Ryzen Owners Zen Garden

[AusWolf]

Stop talking about the X3D! Some people are fighting with their systems to get their SoC voltage right, and here I am, fighting with my urge to get a 7800X3D just because why not. All this X3D talk is not helping.

 

[Space Lynx]

Stop talking about the X3D! Some people are fighting with their systems to get their SoC voltage right, and here I am, fighting with my urge to get a 7800X3D just because why not. All this X3D talk is not helping.

I am still considering just saying yolo and getting one myself. I'm not worried at all about soc damage... bios update fixed all that already.

 

[Mussels]

SVI3 is the CPU "die sense". Any other Vcore or VSOC under the board sensors block is taken from some point along the VRM or the socket, unless stated otherwise (Crosshair VIII). Trust the CPU sensor.

@Nordic 1482 is fine. CO disproportionately affects the 5800X3D, it is your main source of performance improvement, power reduction, and thermal reduction all in one.

Think reous is on OCN, if you wanna hit him up to ask for a BIOS.

VID vs Volts
Voltage requested vs real-time voltage - LLC for example always offsets the voltage under load to prevent it drooping

That shows 1.2V is requested/"Sent" by the board, which then an in-line sensor is noting is 1.236v in real time.

If you run a stress test or change memory settings (more RAM sticks, higher speeds) you'll notice that 1.236v either go down (droop) or up (LLC kicking in too hard) or wobble a little (the goldilocks zone)


My motherboard has "VIN6" which seems to be the SoC voltage as set in the BIOS
Why 1.15v becomes 1.136v is likely that the controller only works in 'steps' - either the actual voltage itself or the component reading it back to software.

It never varies

While the reading Zentimings and HWinfo use from the CPU, shows it changing under load/idle conditions - but very little wiggle room overall

I ran testmem5 to get this to occur, but notice how 1.15v becomes 1.131v at idle, 1.125v at load, and 1.138v when that load suddenly ends?

If the boards ramp up the power to prevent droop and the load ends, you end up with far more than expected - but this tech has been around for ever and rarely do the automatic settings do insane things.

The other part seems to be that for whatever reason they scale that voltage up infinitely, instead of a safety limit (Which AMD has now enforced on the CPU side)

 

[tabascosauz]

VID vs Volts
Voltage requested vs real-time voltage - LLC for example always offsets the voltage under load to prevent it drooping

That shows 1.2V is requested/"Sent" by the board, which then an in-line sensor is noting is 1.236v in real time.

If you run a stress test or change memory settings (more RAM sticks, higher speeds) you'll notice that 1.236v either go down (droop) or up (LLC kicking in too hard) or wobble a little (the goldilocks zone)

VID is not any of the above........they're all VSOC, just to varying degrees of accuracy and in different places. SVI3 is die sense, board sensor is socket sense, and VRM sensor is well VRM if applicable.

 

[Space Lynx]

VID is not any of the above........they're all VSOC, just to varying degrees of accuracy and in different places. SVI3 is die sense, board sensor is socket sense, and VRM sensor is well VRM if applicable.

I'm just going to run latest BIOS, set ram to EXPO if its qvl for mobo, and play games, and I will get the same fps in games as the rest of you spending hours tinkering lol

 

[Mussels]

Can you provide a timestamp where Steve showed 1.8v probed?
I've only saw 1.45v max in his video, unless you're reffering to some other video/test.

I can't find the 1.8v in the GN video, his shows 1.10V BIOS -> 1.45 probed as you said

The 1.8v was from a user on reddit who probed his own board, but it's been lost in my browser history now - the downsides to having a few hundred tabs open at a time.
I quoted it and screenshot it in one TPU thread on the topic, then wrote about it in others - but cant find where i originally copied it to/from right now.

A lot of their threads got merged into this one mega-thread, so it may be in here
Megathread for AM5 (Ryzen 7000) Damage/Burn-out/EXPO/Voltage issues : Amd

 

[Space Lynx]

I can't find the 1.8v in the GN video, his shows 1.10V BIOS -> 1.45 probed as you said

The 1.8v i believed was from a gigabyte board, but with dozens of people quoting these videos all over the place i'm not sure where I saw the 1.8v now

wouldn't updating to latest BIOS fix all these voltage issues though? so this convo shouldn't even be a convo anymore, it should just be, go update to latest BIOS.

 

[Mussels]

VID is not any of the above........they're all VSOC, just to varying degrees of accuracy and in different places. SVI3 is die sense, board sensor is socket sense, and VRM sensor is well VRM if applicable.

