I saw this behaviour only on this Asrock board. On my previous AMD builds I used Gigabyte and Asus, but may be this is a general Ryzen thing so may be all boards for this generation AMD cpus suffer from the same.

Another weird thing I can see on these Ryzens are temp spikes under load. You run, in example, the AIDA64 stress test, and after a while the cpu temp settles at 57°, and from time to time HWINFO64 registers temp spikes up to 65° in a very short timeframe. 

If you want to set manual voltages and also get the power savings, use the pstate overclock method with offset voltages, adjust LLC (I suggest LLC 3 as max. If you need more LLC it's because you set a weak vcore from the start) and the Ryzen balanced power plan on Windows.

Cheers. 

Anyway, it's not related to your question so I'm sorry to hijack your post.

Cheers.

1.4+v with all cores going in CineBench...hmmm.

To be honest, on stock/auto it's around 1.25v for all-core voltage on 1700X on the TaiChi.  In my case it'll happily do all core loads on just under 1.10v!

What is causing these spikes? It's XFR (Extended Frequency Range) which is a feature of Ryzen which is ONLY enabled when running at stock (i.e. Auto settings). XFR monitors your CPU thousands of times per seconds and dynamically adjusts the overclock and voltage on the fly.

BUT BUT BUT 1.5v WILL DEGRADE RYZEN!!!!!1111

This is correct - however when you see these large voltage spikes they are in fact not over all cores/the whole chip. When your Ryzen CPU boosts to its maximum performance it will bump the voltage momentarily over 1 or 2 cores and it is NOT the same as running high voltage over the whole CPU. The peak voltage is usually when the CPU comes 'off load', are for a split second and WILL NOT damage your chip.

If you overclock your Ryzen CPU/lock your voltage in any way, XFR is disabled which is why you no longer see the voltage spikes. Of course overclocking will net you more performance, but the Auto/Stock settings are SAFE.

This is not an Asrock bug, it's a Ryzen feature, it happens on both my Asrock Taichi and Asus Crosshair VI.

AMD know more than any one of us about what is/isn't safe for their chips and they wouldn't run this way otherwise.

From what I read, I am in agreement: Max vCore of ~1.55v under single core XFR boost is by design.

What Gigabyte did when they killed a bunch of chips with high vCore ~1.70v, I do not know.  Was it on auto mode?

What I can say is that from P3.00 to P3.10 on the TaiChi went from 1.55v to 1.50v max. vCore under XFR single-core.

You say "lock your voltage" = disable XFR.  That's not true.  As long as you don't change clock speeds or actively disable CPB, then you can indeed lock or offset the vCore and XFR will still work fine.  

I set my vCore to fixed 1.30v and XFR worked fine but vCore SVI2 TFN did not scale down when idle.

Offset mode is working for me with -0.10v and forcing LLC to L4: I see a max. of 1.369v and it drops under idle.

It is a fact that voltage is always kept on the higher than optimum setting so that all chips that leave the factory will work at their advertised max. specs.  For the vast majority, the chips will work below the auto voltages.  It's been the same for Intel, AMD and anyone else.

My old Intel Core2Duo used more volts than necessary, and I could overclock 25% on less than the stock auto max. volts!

One thing that seems to be true is that these modern CPUs are very hard to read data from in the OS, and specifically Windows.

They are self-regulating and so are doing things at rates the OS beyond what the OS can report.  I guess results are sort of averaged out, or a mini-snapshot of the real truth.

The on-chip SMU may request a vCore, may or may not actually get it, and may distribute less than this voltage to the internal parts of the chip (core, cache etc.).  I have a feeling much of the SMU's control is removed when the clock frequency is altered and the chip is made 'dumb' and will stick to distributing voltage by a table or by whatever comes in - in the case of overclocked frequency, whatever the vCore in BIOS is will go to the cores.

It's still a little annoying that AMD aren't just completely clearing this up once and for all.  Of course, they don't want to give out too much info on how their intellectual property works, but this is just general user information.

If they have stated that with overclocking don't go over a max of 1.45v, then it looks very bad to the uneducated when they see 1.55v and they aren't even overclocking!

C'mon AMD, educate us :)

The reason for the max vCore ceiling when overclocking is because all cores will be running hard pulling an awful lot of current.  A high vCore and all that current means an extreme stress on the CPU.

With respect to our (or my) issue, single-core CorePerformanceBoost and XFR, the seemingly high vCore spikes to ~1.55v and sustained <1.40v aren't an issue due to such low current draw when the other cores are idle.

You can actually watch the vCore drop, and XFR temporarily disable if you fire up a CPU heat stress like Prime95 on 1 or 4 threads (1-2 cores) and throw some other workload at other cores.

Leave a 1-4 thread stress test running and watch the one or two core clocks hit their max CPB+XFR speeds.

Launch another task with affinity set to another core, and then watch the vCore drop and the clock speed on those cores running the stress test drop to the all-core max+XFR frequency.

The 1700X has the following modes when stock/not over-clocked/clock-locked:

1-Core to 2-Core 3.8GHz + 100MHz (XFR) ~1.55v max.

3-Core to 8-Core 3.4GHz + 100MHz (XFR) ~1.3v max.

1-Core to 8-Core 2.2GHz idle ~0.8v

I can't honestly believe that seeing a 1.55v vCore on stock/auto settings is bad.  To many people have reported it and AMD have said it's normal.

However, the annoyance is that they haven't drawn a graphic or written a document to literally state the normal working ranges, and then to also state what is advisable for different overclocking scenarios.

I've seen 1.35v and 1.45v as two possible ceilings for overclocks (assuming all-core and 24/7 fixed vCore).  Now waht about if someone wanted to (not sure why) make their 8-core a 4-core by disabling cores and overclock those?  Is it just under all-core that those numbers are ceilings or under any circumstance?  If it's under any circumstance, then the question comes back to - what the heck is going on in single/dual-core XFR on stock?

AMD could silence a lot of armchair engineers if they just published that info.  They don't even have to go that nitty-gritty, just to say something like:

"Changing clock speed and/or disabling cores will disable some of the functionality of the on-chip power management.  For these reasons, we suggest a safe max. of 1.45v fixed vCore, ideally a 1.35v or under for reasons of prolonged longevity.  In automatic stock configuration, the on-chip power management is in fully enabled. A vCore up to 1.55v may be observed, this is normal behaviour as the actual cores will see less voltage."

But nope, and their own Ryzen Master uses the highest core VID as it's value for vCore [which is the requested voltage from the core, as far as we know, and not the actual voltage - kinda useless data].  So really that'd a lot of use!

Whilst I will continue to wish for a decent answer and explanation, I do feel that I am happy with my system running as it does with the -100mV offset.