The final word on idle voltages for 3rd Gen ryzen
Hi, everyone. I've spoken to many of you publicly or privately over the past 48H to better understand why you are seeing idle voltages the community considers to be high. Some of the back-and-forth was covered in this thread, but I wanted to submit my own post to bring more visibility to this topic. We have a final answer for you.
Understanding What's Going On
We have determined that many popular monitoring tools are quite aggressive in how they monitor the behavior of a core. Some of them wake every core in the system for 20ms, and do this as often as every 200ms. From the perspective of the processor firmware, this is interpreted as a workload that's asking for sustained performance from the core(s). The firmware is designed to respond to such a pattern by boosting: higher clocks, higher voltages.
The Effect of This Pattern
So, if you're sitting there staring at your monitoring tool, the tool is constantly instructing all the cores to wake up and boost. This will keep the clockspeeds high, and the corresponding voltages will be elevated to support that boost. This is a classic case of observer effect: you're expecting the tool to give valid data, but it's actually producing invalid data by virtue of how it's measuring.
What about Ryzen Balanced vs. Windows Balanced Plan?
By now, you may know that 3rd Gen Ryzen heralds the return of the Ryzen Balanced power plan (only for 3rd Gen CPUs; everyone else can use the regular ol' Windows plan). This plan specifically enables the 1ms clock selection we've been promoting as a result of CPPC2. This allows the CPU to respond more quickly to workloads, especially bursty workloads, which improves performance for you. In contrast, the default "Balanced" plan that comes with Windows is configured to a 15ms clock selection interval.
Some have noticed that switching to the Windows Balanced plan, instead of the Ryzen Balanced Plan, causes idle voltages to settle. This is because the default Balanced Plan, with 15ms intervals, comparatively instructs the processor to ignore 14 of 15 clock requests relative to the AMD plan.
So, if the monitoring tool is sitting there hammering the cores with boost requests, the default plan is just going to discard most of them. The core frequency and clock will settle to true idle values now and then. But if you run our performance-enhancing plan, the CPU is going to act on every single boost request interpreted from the monitoring tool. Voltages and clock, therefore, will go up. Observer effect in action!
Okay, Rob. Shhhhh. Just Tell Me How I See Voltages? I Just Wanna Check!
CPU-Z does an excellent job of showing you the current/true idle core voltage without observer effect. In my example image, I've configured a Ryzen 9 3900X with all the same things we would advise the public to use: Windows 10 May 2019 Update, the latest BIOS for the Crosshair VIII, and chipset driver 1.07.07 (incl. the AMD power plan). Yes, we're monitoring the behavior of the core, but we can see that idle voltage looks great. The tool is not compelling the firmware to boost when it's not needed.
Is There Anything Else I Need To Know?
Yes, actually. The Ryzen CPU depends heavily on a low-power state called cc6 sleep. In this sleep state, core clockspeeds and voltages are basically nil as the core is sleeping and gated. It is not possible to report out the state of the core in this sleep state without waking the core, probing the status, and killing the power savings of cc6. Therefore, MOST tools can only show you the last clock and voltage of the core before the core went to cc6. So if you were at full 4.5GHz+ boost @ 1.48V, then the core went to sleep, many tools might show the core(s) stuck at that value. The tool just doesn't know any better.
However, the latest version of AMD Ryzen Master can uniquely show you clocks and voltages in a cc6 state. No other tool can do it. Neat piece of info for the people looking to understand how their core behaves!
tl;dr: Observer effect bad. You can't always trust your tools. CPU-Z gives you the right idle voltage. We'll look at the rest. Thank you everyone for your reports and insight, which helped us get to the bottom of this once and for all.
