Hello everyone! I started my undervolting journey on Ryzen a while ago. Now I got a bit more time go do more testing and trying out stuff.
Right now I’m at -0.06~ V, just to stay safe. I keep checking the Cinebench scores and everything looks okay. I can’t test for thermals because I can’t keep a constant temperature in my room.
Now comes the weird stuff: ever since trying to undervolt the idle voltage is not ever dipping below 1V. Also, from the HWiNFO64 logs looks like I’m getting an higher voltage during Cinebench runs. Before undervolting I was at 1.24V and now it’s more like 1.3V. To top it all off the peak voltage during normal use is still 1.48 - 1.494V.
I don’t have a clue of what’s going on, literally zero. I thought that having AI Suite 3 installed (I used it to control the fan speed) was overwriting my BIOS settings so I uninstalled it and rebooted two times. The BIOS is reporting a lower voltage in the right column while browsing it.
How do I go about testing properly what I’m doing? Is it really effective what I’m doing (just negative offset into the BIOS + XMP enabled, nothing else changed)?
One comment I’ve seen pop up a few times is to change max processor state in Windows power management to 99%, and you won’t see such high voltages or voltage spikes.
Having said that, there’s also this post from AMD saying that 1.5V is within operating voltage for these chips.
I saw that post, red it many times trying to extract more information regarding my findings (?). But still this only explains the voltage peaks, which are due to the system “waking up” cores, if I understood the matter correctly.
This still doesen’t explain why am I seeing higher voltages during Cinebench. But, even if voltages are higher, I don’t see temps doing much worse than they were when I wasn’t trying to undervolt it.
It’s really an headache trying to mess with these highly unpredictible CPUs.
Yeah not really weird that single core turbo boost hits that with a good cooling solution.
It also kinda depends on the said motherboard as well.
The voltages reported in HWinfo etc are often a couple of mV’s off, depending on where the super io chip is located on the board.
With most boards the super io chip does not talk directly to the pwm controller.
But rather get its measurements from the pcb, which has some resistance.
Thats why LLC is a thing.
Looks like pretty normal behavior to me.
Higher end Asus boards are generally pretty accurate.
I wouldn’t be too concerned about it really, unless your temps are going trough the roof.
I would say run prime95 small fft’s and see what the all core voltage is sitting at during a 30 minute test or so.
And monitor temperatures, i mean a single core boost voltage spikes of 1.48V / 1.49V
should not be a real issue.
I believe AMD themselves also stated something like that.
But i get what you mean, voltages on Ryzen are a bit weird.
I could run Prime95 just for a bit because my ambient temp is 26°C, no AC on and the CPU is cooled by a Noctua NH-L9x65 in a Silverstone RVZ03 so it’s not under ideal circumstances.
I recorded a maximum of 1.3V under full stress test, no AVX instructions (those would’ve made my CPU shoot up to 90°C and above even faster).
Exactly what I recorded during a Cinebench run. I guess I should try to revert everything back to stock and see if the voltage is higher or lower during heavy workloads.
Is it safe running at 1.3 volts fluctuating all core voltage? It’s more like 1.28-1.26V.
I also have a question: do you know if AI Suite 3 overwrites the BIOS settings? I was using it to control my fans and I would really want to keep using it.
I’ve never had any issues with it. It’s just that, in the beginning, I was worried that using AI Suite was screwing with the changes I made inside the BIOS. Thanks for all the answers, at least I’m getting somewhere now!
If anyone else can instruct me more regarding Ryzen and it’s behaviour with my setup I’d greatly appreciate it.
Unfortunately, as much as MisteryAngel helped me out, I feel like I need more notions and feedback from people that really like to tune these processors.
Just did a test run and, as I recorded previously, the load voltage for all core workloads is higher when I’m using the negative offset. In Cinebench I recorded around 1.25V without negative offset.
The AMD engineers did some black magic shit on these CPUs!
P.S. the negative offset made me score around 50 point higher consistently. The more I lowered it, the more the score was consistently up.
So here are my findings so far and the solution to this topic:
The correct value to monitor when making any voltage change in BIOS or software is CPU Core Voltage (SVI2 TFN) in HWiNFO64.
Extensive testing in EVERY application used is needed. I didn’t have any issues with any application I use day to day but World of Tanks that crashes my graphics drivers passed -0.05V.
Before running AI Suite 3 is necessary to take note of all the changes made in BIOS because the auto calibration the program needs to work will override everything (beside XMP settings)
This is basically everything I know for right now. I don’t know if I can make the better undervolt stick by doing anything but that’s about it.
