AMD Releases Second Official Statement Regarding Ryzen 7000X3D Issues

[R-T-B]

Yeah but unfortunately they have locked soc voltage to 1.300v on my 7950x, x670e taichi, so back to an older bios ffs...

If you aren't pairing that 1.3+ volts with some sort of exotic cooling, you may be asking for a burned CPU.

I really have a hard time believing anyone should be running greater than 1.3v, the benefits are few and the dangers are seemingly many.

So.........instead they're just blaming VSOC? While for 7 months now, people everywhere have been defaulting to 1.35V VSOC and no one at AMD so much as batted an eyelid? And so now suddenly a reduction of 50mV will magically prevent physical damage?

I'm not sure what conclusion buildzoid came to in his video, but no evidence so far clearly points to VSOC. VSOC max current draw remains low on chiplets. The visible evidence of damaged pads and areas are all far away from VSOC pads and IO die areas.

I agree there is likely more to the story here, but I personally wouldn't be running over 1.3v SOC until we know the whole picture.

I doubt that very much, considering there are clearly samples of Raphael out there that require that VSOC to hit 3000MHz UCLK. It's not like they got rid of IO die silicon quality variance overnight.

I'm not a fan of disabling features either. Perhaps they could just plaster a big warning on voltages past a point, and trip some efuse or something that I'm sure they could check warranty wise.

So, basically (if I were to take Yuri's statement as fact) an SOC voltage of less than 1.3 volts is safe. So, my previous voltage setting of 1.245 and 1.25 were safe, at least according to him.

Yes, if their obeyed and the alleged bug of copying VDIMM as SOC voltage isn't real.

My VDIMMs run at around 1.35v in EXPO, so even if the board did do that, hopefully nothing would just explode.

 

[trparky]

Yes, if their obeyed and the alleged bug of copying VDIMM as SOC voltage isn't real.

My VDIMMs run at around 1.35v so even if they did do that, hopefully nothing would just explode.

You mean this?

 

[tabascosauz]

So, basically (if I were to take Yuri's statement as fact) an SOC voltage of less than 1.3 volts is safe. So, my previous voltage setting of 1.245 and 1.25 were safe, at least according to him.

I mean, just update to one of the new BIOSes and see what the board defaults to under EXPO now. All the vendors have made a big deal of safe limits in their new BIOS releases.

Reading the original statement again, it reads like AMD hasn't gotten to the bottom of it yet. It's claimed that numerous rails now receive caps, and maybe the only reason VSOC is mentioned is because it's the most prominent rail that users will actually adjust and interact with. Most people with even superficial knowledge of Ryzen don't touch Vcore directly and the minor rails aren't worth mentioning, but almost anyone who has to delve into anything memory-related will have to familiarize themselves with VSOC.

If you aren't pairing that 1.3+ volts with some sort of exotic cooling, you may be asking for a burned CPU.

I'm not a fan of disabling features either. Perhaps they could just plaster a big warning on voltages past a point, and trip some efuse or something that I'm sure they could check warranty wise.

VSOC doesn't draw nearly enough power on non-APUs for it to become a thermal problem, unless there is some sort of short or hardware issue. Hence the speculation over what causes the hot spot/burned area.

To be fair, AMD only guarantees 5200. I don't like it either and I think it's a scummy move to be advertising "sweet spot" speeds on official slides that not all CPUs can hit, but that's their policy.

You mean this?
View attachment 293481

Ryzen Master has always displayed VDIMM as "MEM VDDIO", so yeah, if the board automatically copied VDIMM to VSOC as alleged (not doubting that it's a possible bug, just not sure it's what's the problem here).

 

[trparky]

I mean, just update to one of the new BIOSes and see what the board defaults to under EXPO now. All the vendors have made a big deal of safe limits in their new BIOS releases.

Right now, there's no new update for my board, at least, no update that mentions a new AGESA version. Still the same 1.0.0.6 AGESA version.
B650 AORUS ELITE AX (rev. 1.x) Support | Motherboard - GIGABYTE Global

 

[pressing on]

B650E-E Strix 1412 now.

I did not have CPU, then used flashback from 801 (bulk) to 1406, then all reports came up and i applied 1409, today 1412, i feel bad... a Bios has not to be updated, but only if needed.

I suspect that there are many users who just don't upgrade their BIOS. Some because they don't want to loose custom settings and/or the hassle of having to reapply them.

It is possible for BIOS updates to be issued through Windows Update, and in this instance I think there is a case for AMD to ask its board partners to do this for all motherboard models.

