I installed the latest Ryzen drivers 17.30, with the "Ryzen Balanced Power Plan" for Windows 10.

Whilst I haven't started getting all of the content software installed, the system doesn't feel sluggish - why would it 32GB RAM and 8x cores at 3.4GHz?

Anyway, I put the 'tweaked' plan on and HWinfo, Ryzen Master and anything that I could see all indicated maxed out cores.

Sure, OK, a screw up in the reporting perhaps.  It's not really going that quick, the temps aren't climbing, the fans are going nuts.

But importantly the voltages looked like they bouncing u to 1.45v on vCore.  

Under the regular stock balanced plan, vCore is reported in the 0.87v region - this is idle, browsers open, monitoring tools, nothing taxing.

Playing with the Power Saver plan, things look even nicer in terms of vCore bouncing to 0.84 but dropping to ~0.7.

Just what is going on here?

If anyone has a guide to keeping things stock and dropping voltages, I'd like to hunt for longevity and low temps as much as possible here.  I guess it's mostly follow the OC guides, omitting the chip-thrashing.

Thanks for the info.  Sorry for perhaps being dense here.

Here is a screen grab of HWinfo readout.

Which vCore are we supposed to be looking at?  OpenHardwareMonitor, which I know cannot access the CPU Tdie, can only provide the vCore data at the bottom, off of the motherboard(?)

UEFI is all set to AUTO everything.

These results are just idling with Win10 x64 Creators power plan set to" PowerSaver" - "AMD Ryzen Balanced" gives me the willies seeing all the stats flat out.  I know what they say, but the temps do actually increase on that setting.

The reason for the 1.475v is because I had run Prime95 on 2 threads.  That caused the 3.9GHz on those two threads and thus the added voltage.  When all cores are maxed (16 threads) in Prime95 the vCores all sit at 1.2v or just a hair under.

Additionally, I set the UEFI vCore to 1.2v LL3 and SoC to 1.0v LL1 and everything was fine until I did a 2 thread test on Prime95, then the screen went black, but I couldn't get back to UEFI without holding the case power button.  I've a feeling I have found that my CPU won't run on that low voltage when boosting 2 threads.  Should that be surprising, probably not.

Annoyingly when I did set the vCore manually in EUFI it did not drop back down when the CPU wasn't being used, but rather stayed at that same voltage all of the time.

Again, I apologise that I am probably making you folks roll eyes here.  I'm having a tough time finding a concise and easy to follow guide on overclocking that I can apply to simply under volting.  It seems I can find different settings but there isn't so much explanation behind them or what to expect - eg. the vCore becoming static.

Admittedly, I did this on the old C2D system and the vCore as static, except for droop on load, hence LLC or whatever ASUS called it then.

Thanks as always.

Muchas Gracias!

Yup, I used fixed mode in UEFI.  I didn't quite understand how offset would work.  I'll go read a bit more about that and then poke around in that again.

I picked 1.2v based on what HWinfo was showing with all cores working.  I didn't really think about the high-boosted low-thread count work.  As such it totally failed at that point.

I wasn't using the C2D as a reference, no for a minute, entirely different silicon!

AMD Ryzen Balanced power plan = 90% min CPU state, and a high vCore.  Well that totally explains why it was a bit warmer than regular Balanced or Power Saver options!

vCore from the motherboard sensor, that's the one at the bottom of the image, and the one that shows in OpenHardwareMonitor, SpeedFan, CPU-Z and most other software.  HWInfo just gives a more comprehensive set of data as it taps into the CPU too.  But we look at the motherboard sensor for this, cool, that clears things up some.

Yes, vCore was on Auto in UEFI, it does a lot of jumping around.  Normally sits around 0.848v when doing something trivial like web browsing, but is constantly moving below this.

So far so good, your detective skills are very sharp.  You knew what I had done here from the data I have provided.

OK, so UEFI vCore to 1.35v using offset mode, not fixed (I think this might be a value I have found AMD recommend for long term use).

I'll follow your suggestions and report back.

I'll leave SoC on auto for now, as it's frankly not gone above 1.0v

vCore (Motherboard) 0.384v - 1.456v

Idle vCore sits around 0.544 more often now.

The NH-U14s did it's job in keeping the Tdie below 60°C (~75-80% fan speed).

vCore was around the 1.2v again, slightly over, with all cores.

The high vCore is when taxing the two threads only.

The first + option after Auto on the Offset mode is 0.00625

This seems a strange number being that 1.35+0.00625 is 1.35625 not 1.456, so I wonder how I attempt math on that.

So now, I just reduce the main 1.35v to the next lowest, and so on, until things become unstable, then bump them back up a little.

Thanks, this does make some sense to me.

I had to reset the power plans to default which has removed the Ryzen one.  All of the cores clocks were pegged in HWinfo and I just couldn't seem to find a way to make them not be.  OK, I appreciate this maybe a cosmetic issue, but it was also not really reflecting vCore movement.

Open cmd.exe as admin and type: powercfg.exe -restoredefaultschemes

That'll reset all of the power plans to defaults, including the drive sleeping and such.  "AMD Ryzen Balanced" will be gone.

I have left everything at auto again, NOT setting CPU to 3400 and Core to 1.35000.

I have just change the voltage mode to offset and hit - key 4 times.  It's now -0.02500.

The vCore doesn't go above 1.408v with LLC3, and sits mostly at 1.392v.  It's a matter of tweaking and this will come down more.

When hitting at 16 threads with Prime95 it's still at vCore 1.2v

Loweset measured vCore is 0.352v

So, thanks Parsec.  I'll poke about some more, I think I know what I am doing now.

