There is a huge fuss lately about ASUS mainboards, keeping high SoC voltages which seem to be the problem of the Ryzen 7000X3D failures. As you will see, other brands also retained high SoC voltages, applying a minimum drop through the new firmwares, ignoring AMD’s guidelines.
I am not going to analyze the problem. You can check out this article at Tom’s Hardware if you want to learn about it.
Suppose you don’t have the time to read it. Briefly, the high VCORE_SOC voltage seems to be the culprit behind several CPU and, consequently, mainboard failures, so AMD issued a new firmware to reduce SoC voltages to 1.3V, which is safer. So according to AMD, the same maximum safe limit for SoC voltage is 1.3V.
(Image credit: Speedrookie/Reddit)
According to recent findings by Gamers Nexus, although ASUS rolled out new BIOS which would allow for a maximum of 1.3 SoC voltage, this doesn’t seem to be the case, so I thought to give a go to a Gigabyte Aorus Master X670E mainboard that I have and share the results with you. I also have an ASUS mainboard, but since this is installed onto a test system, I don’t have the luxury to mess with it.
I soldered three wires to the mainboard, which I connected to a LabJack U3-HV, to monitor the following voltages at the same time:
VCORE
VCORE_SOC
VDIMM
I will measure the voltages in three different cases:
With an older BIOS (F7) & EXPO disabled
With an older BIOS (F7) & EXPO enabled
With the newest BIOS (F10d) & EXPO enabled
Gigabyte states that the latest BIOS limits SoC Voltage to 1.3V max.
F7 BIOS Without EXPO
With the old bios and without EXPO enabled, Vcore_SOC and VDIMM are at low voltage levels. These readings are with the system at idle.
F7 BIOS With EXPO
Once EXPO is enabled, Vcore_SOC goes through the roof, reaching 1.42V. These readings are with the system running Prime95.
F10d BIOS With EXPO
With the new BIOS, which claims to limit Vcore_SOC to 1.3V, the screenshot above shows this is not true. The difference is 0.056V, with Vcore_SOC exceeding 1.36V. So I fail to see where the fix is that GBT promises in the F10d changelog. So ASUS is not the only one to “cheat” in Vcore_SOC voltage. If I test more mainboards, I will find several with Vcore_SOC exceeding 1.35V.
AMD’s newest statement in this issue is the following:
We have root caused the issue and have already distributed a new AGESA that puts measures in place on certain power rails on AM5 motherboards to prevent the CPU from operating beyond its specification limits, including a cap on SOC voltage at 1.3V. None of these changes affect the ability of our Ryzen 7000 Series processors to overclock memory using EXPO or XMP kits or boost performance using PBO technology.
Some mainboard manufacturers don’t care about AMD’s fix, applying higher SOC voltage by adjusting the voltages on the mainboard VRMs.
Where did Gigabyte claim that they have limited VSoC to 1.3V in F10d BIOS update? You mean “5. Follow AMD guidance to limit max. SOC voltage to 1.3V for Ryzen 7000 series X3D and non-X3D CPUs.”? No, that’s not it. It means “YOU SHOULD follow” the AMD’s guidelines and set the limits your self, because they have not.
What was the voltage read from the board sensors/hwinfo64? Did the board sensors match your sensors? Or were they inaccurate so the BIOS was sending what it thought was appropriate voltage but wasn’t?
Yes, this voltage points are a little differents than direct contacts at CPU, so basically seems fine. Also PRIME95 is not case, who people are using on daiyl basis or often for stability testing (its most crazy stress program ever, modern CPUs from Intel donwclocking turbo ratios with PRIME95 a lot)
I appreciate the sentiment, but you’re talking about how things ought to be designed in theory vs how they are designed in practice. Sadly, nothing actually makes it to production with that ethic in tact. Not even nuclear power plants, airplanes, or space shuttles are engineered to handle the worst case scenario.
Furmark should not be messing with voltage limiters and I highly doubt it is, it’s just a benchmark program and not a system tuning program like Ryzen Master, MSI Afterburner etc. therefore the voltage limiter is not limiting the CPU to 1.3v on the vSOC when it comes to any software running on the PC.
This is assuming Gigabyte aren’t doing an Asus by having a special setting in the BIOS that can be configured to “optimize” the system for certain benchmark programs (which can in theory be designed to bypass limitations).
@crmaris did you check if such a setting exists in the Gigabyte BIOS and disabled it if found?
What is the board/hwinfo64 reporting for the voltages? Does the board know it is providing the CPU with too much voltage or are its sensors wrong which tricks the BIOS into thinking it’s doing its job and limiting the voltage correctly?
Are truly only X670E motherboards impacted here, or are B650 experiencing the same problems (perhaps in lower amounts so taking longer for the damage to become apparent)? It’s a bit of a minefield currently to try and put together a new system.
the voltage read points on most motherboards are connected to the power plane before the CPU socket so you get higher measurements than what the CPU actually gets. This is even more of a problem if you don’t use a GND connection near the CPU.