You're thinking of VID as in the CPU voltage, I'm saying it's the same thing here - it's a requested voltage vs the supplied voltage.

CPU VID, CPU voltage
SoC VID, SoC voltage

The SoC is still part of the CPU and works the same way, it requests a voltage but then doesn't check if the actual supplied voltage is higher or lower than that value - that's in the motherboards hands, and how voltage offsets/curve optimiser work.

wouldn't updating to latest BIOS fix all these voltage issues though? so this convo shouldn't even be a convo anymore, it should just be, go update to latest BIOS.

No ones fully confirmed that yet, to my knowledge
They're safer, sure - but no one knows if EVERY problem is fixed

As an example if it was forcing 1.10v to 1.45v, now it may force AMD's new limit at 1.30v - what if you wanted 1.10v and still cant get it?

 

[tabascosauz]

You're thinking of VID as in the CPU voltage, I'm saying it's the same thing here - it's a requested voltage vs the supplied voltage.

The SoC is still part of the CPU and works the same way, it requests a voltage but then doesn't check if the actual supplied voltage is higher or lower than that value

Never have I seen anyone refer to a SVI2 source as being VID. SVI Vcore isn't VID, and haven't seen anyone prove that VSOC is any different.

Not sure about the board sensor as it's just routed through the SuperIO, but it's never really seriously relied on outside of ROG boards.

The VRM sensor has nothing to do with VID; it's just literally the voltage going through either Vcore or VSOC domain in the PWM controller, and usually not labelled as Vcore or VSOC but just voltage over a particular VRM loop.

I'm just going to run latest BIOS, set ram to EXPO if its qvl for mobo, and play games, and I will get the same fps in games as the rest of you spending hours tinkering lol

Well, for your sake I sure as well hope that an AM5 CPU at least matches a 5800X3D lmao

 

[Space Lynx]

Well, for your sake I sure as well hope that an AM5 CPU at least matches a 5800X3D lmao

I am not sure what you mean, I run a 5600 non-x, I paid $130 for it, my buddy irl is going to give me $100 for it, so i just have to sell mobo and ram, probably at a big loss, but I am ok with that as I got a bonus from work recently.

My plan at the moment is 7800 x3d + asrock b650m mobo that hardware unboxed gave the thumbs up on in a recent video and only costs $120, and then some 6000 cl 30 ram at $129 2x16gb and ram is qvl for mobo.

that's my current plan... but I have not decided 100% yet... 8000 series will be out in 1 year or possibly less than 1 year. so maybe I will just wait.

 

[Mussels]

I am not sure what you mean, I run a 5600 non-x, I paid $130 for it, my buddy irl is going to give me $100 for it, so i just have to sell mobo and ram, probably at a big loss, but I am ok with that as I got a bonus from work recently.

My plan at the moment is 7800 x3d + asrock b650m mobo that hardware unboxed gave the thumbs up on in a recent video and only costs $120, and then some 6000 cl 30 ram at $129 2x16gb and ram is qvl for mobo.

that's my current plan... but I have not decided 100% yet... 8000 series will be out in 1 year or possibly less than 1 year. so maybe I will just wait.

What part wasn't clear?

AMD have put out a voltage cap at 1.30v, but boards can still lie and send 1.30v when the BIOS and software says less. If that happens to you, you're always over-volting the SoC no matter what you choose in the BIOS.

 

[Space Lynx]

What part wasn't clear?

AMD have put out a voltage cap at 1.30v, but boards can still lie and send 1.30v when the BIOS and software says less. If that happens to you, you're always over-volting the SoC no matter what you choose in the BIOS.

i never said you were not clear... I quoted @tabascosauz ...

 

[Mussels]

My bad

You get one free like as compensation for the emotional damage of me not paying attention

 

[AusWolf]

I'm just going to run latest BIOS, set ram to EXPO if its qvl for mobo, and play games, and I will get the same fps in games as the rest of you spending hours tinkering lol

All it took me is "enable EXPO -> restart -> set VSOC to 1.1 V -> restart -> set VSOC back to Auto -> enjoy".

wouldn't updating to latest BIOS fix all these voltage issues though? so this convo shouldn't even be a convo anymore, it should just be, go update to latest BIOS.

The latest AGESA fixes the maximum VSOC at 1.3 V, which is safe, but still a bit too high, imo. Dropping it to 1.2 V with the above method saves me at least 5, if not 10 °C on my idle temp.

 

[Ruru]

i would ask him to come down to 285 eur shipped. its used after all.

