Matisse (Ryzen 3000) overclocking/undervolting

[Zach_01]

My original plan had been to install 3800X Set PBO etc and leave it well alone. The only reason I have for fiddling at all is because I saw consistent voltages right up to 1.5v and it just didn't 'feel' right.

More or less this is false thinking or feeling...
This is how ZEN2 is working and programmed by AMD. It is said alot of times that high voltage for low clock/load and eventually low power current (EDC/Ampere) does not hurting anything (PPT/Watt stays low). These chips have internal monitoring and regulation for clock/voltage in conjunction with temperature to preserve performance and longevity of silicon. Its called FIT (silicon FITness controller).
If you have it on auto for PB and PBO and increase cooling, clocks will go up for allcore loads and it will sustain the rated single core boost more (effective clock goes up on all scenarios).
Additionally If you keep power current tamed while keeping it cool clock increasing further and its more sustained.

 

[liuyc_2007]

Disclaimer: These are tests on my CPU /mobo and may or may not apply to anyone else. (R7 3700X is shoved into a cheap board - Asus Prime X370-Pro)
Also, it is Cinebench R20 only, I haven't checked for stability in other applications at any of the voltages.
Tests were done in a closed case (but with 5x 140mm fans, and boxed cooler AMD Wraith Prism RGB), in my attic where it's quite hot !

Conclusions for the wall of stuff above:

• Ryzen 7 3700X (and my guess the entire series) seems to overclock itself simply based on Core Voltage. By just changing the core voltage to different values, the CPU gets slower or faster
• The CPU worked without any crashes for the entire range, from 0.925 volts to 1.45 volts (where it was throttling). At no point it showed instability (in Cinebench R20 anyway)
• It appears to work as low as 0.925 volts, while being VERY cool (just a few degrees above minimum). Will redo power consumption tests at this insanely low voltage !
• No performance improvement all the way up to 1.275 volts (even regressions at around 1.1v-1.2v - tests were repeated)
• Going from 1.287 all the way to 1.35 seems to improve performance significantly - 4826 cb / 4092 cb is 18% )
• The CPU stopped getting faster at voltages higher than 1.35, but I will test at "near" that voltage later, to see if there's a "max sweetspot" on my system, and also test for stability in other applications around those voltages
• Max Temperature increased constantly with every voltage point, from as low as 56 to the throttling point: 95
• I'm quite convinced that with a much better and expensive cooler more performance can be extracted at around 1.375-1.4 volts, but these tests are done without spending any more $$$
• ASUS is f*cked up with their 1.4+ default "Auto" voltages, they seem to be NOT NEEDED for this CPU, and it can be made to run cooler and more efficient at lower (1.35 or so)

I hope my efforts here help you find your own voltage/performance sweet-spots.

Good luck and please share the findings !

---
One final test after finding that voltage sweet-spot:
View attachment 127189

@1.3625 volts (bios setting) - I think this is as high as it would go.
Will stick with this voltage and run other tests and play games to see if stable, and come back with any significant observation.

Hi OP,

Quesiton for you, did you set a manual fixed vcore then leave CPU ratio and all other settings on AUTO? Has your system been stable at fixed 1.36v voltage? Any CPU degradation after setting so?

I got a 3600x+Gigabyte X570 Aorus Elite borad, I am experiencing annoying motherboard coil whine from the VRM seciton whenever moving my mouse slowly in Windows especially in Chrome. I think that coil whine is caused by CPU/VRM's frequent voltage change. The only fix i found is to set a manual fixed CPU voltage (Vcore) to 1.356v and leave all other settings stock, then the coil whine stops. I just wonder if setting manual vcore without touching CPU ratio is a safe and proper way for long term use.

Any input would be appreciated, thanks!

That's going by the default voltage set by AMD which is more than it needs to run at default speeds.
I've been running my 3600X using 1.25v's set manually and it's been doing just fine, the chip I'm posting with right now (2700X) is only using 1.23v's and has no issues - Been running it that way since day one.
Zen 2 should be more efficient than even than the 2xxx series chips so it's not a stretch to believe the new chips coudn't do the same by comparison. If an issue does appear, all you'd have to do is raise it up, even back to stock if you want.

Hi Bones,

Quesiton for you, did you set a manual fixed vcore then leave CPU ratio and all other settings on AUTO? Has your system been stable at fixed 1.25v voltage? Any CPU degradation after setting so?

I got a 3600x+Gigabyte X570 Aorus Elite borad, I am experiencing annoying motherboard coil whine from the VRM seciton whenever moving my mouse slowly in Windows especially in Chrome. I think that coil whine is caused by CPU/VRM's frequent voltage change. The only fix i found is to set a manual fixed CPU voltage (Vcore) to 1.356v and leave all other settings stock, then the coil whine stops. I just wonder if setting manual vcore without touching CPU ratio is a safe and proper way for long term use.

Any input would be appreciated, thanks!

 

[Bones]

I manually set the voltage and speed it runs at.
The BIOS in my x470 Tachi Ultimate will let me set it's speed manually and I did so, everything to that end is still working well for me with CPU voltage set below 1.30v's - I can't recall ATM where it is but it's below that voltage value for sure.
Some boards are different in how things are implemented in the BIOS but all do the same thing in the end.

Yes, as long as you have the voltage set within a certain range, based on cooling and temps seen UNDER LOAD (Not idle) it's fine for long term use. As for the mouse noise I've experienced it before, you can try a different mouse if you have one and see it that changes anything.