Thank you for your input! I am sorry but I don’t agree on this: GND is the same anywhere in the circuit. Unless you mean reference voltage because in DC you basically measure voltage difference between two different voltage levels. Still, VRMs have the same reference (zero voltage) as what the PSU provides.
On the second part I agree with you:
When the load is applied, wherever in the line AFTER the VRM you measure it, you will get the same voltage, UNLESS there is a resistive part in its pathway. It can be notably lower if you have increased resistance on the PCB trace but still the remote sense would increase the VRM’s output to compensate for this so you won’t be far away. BUT, given that there is a feedback loop which isn’t affected by the main line’s resistance, since I am sure there is a sense line, you cannot know from where the mainboard’s sensors take their readings or if the mainboard’s voltage points are connected to it. In my case the mainboard’s sensors where close to my readings. The best would be to be able to get readings directly under the socket of course, but I don’t want to do this for several reasons.
Hello, So iff i dont use expo on 7950x, i shouldnt be fear of burnout? This is my future upgrade (or 9000x)… Thanks.
I’ve just check and bios f10d is no longer listed. Only f9 is listed.
Yes, talked to GBT this morning about this. They will release F10 final, which basically is the same with F10d
Where did Gigabyte claim that they have limited VSoC to 1.3V in F10d BIOS update? You mean “5. Follow AMD guidance to limit max. SOC voltage to 1.3V for Ryzen 7000 series X3D and non-X3D CPUs.”? No, that’s not it. It means “YOU SHOULD follow” the AMD’s guidelines and set the limits your self, because they have not.
What was the voltage read from the board sensors/hwinfo64? Did the board sensors match your sensors? Or were they inaccurate so the BIOS was sending what it thought was appropriate voltage but wasn’t?
Yes it was almost similar to what I got from my sensors.
Yes, this voltage points are a little differents than direct contacts at CPU, so basically seems fine. Also PRIME95 is not case, who people are using on daiyl basis or often for stability testing (its most crazy stress program ever, modern CPUs from Intel donwclocking turbo ratios with PRIME95 a lot)
When you design something, you always take the WORST case scenario in mind.
I appreciate the sentiment, but you’re talking about how things ought to be designed in theory vs how they are designed in practice. Sadly, nothing actually makes it to production with that ethic in tact. Not even nuclear power plants, airplanes, or space shuttles are engineered to handle the worst case scenario.
true
Furmark should not be messing with voltage limiters and I highly doubt it is, it’s just a benchmark program and not a system tuning program like Ryzen Master, MSI Afterburner etc. therefore the voltage limiter is not limiting the CPU to 1.3v on the vSOC when it comes to any software running on the PC.
This is assuming Gigabyte aren’t doing an Asus by having a special setting in the BIOS that can be configured to “optimize” the system for certain benchmark programs (which can in theory be designed to bypass limitations).
@crmaris did you check if such a setting exists in the Gigabyte BIOS and disabled it if found?
Nope I didn’t tune it for a specific benchmark. I just did what 99% of users will do, enter BIOS and enable EXPO profile leaving all rest in auto.
What is the board/hwinfo64 reporting for the voltages? Does the board know it is providing the CPU with too much voltage or are its sensors wrong which tricks the BIOS into thinking it’s doing its job and limiting the voltage correctly?
HWinfo’s readings were inline with what I got from the labjack.
Are truly only X670E motherboards impacted here, or are B650 experiencing the same problems (perhaps in lower amounts so taking longer for the damage to become apparent)? It’s a bit of a minefield currently to try and put together a new system.
I don’t know of this, sorry, because I don’t have many mainboards to check. Everything I have bought it on my own.
the voltage read points on most motherboards are connected to the power plane before the CPU socket so you get higher measurements than what the CPU actually gets. This is even more of a problem if you don’t use a GND connection near the CPU.
Thank you for your input! I am sorry but I don’t agree on this: GND is the same anywhere in the circuit. Unless you mean reference voltage because in DC you basically measure voltage difference between two different voltage levels. Still, VRMs have the same reference (zero voltage) as what the PSU provides.
On the second part I agree with you:
When the load is applied, wherever in the line AFTER the VRM you measure it, you will get the same voltage, UNLESS there is a resistive part in its pathway. It can be notably lower if you have increased resistance on the PCB trace but still the remote sense would increase the VRM’s output to compensate for this so you won’t be far away. BUT, given that there is a feedback loop which isn’t affected by the main line’s resistance, since I am sure there is a sense line, you cannot know from where the mainboard’s sensors take their readings or if the mainboard’s voltage points are connected to it. In my case the mainboard’s sensors where close to my readings. The best would be to be able to get readings directly under the socket of course, but I don’t want to do this for several reasons.
https://www.youtube.com/watch?v=l8r4LVV_jsQ