So you are saying you have vCore in EUFI set to "Fixed", but it's dropping the voltage lower when not doing anything?  That's definitely not my experience with using the fixed setting - parsec seems to agree that fixed means just that, static, unchanging no matter what the load.  Obviously under heavy stress it could 'droop' which is where more aggressive LLC (lower numbers) comes in.

I notice your post here:

http://forum.asrock.com/forum_posts.asp?TID=6175&KW=&PID=36821&title=x370-taichi-oc-tips-for-begginer#36821

datonyb wrote: ok so heres the full settings im running now  in p states first and second section 9c 8 28 this is 3.9ghz and 1.3 volts for some reason third box is set to  7c 10 6c this translates apprently to 1550 mhz and .875 volts (i know i didnt set this)   i then on oc tweaker leave everything to auto apart from cpu volts =overclock mode cpu core = fixed cpu volts = auto llc level 2 soc = fixed soc volts 1.100 llc = level 2 in advanced tab cool and quiet off c6 off global c state is on auto this is working like a dream for me on bios v3.1 max volts never above 1.3v and auto downvolts down to silly low figures without issue on idle

But you are using P-state options, which is why I think the 1.3v is coming in.  The rest is all based off of that.

Without fiddling with that stuff yet, I think by default vCore is going to be higher when 2 threads are stressed.  I am playing with offset negative in hopes of reducing the rather high short-term vCore when single threaded tasks occur.

I've also got to work on seeing if that buried Fan control hysteresis will actually work.  So far it did nothing.  It's the one in the Advanced tab > AMD CBS > (?)

OK, so it might be worth me just going to v3.10 if not for the potential for it to down volt itself, but to get the fan control options as well.

Mine is sitting around 23°C Tdie on idle and 25.5°C T-socket on idle.  NH-U14s A14 is on 20% duty ~350rpm, Fractal GP-14s on medium setting (7v).

I'll take a gander at techcity and see what notes I can take.

Thanks for the offer to chat.  I may take you up on that if I get real stuck.  It'll be a message sent from the RPi crying "Help" when something won't turn on :D

Aside from these tweaks, every morning I put this machine on I have a great big smile on my face as it boots quicker than I can put a T-shirt on, and is taking everything I can throw at it, and quietly!

I thought I'd have another little play around.

Dropped Fixed Voltage to 1.325 (minus 4).

CPU MHz still drop to ~2.2GHz and boost up to XFR speed to ~3.9GHz, so all is working there.

I think I still need a little clarification on just what numbers I should be monitoring:

Individual Core VID I think might be a bit bogus - well 1.55v is shocking if not!

We know that the frequency isn't strictly true either, as when using "Ryzen Balanced" powerplan, the frequency is pegged at max always.

The bottom screen grab shows two lots of info that might be at least believable.

However, who's numbers are correct?  I know the 'Tdie' is a real sensor on the chip, and the 'CPU' temp (lower) is the Tsocket sensor in the motherboard.

But regarding voltages?  I set it to 1.325 LL2, so they are both quite close.

My temps are still nicely low, and I don't see 1.4v+ any longer, so that's good.  I haven't yet ventured into P-states, but could do so.

Maybe I should check out UEFI 3.1 and see if the vCore drops off more when idle, as this is pretty static, even though the low temp shows that it's clearly not burning the chip up.

I found that if I edit the voltage lower down the page it makes a difference, whilst the upper does not.

Maybe I need to jump to p3.10.

I am still very impressed at just what this system is capable of.  It's mildly annoying that the router and modem take longer to start than this PC gets to the Windows desktop, I never expected that to be the case.

Note: These are not actual screen shots of what I have edited.

In the top image I have it set to Auto.

In the bottom image I have it set to Fixed mode, 1.325v Level 2 for both LLC options.

It looks like there is still some playing around here to get the most out of it.  AS far as I am concerned, I am totally happy with the stock performance and have no need to disable that.  But why have extra volts running through it when they aren't needed?

- The CPU vCore will spike to even 1.5v when XFR turbo boosting.

- Running on dynamic vCore with these transient bursts is safer than a higher vCore (1.35v) permanently.

However, I still see that these "transient bursts" are not so transient when running stress tests like Prime95 on 1 or 2 cores.  Maybe the 1.495v might be a spike, but it's still not coming down from 1.395v - which is basically 1.4v in my book.

When all threads are being stressed vCore drops down to just under 1.2v with the default "Auto" settings.

Fixing 1.3v on the vCore in EUFI stops the chip from pulling anything else than 1.3v.  XFR still worked perfectly on my chip at least.

So, what I am aiming for is limiting vCore to 1.3v max draw, as it has zero need to go above that.  I may be able to adjust this in the P-state settings, which is my next point of call.  However the XFR boost frequency is hidden as a ghost P-state and governed by the on chip system management.

For us that don't want to overclock but instead want to achieve a best possible optimised efficiency / longevity setup are we stuck between a hard place and a rock that has just been chucked down?  I shall investigate more.

Right, political waffle and speculation aside; the SMU!

Apparently, and I have no way to test, the power management on the actual chip requests # volts but may actually use something else.  Does the SMU get damaged by pulling too many volts, what is it's safe limit?  The indication is that the SMU provides safe voltage (1.365v?) internally where needed.

Now, when one ants to overclock, the SMU is gonna pull and keep pulling more power in and distribute it on a less than personally optimised graph.  Why is it not optimised?  Because it's meant for normal operations, not overclocking.  So, when overclocking it's back to the old manual affair of setting up your own voltages.

Still I think I will play with P-states to achieve an optimal low idle voltage and not too high load voltage.  It'd be nice to set up 2-core and all core volts but we'll see.