Vcore vs CPU Core Voltage (SVI2 TFN) - Ryzen 5900X 1.47v Vcore / 1.52v SVI2 TFN / 1.50v VID

[PiersJH]

Hello,

My PC is only a few weeks old (there are no old components in it) and I have recently found the CPU requesting too much voltage, and it being supplied with too much (as per AMD specification for the 5000 series). Which measurement is more accurate?

I should note, that this is with PBO enabled, BUT with motherboard limits set to stock (EDC etc. all at stock values), no 'AutoOC' and a large undervolt per core due to this issue.

This problem didn't start until a few days ago and it's the reason why I applied an undervolt.

I'm using the latest available AMD chipset drivers, latest Asus BIOS (a couple of months old), and latest Windows 11 updates - essentially, all software and firmware is up to date.

Thank you for any help.

Edit:
- First screenshot was taken after 3.5 hours of playing Civilisation VI locked to 60fps.
- Second screenshot was taken after 15 hours of using CoreCycler (tests stability of undervolts per-core using Prime95 and SSE (not AVX(2))

 

[PiersJH]

Here's another screenshot. This one shows maximum voltages and temperatures from just running Cinebench R23 single core on a loop four times (as well as having a couple of programs open). Again, the SVI2 TFN is above 1.5v.

 

[Martin]

What is your PBO Scalar? It's also shown in main HWiNFO window.

 

[PiersJH]

What is your PBO Scalar? It's also shown in main HWiNFO window.

Thank you for the reply - I truly appreciate it.

- It's set to 1x. The only thing I touched under PBO is the Curve Optimiser - everything else was left on stock values or disabled (power limits, AutoOC, etc.).

- I've just reset the BIOS again to see if maybe something went wrong last time (I can't see how, but Asus seems to have dropped quality control in recent years) and am conducting the same CBR23 SC test.

- What would you personally use to test single core for a multi-hour loop so I can leave the PC with HWiNFO running again to see if the same issue crops up? Cinebench only has a 30 minute allowance.

- I've also updated to the latest non-beta HWiNFO version, although the changelog didn't mention anything relating to this chipset and CPU.

On a different note, do you know how the CPU fan speed is calculated? I thought at first it was the H150 'Pro XT' values added up, but that comes to over 7,000 with my current fixed fan speed, yet the value in the BIOS and reported in HWiNFO is ~5,400 - but there's no fan plugged in to it.

 

[PiersJH]

Update: on stock (as explained above), with Cinebench R23 single core, it's so far showing this :



I've set it to run for 400 minutes and will update this thread when I check on it.

 

[PiersJH]

Here's an update. The SVI2 TFN went outside of acceptable limits briefly - see post above - and then I reset and enabled logging for nearly 3 hours. All of this running a constant CB R23 single core benchmark. Here's some data, including using the excellent tool made by Thomas Barth.

I can't explain why the post above has SVI2 TFN outside of specification.

 

[Martin]

I don't know why it behaves so and there are more users questioning the same.
However, the CPU has other mechanisms to protect from damage and unless you modify the PBO scalar even such voltage should not reduce the reliability.
Maybe someone else can comment on this or you seek for advise on other forums, i.e. overclock.net

 

[Reknez]

I noticed that my VID would sometimes get to 1.506 V after updating to a newer HWiNFO64 version (I do not know which one exactly it started happening at). It never did that before, but now it does it regularly, only VID though, SVI2 TFN is lower because of vdroop.

 

[PiersJH]

I noticed that my VID would sometimes get to 1.506 V after updating to a newer HWiNFO64 version (I do not know which one exactly it started happening at). It never did that before, but now it does it regularly, only VID though, SVI2 TFN is lower because of vdroop.

Interesting observation. The question for Martin is whether this is likely due to increased accuracy within HWiNFO, or whether something else is at play.

Which LLC setting are you using on your board?

@Martin I've contacted AMD and asked about SVI2 TFN (core) and they've stated the values I've reported are normal. If you are interested, I would suggest making the SVI2 TFN values for core voltage red (like telemetry) when 1.551 or higher (any value 1.550 or below is normal) - this would provide clarity. I would also add a disclaimer in the tooltip that this value can change hundreds of times per second and HWiNFO's default polling rate of 2,000ms, along with AMD's proprietary system of reading values may not provide enough resolution for an accurate overview.

What do you think?

I'm happy to privately provide the email from AMD support (I already have your email address from speaking privately with you).

 

[Martin]

I will leave marking of the values above the threshold up to the user. BTW, the CPU cannot request > 1.550 V via VID, it's the SVI2 maximum VID.

 

[PiersJH]

I will leave marking of the values above the threshold up to the user. BTW, the CPU cannot request > 1.550 V via VID, it's the SVI2 maximum VID.

Under VID (effective), it reads 1.55v, with up to 1.531v VDDCR CPU/SVI2 TFN. The VID's never read over that value as I'm aware it's the limit.