Overclocking and performance numbers don’t add up here.
Overclocking and performance numbers don’t add up here.
I thought this adjustment applies mainly to Ryzen 3000 models. I saw the same effect on my 5900X during my research. It seems related to clock stretching. Keep this option off for optimal results and experiment with PBO plus curve optimizer (using a negative offset, but be cautious of WHEA errors especially at lower voltages or when the CPU isn't heavily loaded). Make sure to try curve optimizer and also think about reducing power to help maintain cooler temperatures, which might boost performance.
Based on my observations, what you're seeing is typical unstable OC behavior where Ryzen can run high clock speeds without problems but then throttles for stability. This happened often when I was tuning the 3900x. Running it at 4.2 all cores with a 1.2 Vcore gave great performance and high scores, while 4.25 all cores still worked for benchmarks but lower cinebench results. It makes sense to confirm your OC's stability by matching performance data. I've also noticed Ryzen Master provides more accurate clock readings than other tools.
It seems you're stretching your clock settings. I suggest Ryzen Master and enabling PBO 1. Right now it's not installed on my PC, but my Cinebench R23 score is around 14900-15000 multicore with all cores at 4.5 and temperatures staying under 73°C (using an NH-U12A cooler and setting fan speeds to 70%). Later I can share the numbers from Ryzen Master in terms of values. You're likely more familiar than me, but adjusting it for small tweaks in BIOS feels too complicated (besides my recent overclock from 3200 CL16 18 18 38 to 3600 CL 15 16 16 36 at 1.4v RAM).