He agreed that I'll pay 270EUR + postage for that

 

[freeagent]

wouldn't updating to latest BIOS fix all these voltage issues though? so this convo shouldn't even be a convo anymore, it should just be, go update to latest BIOS

Personally, I doubt it’s a software problem.

 

[R-T-B]

My bad

You get one free like as compensation for the emotional damage of me not paying attention

When do I get one? I've been needing one for some time now.

We're talking about CPUs, not mod smackdowns or something, right?

 

[tabascosauz]

GN video is out. Lots of very cool imagery. Just some thoughts:

1. The acoustic scan result is interesting - the bubbles(?)/voids in the indium solder application could certainly explain some of the wack core deltas that some samples have.
2. Crazy how little thickness the base die and Vcache take up. Maybe the silicon spacer isn't to blame for X3D thermals, as much as cache being in the way of heat dissipation from the CCD.
3. So allegedly the theory goes high VSOC >> exponentially more leakage and degradation, where low SP samples have a head start >> damage is extensive enough to short and deliver massive current in small area >> silicon melts/indium solder melts/heat builds up >> boom takes path of least resistance?

Steve saying that the precursor degradation is irreversible......obviously, everyone knows that's how silicon degradation works, but in past generations "degradation" due to bad SP quality and/or unsafe OC was often visibly obvious. On AM5 there haven't been many consistent, visible signs of this sort of "slow damage" over time. As an existing [Asus] owner, you'd just have to update to a safe BIOS and hope that there's still enough life in the CPU to last for x years..........and hope that the board now has working OCP that can hopefully limit the extent of damage (ie. saving the board) if it does happen. Because AMD sure as hell won't replace AM5 CPUs en masse for preventative reasons.

I understand that the lab is just being honest when they say they can't isolate any of the known factors (thermal runaway, short, VSOC degradation) from dozens of potential manufacturing or environmental factors (just rattled off like, 30+ non-exhaustive examples), but the video hardly arrives at the "Truth About AMD's CPU Failures" - it just leaves more questions than answers.

If anything, GN's investigation confirmed just one thing: AMD won't be the one to deliver that "Truth", pretty clear from their one and only official statement on the matter that they believe the new AGESA voltage limits will make the problem magically go away.

 

[AusWolf]

GN video is out. Lots of very cool imagery. Just some thoughts:

1. The acoustic scan result is interesting - the bubbles(?)/voids in the indium solder application could certainly explain some of the wack core deltas that some samples have.
2. Crazy how little thickness the base die and Vcache take up. Maybe the silicon spacer isn't to blame for X3D thermals, as much as cache being in the way of heat dissipation from the CCD.
3. So allegedly the theory goes high VSOC >> exponentially more leakage and degradation, where low SP samples have a head start >> damage is extensive enough to short and deliver massive current in small area >> silicon melts/indium solder melts/heat builds up >> boom takes path of least resistance?

Steve saying that the precursor degradation is irreversible......obviously, everyone knows that's how silicon degradation works, but in past generations "degradation" due to bad SP quality and/or unsafe OC was often visibly obvious. On AM5 there haven't been many consistent, visible signs of this sort of "slow damage" over time. As an existing [Asus] owner, you'd just have to update to a safe BIOS and hope that there's still enough life in the CPU to last for x years..........and hope that the board now has working OCP that can hopefully limit the extent of damage (ie. saving the board) if it does happen. Because AMD sure as hell won't replace AM5 CPUs en masse for preventative reasons.

I understand that the lab is just being honest when they say they can't isolate any of the known factors (thermal runaway, short, VSOC degradation) from dozens of potential manufacturing or environmental factors (just rattled off like, 30+ non-exhaustive examples), but the video hardly arrives at the "Truth About AMD's CPU Failures" - it just leaves more questions than answers.

If anything, GN's investigation confirmed just one thing: AMD won't be the one to deliver that "Truth", pretty clear from their one and only official statement on the matter that they believe the new AGESA voltage limits will make the problem magically go away.

So basically, the new BIOS protects you from your CPU blowing up due to extreme high SoC voltage, but it doesn't do anything against a potential slow degradation due to anything else. Awesome.

How much more proof does AMD need to see that stricter regulations are needed against board partners?

Edit: I'm starting to believe that people who swear by disabling turbo to preserve longevity actually have a point.

 

[freeagent]

How much more proof does AMD need to see that stricter regulations are needed against board partners

I don't think that is the problem..

I think they just rushed this out.. without doing their due diligence first.

 

[AusWolf]

I don't think that is the problem..

I think they just rushed this out.. without doing their due diligence first.