 

[trparky]

It is possible for BIOS updates to be issued through Windows Update, and in this instance I think there is a case for AMD to ask its board partners to do this for all motherboard models.

I really don't like that idea at all. Microsoft can't even get driver updates right through Windows Update and now you want them to update something even lower level than that? Not just no but hell no!

Ryzen Master has always displayed VDIMM as "MEM VDDIO", so yeah, if the board automatically copied VDIMM to VSOC as alleged (not doubting that it's a bug, just not sure it's what's the problem here).

I think I'm going to just run with EXPO disabled until a new AGESA version is released. Probably 1.0.0.7 or at least some kind of letter after the 6 in 1.0.0.6.

 

[tabascosauz]

I suspect that there are many users who just don't upgrade their BIOS. Some because they don't want to loose custom settings and/or the hassle of having to reapply them.

It is possible for BIOS updates to be issued through Windows Update, and in this instance I think there is a case for AMD to ask its board partners to do this for all motherboard models.

For prebuilts and laptops sure, but on the retail market there are a staggering amount of boards that vendors would have to account for and push individualized capsule BIOS updates for. Suffice to say this isn't happening anytime soon, even though most vendors (including Gigabyte now) have moved to the file format.

 

[tpa-pr]

I really don't like that idea at all. Microsoft can't even get driver updates right through Windows Update and now you want them to update something even lower level than that? Not just no but hell no!

You've hit the nail on the head. We're in discussion about disabling the UEFI capsule service by default on our workstations because of the amount of machines we've had bricked by failed/partial BIOS installs from Windows Update.

Anyway, i'll be updating my BIOS when I get home tonight as a precautionary measure. But since I've never had EXPO or any sort of overclocking enabled I assume everything is OK with my setup. I guess only time will tell.

 

[MarsM4N]

@crmaris from "Hardware Busters" made some interesting observations. He monitored some high idle power spikes. Not clear what BIOS versions he used. Maybe he can clarify.

Quote: "There are some interesting facts here, which I didn’t pay much attention to during the reviews because I only look at the average values and not the peak ones in idle. In the 7950X3D, there is a high spike during idle at 130W, which is unjustified because the peak CPU load is only 3.53%. Even with the Curve Optimized enabled and a -15 setting, the idle power spike is close to 125W, so something is happening there. On the 7800X3D, the spike during idle stays low, but this is not the case for the 7900X, which has an idle power spike at 109W, while the peak CPU load at idle was at 5.12%, so these 109W are not justified, either."

https://hwbusters.com/cpu/amd-ryzen-7000-series-burn-issues-our-investigation-based-on-real-data/

Yeah but unfortunately they have locked soc voltage to 1.300v on my 7950x, x670e taichi, so back to an older bios ffs...

In case you missed the memo, they limited the voltage so your CPU doesn't blow up. But ey, if you're into blowing up CPU's (+ mobo's), go for it. Good for the economy.

 

[R-T-B]

You mean this?
View attachment 293481

I think that's it but don't quote me on the theoretical bug, if it exists at all.

I think I'm going to just run with EXPO disabled until a new AGESA version is released. Probably 1.0.0.7 or at least some kind of letter after the 6 in 1.0.0.6.

If you have an x3d and can't afford downtime that's probably safest no doubt.

As for me? Mine isn't an x3d, and I have a spare PC, so I'll risk it for now.

 

[trparky]

If you have an x3d and can't afford downtime that's probably safest no doubt.

Mine isn't an X3D, it's a plain X-model chip.

I have a spare PC, so I'll risk it for now.

I have a spare PC too but that doesn't mean I like to play with fire.

 

[Metroid]

Again, you should read what you quote, MSI literally doesn't say anything about VDIMM lol...........

If you are referring to the alleged bug of BIOS copying VDIMM into VSOC and VDDIO, 1usmus does have a tweet mentioning it, but none of the board vendors make reference to it.

In any case, there are a lot of 1.4V EXPO kits out there, but if it still exists such a VDIMM=VSOC bug doesn't seem to be a widespread one. If it was a major AGESA bug, then everyone with a decent Hynix kit would apply EXPO and immediately see 1.4V SOC set, which is not the case. So maybe board and/or BIOS specific. The board vendors are more hinting at a different form of VSOC overvoltage.

Not sure about the other vendors, but MSI's response seems to be removing manual and positive offset voltage adjustments across different rails. Which makes a lot more sense than AMD pinning the blame on VSOC, MSI's focus seems to be on dying X3Ds in particular, and VSOC has nothing to do with V-cache.

