ryzen OC / undervolting

[smoothnobody]

been chasing my tail trying to figure out how ryzen works. ryzen master and HWinfo not showing the same results is confusing enough and still don't know which to believe. wanted to believe ryzen master since its AMD's program, but having doubts after a few OC attempts. ryzen master tested 4.5ghz @ 1.32v stable. but when i manually used these settings in bios and stress tested with aida64 it crashed all the way up to 1.4ghz. i'm hearing about ryzen master clock stretching and you can't believe what it shows you. also hearing about people making claims that reducing voltage reduces temps and increases boosting but that can be misleading cause even though monitoring programs show you what you want to see if you benchmark your performance actually goes down. watched a video from steve at gamers nexus that confirmed this and basically said we don't know how these chips work yet and you can't believe what you see. that video is over a year old. have we figured out how these things work yet? will OC or undervolting actually give me better results or should i just leave it stock and let amd's secret sauce do its thing?

[Cyracus]

Make sure hwinfo is up to date, softwares needed an update to give accurate info from ryzen 3000.
After a lot of fiddling with ryzens auto boosting, I wasn't happy with voltage hitting 1.45v in single/lightly threaded tasks (I heard this was too high and wanted it lower for peace of mind), just under 4GHz peak load clocks, set it to throttle at 80, low of 3.7GHz. Could get a -0.05v offset (lower crashed), no change in voltage reporting so gave up on that.

I started experimenting at 3.8 1.75v, it didn't like 3.9 there. Bumping it up one step for me to 1.1875v I got an all core 4.175GHz, max temp of 76 in testing (except p95 which I stopped when cpu temp hit 90). 4.2 required a larger increase in voltage, performance was worse in testing (cb15/20 and blended bmw)
Overall results, higher clocks (single and multi-threaded) and lower voltage and temps, granted all this is with a small cooler in a tight box.

If you want a quick dirty OC and you have the cooling to handle it, set voltage to 1.3 and try 4.3GHz, if it works raise clocks until it doesnt, then try lowering voltage the same way.

Spoiler

I saw a claim somewhere that someone at tsmc said for their 7nm 1.3v is their recommendation, Idk, I can't find any source or alternate info, regardless no longer my problem 1.2 gotta be safe