 

[liuyc_2007]

I manually set the voltage and speed it runs at.
The BIOS in my x470 Tachi Ultimate will let me set it's speed manually and I did so, everything to that end is still working well for me with CPU voltage set below 1.30v's - I can't recall ATM where it is but it's below that voltage value for sure.
Some boards are different in how things are implemented in the BIOS but all do the same thing in the end.

Yes, as long as you have the voltage set within a certain range, based on cooling and temps seen UNDER LOAD (Not idle) it's fine for long term use. As for the mouse noise I've experienced it before, you can try a different mouse if you have one and see it that changes anything.

So you set fixed cpu voltage and fixed cpu speed like 4.1GHz at 1.3v correct?

What I want to do is just setting a fixed cpu voltage and leave cpu speed on auto, not sure if that will be a proper way.

 

[Bones]

Mine was at stock speed.
You'd probrably need 1.35v's to run it at 4.1GHz, just be sure your load temps aren't getting out of hand with your settings whatever they wind up being and you'll do fine. It will take some experimentation to find out what the lowest voltage your system needs to run there with stability.

 

[nwtech]

I have an R5 3600, set the vcore to 1.325 and 4.4GHz on all cores by Ryzen Master. Can this affect the CPU life? Affect the cilice?

 

[thesmokingman]

I have an R5 3600, set the vcore to 1.325 and 4.4GHz on all cores by Ryzen Master. Can this affect the CPU life? Affect the cilice?

It depends on your individual silicon. 1.325v is the ballpark for maximum FIT voltage for high current loads, but... each chip is different. You should run a test to see where your chip's threshold is. You can do this by maxing PBO, stock everything else, and load it up with a heavy test app, not cinebench. Then watch your voltage, where it ends up averaging is about your maximum safe voltage. Stay under that maximum.

 

[nwtech]

That way you say?

I left everything in stock, the PBO left it at 200% (maxing) and the maximum after stress was voltage v1.469



_Note: Sorry, I'm not fluent in English.

 

[nwtech]

4.4GHz (All Cores) 1.3vcore. Temp máx: 81°

Acceptable for the cilice?
Does it affect how to keep it that way in the future?

 

[learners permit]

These cpus are beasts at incredibly low voltages. My 3800x runs 4.5Ghz all cores at 1.1v . I know some of you are thinking BS dude not possible. See here.

 

[R0H1T]

It is possible, basically all Ryzen chips operate past their peak efficiency curve, much like Intel. If you lower their clocks a little bit you can definitely save a lot of power by dialing down the voltage, at low clocks they're way past what Intel can offer even if we count Intel's peak efficiency!

 

[tabascosauz]

These cpus are beasts at incredibly low voltages. My 3800x runs 4.5Ghz all cores at 1.1v . I know some of you are thinking BS dude not possible. See here.

I don't doubt that you have a good chip, but show us Effective Clock and SVI2 Vcore in HWInfo during, not after these benchmarks. Ryzen Master is not an effective long term monitoring program.

103C at just 1.1V has alarm bells ringing, I don't care what your clocks are.

Also, "stable" is stable through heavy gaming, half an hour or more in P95 Smallest/Small, intense settings IBT, and multiple runs of CB in short order. Ryzen Master's built in "stress test" is entirely useless and doesn't even stress the CPU properly, giving the impression of stability at some absurd clocks/voltage settings.

The slow RAM won't have a significant impact on your CB scores, but I can't say they aren't holding you back either.

 

[R0H1T]

Part of it could be that he has only 8c/8t with SMT disabled IIRC.

 

[thesmokingman]

That way you say?

I left everything in stock, the PBO left it at 200% (maxing) and the maximum after stress was voltage v1.469



_Note: Sorry, I'm not fluent in English.

First, it is hard to tell what you are doing and what I'm looking at. You will never see a 1.4v+ when all cores are maxed out under load. That amount of voltage is way over the general range of max FIT voltage for high current loads.

Judging from your screen Ryzen Master is saying your max peak voltage is 1.268v. If that voltage was attained while running an all core load, then that is your max safe voltage.


Read this link on how to setup hwinfo for monitoring. Stop using all other forms of monitoring as 99% are not accurate with Matisse.

Build: 3970x, dual 2080ti, 8TB m.2 RAID = Render Monster

At 95c the chip throttles down. That's kind of the point to the test.

hardforum.com

I don't doubt that you have a good chip, but show us Effective Clock and SVI2 Vcore in HWInfo during, not after these benchmarks. Ryzen Master is not an effective long term monitoring program.

103C at just 1.1V has alarm bells ringing, I don't care what your clocks are.

Also, "stable" is stable through heavy gaming, half an hour or more in P95 Smallest/Small, intense settings IBT, and multiple runs of CB in short order. Ryzen Master's built in "stress test" is entirely useless and doesn't even stress the CPU properly, giving the impression of stability at some absurd clocks/voltage settings.

The slow RAM won't have a significant impact on your CB scores, but I can't say they aren't holding you back either.

I already posted in his thread DEGRADATION. His screens are full of missing info, smh...

 

[learners permit]

Would not the resultant = the sum of the factors. Look at the numbers it is apparent that nothing is out of line. The thing that eludes folks is the die shrink effectively gives better efficiency so massive amounts of voltage are in the history books or should be by now. It's all in the current regulation now so a top tier board with kickass VRM's is required to effectively power the cores minus transient disturbances.

Analogy for the masses. Can you build a 1000HP V8 with a nodular iron crankshaft? In short yes if you'd like the output at maximum for 1 whole second.

@ Tabasco thanks for pointing out the metrics that will "finally settle this dispute"? My computer geek friend promises to make himself available this Sunday to help me properly configure the rtss osd for hw info before the last suicide air run for this poor old 3800x before the XT settles in and gets comfortable.