[SOLVED] Vcore offset not working?

[Deleted member 2720853]

I have a Ryzen 9 3900X, B550 Aorus Pro on the latest BIOS update. AGESA 1.0.8.1, Windows 10 2004. Ryzen Balanced power plan. AMD Chipset software and Ryzen Master installed.

I want to set a -.1 offset on my voltage in the BIOS. I did so by setting "CPU Vcore" to "Normal" so that the "Dynamic Vcore" option would not be greyed out. Then I set Dynamic Vcore to -0.102V.

I am now in the OS, I see the exact same voltages while benching as I did before.

Why is this not working and how do I undervolt my Ryzen?

 

[Lutfij]

Perhaps this thread might help.

 

[Deleted member 2720853]

Perhaps this thread might help.

So what I gathered from that thread is that I should enable PBO and set the thermal limit to like 85C?

 

[Deleted member 2720853]

Update: My Vcore offset seems to be working, at least I think.

With the offset unchanged, I get 4.02 all core in Cinebench, with a temperature of almost 80C instantly.

With the offset in place, I see temps decrease by at least 10C, with all core boost clocks higher, even reaching 4.1 GHz.

But Ryzen Master and HWiNFO report identical voltages while benching, no matter the setting on the offset.

Should I just ignore reported voltages and is my offset actually working?

 

[drea.drechsler]

Update: My Vcore offset seems to be working, at least I think.

With the offset unchanged, I get 4.02 all core in Cinebench, with a temperature of almost 80C instantly.

With the offset in place, I see temps decrease by at least 10C, with all core boost clocks higher, even reaching 4.1 GHz.

But Ryzen Master and HWiNFO report identical voltages while benching, no matter the setting on the offset.

Should I just ignore reported voltages and is my offset actually working?

VCore is a constantly changing thing so trying to figure out what it is from watching the bouncing values is pretty hard. The thing to do is use HWInfo and look at the VCore AVERAGE value over a period of time. Compare that doing the same sort of things (a constant, all-core heavy stress test is especially good for this) and see if it's changing any with an offset value. Click the little 'clock' button to restart the averaging period as needed.

Also look at "CPU Core Voltage (SVI2 TFN)" in HWInfo. That's the TRUE core voltage that the CPU reports back. The "VCore" reading is the output of the VRM before VDroop. So it can vary a lot, or very little, depending on load and how much LLC you've dialed in.

The improvement in CB20 score is a typical effect of running the processor cooler, which the lowered voltage offset is apparently helping with. That's good, but make sure SINGLE thread score isn't being hurt by pulling the core voltage back too much. During light loads, the processor won't boost as high if it doesn't see sufficient voltage headroom.

 

[Deleted member 2720853]

Okay, I figured everything out. My Vcore offset WAS being applied, however the CPU did not boost past 4.3, as expected. Temps were indeed lower.

What I ended up doing was finding out that the reason why my PC was somehow spiking to 1.55V and crashing on a manual OC, was because I set the values higher than what ClockTuner told me. It told me to set first die on 4350 and second on 4200. I got greedy and set first on 4375 and second on 4225.

It ran fine in games, didn't crash during normal usage. But as soon as I started an all-core workload and let it sit there for a while, it began either crashing or rebooting. Prime95 Small FTT Torture Test (without AVX) worker threads were stopping.

So I figured I'd bump the frequencies down to what CTR told me, along with a fixed voltage of 1.25V. Now, Prime95 threads aren't stopping, it's fully stable. Ran an hour of Cinebench R20 and there were no reboots. Cinebench score is about 7550. Temps are even lower due to fixed voltage of 1.25V.

I am going to keep it like this, I really don't care about my single thread performance suffering a little.