[SOLVED] Ryzen 5 3600 OC - idle, load voltage, LLC

[Milicek]

Hi. I probably have a release batch sample so don't freak out because my OC is not good - new samples seem a lot better binned.

I used 4.1 Ghz all core on 1.325V for a year, but now I started digging into it a bit more....

I wanted to ask, because i haven't found a clear answer for that. Is it a good thing to have idle voltage lower than the load voltage?

I achieved this by LLC, I set VCore voltage as 1.3V, LLC mode 2 (MSI B450 Tomahawk MAX)

when idling, i'm on 1.304V, In low-medium load I sit on 1.312V-1.320V and in high/extreme load I sit on 1.320V.

I'm just asking if thats a good thing or not, because I don't know how to test if 1.304V is enough for lite loads // single core operations. For middle-extreme loads I have stability, tested that with OCCT. Is 1.304V idle, 1.320V in load better than 1.325V 24/7 in manual OC, or not?

I've just seen people trying to counter Vdroop, but its the absolute opposite of this ;D (for example 1.320V idle, 1.304V load)

 

[drea.drechsler]

I've also a launch 3700X and it's pretty bad too. They're binning much better now...if you can find a recent build.

When set up stock Ryzen can very much deal with a lot of VDroop as it needs higher voltage to hit the high boosts (up to 4.4Ghz spec'd for a 3700X, up to 4.2Ghz spec for a 3600) in light threaded workloads. So you're also losing up to 100Mhz of lightly threaded performance by locking in on a fixed 4.1Ghz all-core overclock, by the way.

But pushing really high fixed overclocks with mature process CPU's it's highly desireable to have rock stable voltage. Some have gotten up to 4.5Ghz clocks (3800x) with good samples at 1.28V or less. That low voltage is to prevent degradation of their CPU. But with voltage that low any VDroop at all is fatal when hitting a heavy all-core processing load. It's also pretty important to do it with limited LLC as the roll-on overshoot and roll-off undershoot is equally fatal to stability, especially for prime95's FFT transitions.

But also important is to know which voltage you're looking at. VCore voltage is at a sense point somewhere after the VRM, so it's before all the VDroop has happened. The STI2 TFN Core Voltage reading (reported by HWInfo64) is reported by the CPU so it's the internal voltage the cores are actually seeing.

All that said, I simply can't get my 3700X stable above 4.2Ghz at anything close to a reasonable voltage...and I'm talking only Cinebench stable, not even Prime95 no-AVX. But, when I run it stock with a tweaked PBO it's just as fast as a 3800X in both light threaded and heavy threaded benches. That's because it still has three cores that regularly boost to 4.4Ghz and that's all that needed to do light threaded work. Yet all the cores working together still hold 4.3-4.35Ghz, dropping to 4.2Ghz as the CPU heats up. So needless to say, all-core overclocking isn't a good option.

 

[Milicek]

Well I'm kinda confused by your answer because i can't tell if there is a clear one.

But to explain why I use 4.1 manual OC, on stock settings my Ryzen was having some voltage spikes that I just didn't like, also in multithreaded things it boosted only up to 3.95Ghz or 4.0Ghz I believe.

My GPU is GTX 1060 so I probably won't notice a difference between 4.1 and 4.2 in games, but in multithreaded operations I will ;D

If I remember from release, they recommended 1.325V-1.35V as a safe voltage for Ryzen 5 3600, so I never went higher.

I can't do 4.2Ghz 1.35V stable in OCCT, in games its probably fine.

That is not the point tho. I ask if lower voltage in idle and pushing higher voltage in load is better than having 1 solid voltage, which is higher than both of these two.. (1.325V solid, or 1.304V idle, 1.320V in load).

I take all of them as a safe voltage, its just a matter of few 0.%% that i'm curious about.

 

[Milicek]

PBO never actually worked decent for me. Could you tell me what settings you've changed for PBO to work somehow decent?

 

[drea.drechsler]

PBO never actually worked decent for me. Could you tell me what settings you've changed for PBO to work somehow decent?

Simply put: set up stock it's desireable to have VDroop to work with the boosting algorithm as it lowers volts when it gets hot. But when set up in a fixed clock/fixed voltage overclock it's desireable to have a steady a voltage since you want the lowest voltage possible to be stable at all loads, especially the highest ones, without degrading the CPU.

But if you have a choice, higher voltage in idle and lower voltage in heavy load is more desireable than the other way around since you can adjust the voltage so it's stable at heavy load. But you have to know where your sense point is to know what you're really seeing, and be very careful just how high the voltage is at idle.

The way I tweak PBO is taking advantage of a 'bug' in the way the boosting algorithm works for ryzen 3000 (maybe 5000, maybe not). What you do is turn on PBO, to manual, then dial in PPT of 330, TDC of 230 and EDC of 10. Yes, 10, it's apparently undefined in the algorithm so it just lets it hit highest possible boost clocks. It still observes PPT and TDC limits, that's why to set them high but you can lower them if you want to use it to reduce thermal output in heavy work loads. Pumping up scalar to 10x helps it hold boosts longer in heavier work loads.

It seems to messes up it's C-State algorithm though so I have to turn off global C-states. When you do that, be sure to set core voltage to AUTO. It doesn't work for all CPU's, but it definitely does for my launch 3700X CPU. Run light threaded and heavy threaded benches (CB20) to be sure it's working right. 6 core CPU's may want it at EDC=8, so also play around a bit. It can't hurt LOL.

EDIT: here's an overclockers thread that discusses it at length:

EDC = 1, PBO TURBO BOOST

www.overclock.net