To be honest, AMD should have just kept the hard cap limitations that were on 5800X3D (hard Fmax limiter, hard Vcore limiter). Either they removed those protections for Raphael-X in the spirit of their advertised "overclocking" to generate more interest in the new X3Ds, or AGESA is bugged so that the protections are worthless, or the board partners fucked it up.

https://twitter.com/x/status/1650791399699218433

MSI said this "As soon as EXPO is enabled, the CPU SOC and CPU VDDIO/MC voltages increase to 1.36-1.4v, sometimes boosting to 1.5V in Windows, causing instant death on the X3D.", I quoted this 3 times and linked the reference.

 

[trparky]

MSI said this "As soon as EXPO is enabled, the CPU SOC and CPU VDDIO/MC voltages increase to 1.36-1.4v, sometimes boosting to 1.5V in Windows, causing instant death on the X3D.", I quoted this 3 times and linked the reference.

But does this happen on non-X3D chips? Like say, the 7700X; which is what I have.

 

[tabascosauz]

MSI said this "As soon as EXPO is enabled, the CPU SOC and CPU VDDIO/MC voltages increase to 1.36-1.4v, sometimes boosting to 1.5V in Windows, causing instant death on the X3D.", I quoted this 3 times and linked the reference.

And? VDDIO/MC is not VDIMM. The "increase" from enabling EXPO is naturally from the board picking an (admittedly obscene) auto-voltage to support higher FCLK and UCLK that comes with running the EXPO profile at 1:1. That's not confirmation that the board is copying VDIMM onto VSOC and VDDIO/MC - you're not running a 6000CL30 kit on 1.0V VSOC without the "increase".

As for the spikes, there are screenshots showing north of 2.0V spikes on Vcore and VSOC......no telling if that's a really uncharacteristically severe case of software monitoring bugs, or something else. 2.0V should be an insta-kill......not long enough to sit in Windows and wait for it to log in the Max column...

 

[R-T-B]

I have a spare PC too but that doesn't mean I like to play with fire.

I don't mind having to RMA something, which they are clear they will cover. YMMV, of course. I don't think it's much of a real fire (as in flames) risk.

 

[trparky]

I don't mind having to RMA something, which they are clear they will cover. YMMV, of course. I don't think it's much of a real fire (as in flames) risk.

Figuratively playing with fire.

 

[R-T-B]

Figuratively playing with fire.

I figured but was being complete just in case you were being literal.

Of course, what we are comfortable with varies. I can by no means blame you for being safe.

 

[Dirt Chip]

But does this happen on non-X3D chips? Like say, the 7700X; which is what I have.

I thought the same question: is this phenomenon is on all sku or just with MCM 2-chiplet\ccd design?
If the letter , maybe the voltage of the non 3d ccd apply by incident to the 3d ccd in some situation though the common memory voltage setting. Just a speculation.

(I'm no too technically educated about CPU topography so I may just blabber nonsense...)

 

[HD64G]

So, the solution was that latest UEFIs won't allow the vendors to have auto voltage of the SOC higher than 1.3V which is already high overclock. Safe SOC Voltage is up to 1.2 if not using exotic cooling.

 

[trparky]

But what's confusing is that none of the recent BIOS versions indicate that that's one of the fixes included in them. Suffice it to say, all of this is about clear as mud.

 

[HD64G]

But what's confusing is that none of the recent BIOS versions indicate that that's one of the fixes included in them. Suffice it to say, all of this is about clear as mud.

Even so, it is very easy to change the SOCV from auto to manual and allow it to reach up to 1,2V avoiding any problem. The root cause is found and anyone can prevent the damage now.

 

[trparky]

Even so, it is very easy to change the SOCV from auto to manual and allow it to reach up to 1,2V avoiding any problem. The root cause is found and anyone can prevent the damage now.

I’ve tried, I can’t seem to figure out how to do that in my BIOS.

 

[AusWolf]

So it was the SoC voltage (and motherboards setting it too high) after all! I'm glad they've resolved the issue.

Question: do we actually need a high SoC voltage for EXPO / memory OC?

 

[trparky]

I still want to know what an acceptable SOC voltage is and if utilities like HWInfo and Ryzen Master is reporting correct values. I keep hearing so much misinformation.

 

[AusWolf]

I still want to know what an acceptable SOC voltage is and if utilities like HWInfo and Ryzen Master is reporting correct values. I keep hearing so much misinformation.

If AMD limited it to 1.3 V in the new AGESA, then I suppose that's the safe limit.