Meanwhile I will watch with popcorn using my comfy manual 1.1vSOC.
So according to what you linked intel set up a fund for the RMAs and released a new stepping to fix the issue, the issue being that signal could degrade after multiple years.
Yep, ASUS does this. Watch the video.
It is mind boggling TBH.
Asus released the 1410 BIOS after this video was shot, that was the first BIOS for me that got 1.30v SOC right on Auto, before that DOCP/EXPO would still be at 1.35v.
Keep in mind that appears to be a reference to two different ways of activating EXPO settings, e.g. the ASUS control set 1.35v+ with EXPO on while the AMD CBS menu set a lower vSoC. AMD has had these submenus on their platforms since AM4, and it's up to the motherboard vendors whether or not to expose them. Generally it's a good idea to expose them since some settings like VDDP and VDDG are hidden otherwise.Hilarious stuff really, this whole "we have two places to set parameters in vendor BIOS'es" is stuff of legends only morons from AMD could come up with.
That's one of the reasons I said that it still feels like we haven't really gotten to the bottom of this.At least one of the reported users lost his CPU and he wasn't even using EXPO.
Just keep in mind that even if the board says it's limiting vSoC to 1.30v, it could be going as high as 1.35v in actuality. Without external hardware testing, there's no way to know for sure.
Famous last words huh, if i understood that GN video correctly, it seems that ASUS motherboards were applying quite higher vSoC even if it was manually set below it.
View attachment 80145
Hilarious stuff really, this whole "we have two places to set parameters in vendor BIOS'es" is stuff of legends only morons from AMD could come up with.
A crappy IOD need more than just 1.3v to scale if you haven't play with one.Lower your SOC voltage. There is no reason for it to be above 1.3V.
Seems more like vdroop + LLC. It overshoots during a heavy load rather than immediately after one. Unless I'm getting the wrong impression of what was showing on the multimeter they used for testing in the video.yup this is called Vdroop.
Voltage droop - Wikipedia
en.wikipedia.org
View attachment 80149
this is why i recommended 1.2875V after droop.
It's VDDCR_SOC.Where is vsoc here ?? I see a VDDCR SOC at 1.25 Is that it ?
So two thirds of the issue are from AMD, weak io die and weak CCD is what is killing the CPUs.
View attachment 80146
Going back to the topic.
My comment early in the thread about short-circuits seems oddly prescient...
Two failure modes GN has identified:
1) Zero Ohm complete shorts in cases where CPU died but motherboard is fine (due to OCP/OTP kicking in)
2) Low resistance shorts... this is where motherboard vendors who somehow put in inadequate OCP protection in their $700 motherboards (*cough*, ASUS) continue delivering power overwhelming to an already-dead CPU leading to a catastrophic failure with desoldering of the IHS, audible cracking of the CPU die, and incineration of the motherboard socket.
My comments:
ASUS - OCP protection did not kick in even at 400W+. Fail. Catastrophic fail.
AMD - Make it right for the end users. And rein in your partners (especially ASUS).
Relevant Wendell-ism:
View attachment 80147
That's how I interpreted it. It seems like Steve kind of overpromised with this video. He spent like 2/3 of the video talking about how horrible ASUS is but that doesn't really root cause the issue. He identified a ton of motherboard and BIOS bugs (one big one for Gigabyte too) however, that doesn't fully explain what's going on. It seems like we've got an explanation for how the socket got damaged on ASUS boards but that's all that's explained in this video.So two thirds of the issue are from AMD, weak io die and weak CCD is what is killing the CPUs.
If you also have a bad bios then you also get a burned mobo (the rapid disassembly means the blowing up damaging the mobo) but the bad bios is not the cause of the CPU failing it just makes it even worse.
The one you quoted is an ASUS board, and I see 1.25. The second pic is ASRock and I see nothing that SOC anywhere in the voltage description. Its a Taichi.It's VDDCR_SOC.
Which on at least the handful of ASRock boards tested by GN and Level1Techs sets it to 1.2-1.25V with EXPO, and closer to 1.0V with Auto settings. Unlike ASUS which defaults to overly high SOC voltage.
So two thirds of the issue are from AMD, weak io die and weak CCD is what is killing the CPUs.
If you also have a bad bios then you also get a burned mobo (the rapid disassembly means the blowing up damaging the mobo) but the bad bios is not the cause of the CPU failing it just makes it even worse.
That's how I interpreted it. It seems like Steve kind of overpromised with this video. He spent like 2/3 of the video talking about how horrible ASUS is but that doesn't really root cause the issue. He identified a ton of motherboard and BIOS bugs (one big one for Gigabyte too) however, that doesn't fully explain what's going on. It seems like we've got an explanation for how the socket got damaged on ASUS boards but that's all that's explained in this video.
What happens when the CPU blows up but the motherboard is intact? It seems like for that one we've got to wait for the failure analysis lab and we're still weeks away on that one. We've got a theory from Wendell about what's happening on the CPU side but it's just a theory (for now).
You're seeing what is set in the BIOS, not the actual readout on the voltage.The one you quoted is an ASUS board, and I see 1.25. The second pic is ASRock and I see nothing that SOC anywhere in the voltage description. Its a Taichi.
Those that have been running their memory at a magical 6400 or a mythical 6600 are going to have relent and settle for 6000/6200.
Then again, they may just go YOLO and keep VSOC at crazy levels for their 1% total system performance increase.
So, looks like for your board/memory combo on that BIOS it sets VDDCR_SOC to 1.25V and on load the vdroop causes it to go to 1.235V. Both of which should be safe. Assuming the software reading is accurate and ASUS isn't using some VRM trickery.Do I see 1.235 ? Thats ASUS BIOS 1410.
As for the issue... it definitely exists, and even if (as is likely) it's an edge case like +12VHPWR this should never happen, period. Proper safeguards should be in place to prevent it.HWINFO on mt Taichi also say 1.235. And I got that motherboard when the Zen 4 first came out, and the bios is original. ALL of my 7950x's are working just fine. I have 5 of them.
Aside from older ASUS BIOS, I don't see this as an issue , aside from some want to make it seem spectacular and a fail.
Well, 5 boxes, 4 different motherboards, 2 vendors, and the one ASUS board is fine with bios that is MONTHS old. It still must be a small issue, way smaller than most are making it out to be.So, looks like for your board/memory combo on that BIOS it sets VDDCR_SOC to 1.25V and on load the vdroop causes it to go to 1.235V. Both of which should be safe. Assuming the software reading is accurate and ASUS isn't using some VRM trickery.
As for the issue... it definitely exists, and even if (as is likely) it's an edge case like +12VHPWR this should never happen, period. Proper safeguards should be in place to prevent it.
What happens when the CPU blows up but the motherboard is intact? It seems like for that one we've got to wait for the failure analysis lab and we're still weeks away on that one. We've got a theory from Wendell about what's happening on the CPU side but it's just a theory (for now).
Yes, if CPU's where burning up all over the place we would probably have known. But even with 5 boxes you are nowhere near to have a statistically significant amount, to say anything about the probability for it to happen, and if only ASUS boards have been running the vsoc out of spec, then only one of your boxes would be at a higher risk.Well, 5 boxes, 4 different motherboards, 2 vendors, and the one ASUS board is fine with bios that is MONTHS old. It still must be a small issue, way smaller than most are making it out to be.