3W 21W or 33.6W for the CPU, depending on which sensor reading you use - all are accurate, but all give a totally different reading.
(5.4W for the CPU Cores, 15W for my OC'd SoC, and the rest for whatever else isn't already counted like the USB and SATA ports i guess - I cant find a solid answer on this, but 10W goes to something else in the CPU package and always has)

Edit from HWinfo website:

Zen 3 Package Power | HWiNFO Forum

They estimate higher than they are, since theres no direct measurement for the total reading.



On the AM5 recent problems, It looks like a mix of a few problems so far:

MSI: A hibernate bug stopping coolers working
Gigabyte/Aorus: SoC voltages being set high no matter what the user sets (and not reporting those values to the user in BIOS/software)

Asus: Automatically boosted SoC way too high to "help" overclock
Even the fixed BIOS is kicking to 1.350v when the users are finding 1.25 works fine


Is it enabling EXPO that does this, or PBO at "motherboard levels" (which are always balls to the wall stupidity) ?

 

[AusWolf]

I've just updated to BIOS version 1.82, which, according to MSi, contains the max. voltage fix, that I can confirm.

However, with a 1.2 VSoC (same as before), my idle power consumption rose from 20-22 W to around 25 W, and with that, my idle temperature is 45 °C whereas it was around 39 °C before with the same fan curve.

So far, I'm not happy.

 

[Dr. Dro]

it is way worse than we thought

The more things change, the more they stay the same

Sadly no bangers like Project X for soundtrack this time around

 

[phubar]

it is way worse than we thought

That is bad but its not related to the idle power spike issue.

Going by what GN is saying not many will be effected by it but that its happened at all is stupid and AMD and the mobo vendors gotta get their crap together.

Is it enabling EXPO that does this, or PBO at "motherboard levels" (which are always balls to the wall stupidity) ?

EXPO has nothing to do with voltages. Its in effect a more advanced form of XMP that does a better job of setting secondary and tertiary timings and Steve says as much in the vid.

The problem is that these motherboards would set key voltages way too high for no good reason when XMP/EXPO were used. Basically either the AGESA or mobo vendors voltage tables weren't right for some reason.

Given that AMD has said they've released new AGESA's that limit the volts you can set it does look like AMD can have some control over this.

I also remember though back a few years ago when mobo vendors were cranking volts up real high on Intel 8000 and 9000 series chips and Intel eventually had to put their foot down on that.

Yeah GN did a vid on that too, looks like they were still doing in it for the Intel 10000 series:

I don't think anyone had their chips burnt out but the motherboard guys were clearly seeing how far they could push the limits back then and I can't help but wonder if they did so again here.

IMO even if AMD's AGESA isn't at fault Steve is right that AMD should've done something about this before it became a issue.

 

[Dr. Dro]

That is bad but its not related to the idle power spike issue.

Going by what GN is saying not many will be effected by it but that its happened at all is stupid and AMD and the mobo vendors gotta get their crap together.


EXPO has nothing to do with voltages. Its in effect a more advanced form of XMP that does a better job of setting secondary and tertiary timings and Steve says as much in the vid.

The problem is that these motherboards would set key voltages way too high for no good reason when XMP/EXPO were used. Basically either the AGESA or mobo vendors voltage tables weren't right for some reason.

Given that AMD has said they've released new AGESA's that limit the volts you can set it does look like AMD can have some control over this.

I also remember though back a few years ago when mobo vendors were cranking volts up real high on Intel 8000 and 9000 series chips and Intel eventually had to put their foot down on that.

Yeah GN did a vid on that too, looks like they were still doing in it for the Intel 10000 series:

I don't think anyone had their chips burnt out but the motherboard guys were clearly seeing how far they could push the limits back then and I can't help but wonder if they did so again here.

IMO even if AMD's AGESA isn't at fault Steve is right that AMD should've done something about this before it became a issue.

AGESA/SMU is very much at fault for reasons he explained in the video, it seems everyone shares a largely equal part of the blame, more so ASUS for their excessive voltage and loose OCP, AMD not providing guidelines, and also the CPU that exploded on first boot without any EXPO applied...

With multiple vendors exhibiting the problem, it displays at the very minimum poor communication and lack of certification, as well as poor quality control by AMD. Launching a new brand to compete with XMP, alongside newer CPUs and a new memory standard, AMD should have had the foresight to establish a certification board and publish detailed guidelines. Gigabyte's board surviving shows that they've been the closest to right in a situation that is so wrong that there's no excuse for this.

Karmic rebalance complete, though. I just hope that all parties affected have their products replaced ASAP, with as little bureaucracy and questions asked as possible.

 

[R-T-B]

Click Bait Bs. Great.

Yeah, as an owner of a 7950x, this is decidedly something I'd like to know about, and NOT "clickbait."