X370 Taichi VCore nasty spikes

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.

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.