I'm even more confused now. Totally not your fault, there's just too many voltage rails on a modern CPU.
What about CPU cores? How do I make sure my board doesn't overvolt the crap out of them for the illusion of added stability?
I'm still not sure how we came onto the topic of overvolting Vcore
Short answer: it's not an issue
Long answer: it's not worth worrying about, because
- Stock Precision Boost functionality is not and has never been an issue, and in the freak event boost algorithm rampant overvolting is actually causing all these CPUs to explode, then all Ryzens post-2019 would also be caught in this net because PB has not fundamentally changed; most aggressive of all in terms of Vcore are actually the oldest CPUs (Ryzen 3000), and they haven't been dying or significantly degrading by the droves
- You just have to trust software monitoring and the fact that up-to-1.5V PB hasn't blown up CPUs in 4 years, because it would be impossible to monitor Vcore at a polling rate that approaches how fast these CPUs are actually controlled, without expensive hardware logging
As to the point that Vcore always degrades, sure, on a bare technical level. But that technical level of degradation also applies to Intel CPUs, which are notoriously hardy. Reality is that the amount of "degradation" experienced is not relevant in the timeframe of the CPU's expected lifespan...........in AMD's case,
if PB works as intended, of which there is so far no evidence to the contrary.
The Ryzen question of degradation can never be discussed with voltage alone - it is current (accelerated by high temp and volts if present) that most readily kills Ryzens. The PB algorithm is designed exactly to take this into account - stock boost behaviour will never apply high Vcore (1.35V+) under high current, which is why you will see high Vcore for light ST but gradually falling as you bring more cores/more load/more current online.
The APUs demonstrate this relationship between volts and current really well. The UMC is both AMD's best DDR4 controller and highly tolerant to VSOC, which is why you'll see record OC scores almost always pushing absolutely obscene VSOC (way north of 1.3V). Given good cooling and knowing what you're doing, benching at high VSOC still shouldn't cause significant damage, but I can guarantee that those 5700Gs aren't long for the world if you decide to start benching the iGPU under those conditions (drawing an easy 30A+ of current). When at 2D idle or when paired with a dGPU, iGPU/SOC power draw is usually in the single digits therefore current draw does not become a problem.
Because VSOC doesn't have anything like PB that naturally scales back voltage when current increases, it's easy to see why running high VSOC at high load/for extended period of time is harmful. But that doesn't apply to Vcore, unless you go out of your way to find unsafe PBO settings/try to break the algorithm (which is not hard), or run a static OC.
docs.google.com
Something must be wrong with my board on Auto settings. Even in all-core workloads, I have at least 1.3-1.35 V on the cores.