That's even worse because that would mean that even AMD doesn't have a clue.

 

[tabascosauz]

So basically, the new BIOS protects you from your CPU blowing up due to extreme high SoC voltage, but it doesn't do anything against a potential slow degradation due to anything else. Awesome.

How much more proof does AMD need to see that stricter regulations are needed against board partners?

No, should be the other way around - the new voltage limits prevent any more continued, accelerated degradation, but do nothing to reverse that which might have already happened. Unless something radically changes as to how [functional] OCP works, they don't seem to be able to prevent the catastrophic failure itself. Only hope to prevent it from ever happening on brand new CPUs/boards from here on out, and hopefully stave off failure on all those CPUs that are somewhere along the line after running months on old BIOSes.

If VSOC is truly the culprit, I expect the problem to no longer be relevant for brand new CPUs, and that's how AMD will sweep it under the rug. Replace the high-profile failures and maybe the occasionally CPU that is close enough to make it to failure in the next few months, then revise their policy (although their RMA should be lax enough to continue allowing replacements anyway through warranty).

But that's assuming that we know the whole story and can solely pin the blame on VSOC. I don't believe that's the case, only Asus has been obscene with VSOC, but other board vendors have also produced these failures (including those that were already below 1.3V).

That's even worse because that would mean that even AMD doesn't have a clue.

I also don't think AMD has any clue what's going on. But if the new BIOSes prevent future CPUs from suffering this fate (or at least prevent it from happening within say, 3-5 years), then they don't *need* to for this PR problem to go away.

Anyway, Asus is a scumbag, but honestly no more than Gigabyte has been for a decade of overvolting. Asus just happened to find the time bomb in VDDCR_SOC (assuming that's the whole story). And AMD gets no sympathy from me. All this and more has been in the cards for a long time, AM4 AGESA has been ringing the alarm bells for a long time. AMD could not care less.

 

[AusWolf]

No, should be the other way around - the new voltage limits prevent any more continued, accelerated degradation, but do nothing to reverse that which might have already happened. Unless something radically changes as to how [functional] OCP works, they don't seem to be able to prevent the catastrophic failure itself. Only hope to prevent it from ever happening on brand new CPUs/boards from here on out, and hopefully stave off failure on all those CPUs that are somewhere along the line after running months on old BIOSes.

If VSOC is truly the culprit, I expect the problem to no longer be relevant for brand new CPUs, and that's how AMD will sweep it under the rug. Replace the high-profile failures and maybe the occasionally CPU that is close enough to make it to failure in the next few months, then revise their policy (although their RMA should be lax enough to continue allowing replacements anyway through warranty).

But that's assuming that we know the whole story and can solely pin the blame on VSOC. I don't believe that's the case, only Asus has been obscene with VSOC, but other board vendors have also produced these failures (including those that were already below 1.3V).



I also don't think AMD has any clue what's going on. But if the new BIOSes prevent future CPUs from suffering this fate (or at least prevent it from happening within say, 3-5 years), then they don't *need* to for this PR problem to go away.

Anyway, Asus is a scumbag. And AMD gets no sympathy from me. All this and more has been in the cards for a long time, AM4 AGESA has been ringing the alarm bells for a long time. AMD didn't care.

Then let's hope that VSOC is the culprit, and people whose CPU hasn't blown up are fine on the new BIOS. I wouldn't want to replace my 7700X before its time is due (unless I decide to pull the trigger on that sweet 7800X3D a bit later).

 

[Klemc]

I did only a quick search, older result found of AMD telling switching from Agesa to openSIL is from 2023-3 !

I hope i receive a replacement CPU (i sent yesterday, it will be in their hands today), then i can wait the AGESA crappy 1.09.x to be out and disable Turbo and also DOCP until next year when it will be stabilized !

 

[tabascosauz]

I did only a quick search, older result found of AMD telling switching from Agesa to openSIL is from 2023-3 !

I hope i receive a replacement CPU (i sent yesterday, it will be in their hands today), then i can wait the AGESA crappy 1.09.x to be out and disable Turbo and also DOCP until next year when it will be stabilized !

openSIL won't be ready for deployment on consumer platforms for years, like 2026 at the earliest. And if it's going to be the same team and people at AMD who are responsible for AGESA right now, it's delusional to think they will treat openSIL any different from AGESA. Open source just means offloading the work to the board vendors if AMD stays this lazy.

If AMD can actually work on root-causing the problem instead of saying that they have "root-caused it already", that would restore a lot more confidence than any of these stupid PR announcements ever could.