AMD Ryzen 9 3900x Voltage Confusion

Eikon, I run that same board. Ill toss my bios settings in here after I finish up some work =) I have made a couple of changes to my system since I started this post a while back, and I have a bit more insight as to how voltage works on these chips. It is so much different that we have ever really dealt with in the past. I found a Buildzoid video where he attempts to explain the per-core voltages.
 
Also, just as a quick follow up. PBO Will bug out at anything over 210a TDC and 280a EDC on MSI Motherboards. I think for Gigabyte boards that number is closer to 240a/240a on both. After all of my testing I can confirm that a good All Core OC is more efficient AND gives you better single and multi-core performance.
There is a way to set these values to 0, essentially breaking PBO... This will break your chip quick... It shuts down almost all of the lifesaving features of your chip and just lets the current flow based on motherboard recommendations... In my opinion and again, this is just based on my testing, avoid PBO tweaking. When I say PBO bugs out, once you go above those settings you will notice Same/Worse performance than stock. So MSI has like PBO Enhanced 1, 2 ,3 4... Anything over 2 essentially fails to provide any kind of boost. If you can even call what BPO does as boosting... More like a shitty heatsoak for you cooling system.

While testing PBO even when I could get my cores to touch 4600mhz they sit there for a fraction of a second... That will not help you in gaming or any kind of long term task. It may give you a slight bump in doing something like, loading a browser, or launching an application. But your clocks will drop to around 4.1 to 4.2 ghz and be using anywhere from 1.35 to 1.45 volts per core to get those shitty clocks. PBO just makes that worse. It lets the cores suck more juice for longer periods of time, saturating your cooling system with un-needed heat. These cores are surprisingly efficient, when given the proper voltages.


Here are my current settings in RM -

1589390177214.png



This is what CPUz Looks like -

1589390247541.png


( I will grab some screen shots of my bios once I can reboot.)

I have found that 4450 is the best Bang for my Cooling buck. My max temps hang around 74c which is about as high as I am comfortable with long term, as I expect this machine to last me another year before I go to a 4000 series chip. I did change my RAM from my initial posting. I was running some Samsung B Die 4200 kit from an Intel machine I had. Which gave me the ability to kind of play with fabric clocks from 1600 to 1900. The ideal fabric clock imo is between 1800 and 1866 depending on your chip. 1900 is the BLEEDING edge of stability id argue for almost ANY Zen2 chip you see being OC'ed online right now. I got a lot of lock ups running long term at 1900, even if I was Prime stable. So I settled in at 1800 fabric with a 3600 memory kit. This allowed me to basically lock SOC voltage in at 1.1 and forget about it. I found that the fabric clock was much more important to stability than even my core clocks for long term use. Lastly, at 4450 I found that my single core scores were nearly optimized. I say nearly because I can ALMOST touch 4.6 all core on this chip, and the gain is not that significant when you consider what I lose in stability and temps. The way I quickly judge this when I am testing OC's is to use the Benchmark built into CPU-Z, my Multi thread score is ~9020 and my single thread score is 550-551. For reference at 4500mhz its only about 553. The gains depreciate quickly as the chip struggles at higher clocks.


https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232861?Item=N82E16820232861

This is the RAM kit I am currently using. I have played with DRam Calc and essentially this kit at XMP is good right out of the box. Its on par with my totally tweaked out Samsung B Die kit which is considered superior. While the timings ARE tighter on the Samsung memory, the gains are not noticeable. This kit also runs at 1.35v where as my Samsung kit was pushing 1.46 to get the speeds DRam Calc suggested. Rather than sacrifice stability for 1 to 3~ performance I just went with a kit that fit my need.

BIOS settings coming soon -
 
Ok, So here comes the BIOS settings... They are... Pretty basic as not much is needed.

First though, I am using the newest BETA Bios found here - https://us.msi.com/Motherboard/support/MEG-X570-ACE

Leave PBO on Auto everything -

1589402028239.png
1589402118562.png


Nothing fancy, no real secrets. Now, I will say if I am going for broke and looking for the BEST posible benchmarks... Ill go through manually edit all my RAM Timings, and manually set VDDG, VDDP, and my CLDO voltages... But you are talking about 3 to 5% gains for a LOT of work... So. This is the simple 24/7 set up im running. I like the AMD Overclock voltage option because it gives me smaller increments.

Let me know if you have any other questions, Ill do my best to answer.
 
Hi, im trying some of this stuff to try how strong my new pc for some hobby and found some benching etc.. Im trying to absorb info here and got confuse.

I tried to change something in bios and makes me sick of so many settings and failed many times specially in ram. I can do only 4.1ghz (CB 7180) and 4.2ghz (CB 7342) with game boost in bios. So i tried reset bios and just do XMP+Disabled PBO+Save and go to RM.

1st try: 1.35v @4.4ghz and worked got CB 7757 at 74°c = happy 😀
2nd try: 1.35v @4.425 and worked until CB stopped
3rd try: same as 1st try and got CB 7776
4th try: 1.2v @ 4.0ghz i lower it to compare to auto PBO with offsets and got CB 7776 and now confused 🤔🤔 (Edit: not happened again, maybe error from CB)

[email protected] = [email protected] = CB 7776 (Edit: not happened again, maybe error from CB)

I notice lower voltage with higher clock produce more heat than opposite🤔🤔🤔.

XMP+Offset (-0.05 to -0.10) got 4.1ghz max @ 1.35-1.36v @55-68°c Cb=7100+
XMP+PBO off+RM 1.2v+4.0ghz @74°c Cb=7776 (Edit: not happed again, maybe error from CB)

One more thing if I just run Game boost in bios+XMP others auto run in CB at [email protected]@50°[email protected] cpu power Compare to [email protected]@74°[email protected]

Which one now is safer? Seems like it on watts produce heat.
 

Attachments

  • 1597953560569_7776 - 1.2 at 4.0Ghz.png
    1597953560569_7776 - 1.2 at 4.0Ghz.png
    932.1 KB · Views: 0
  • 7776 - 1.35 at 4.4Ghz.png
    7776 - 1.35 at 4.4Ghz.png
    929.5 KB · Views: 0
  • Fail 1.35 - 1.3625 - at 4.425Ghz.png
    Fail 1.35 - 1.3625 - at 4.425Ghz.png
    686.2 KB · Views: 0
Last edited:
Update: I just updated to latest Bios AGESA ComboAm4PI 1.0.0.6:

BIOS: XMP + Disable PBO + CPU Override 1.325v
RM: set 4.500/4.425 and 1.3625v

Previous Bios 1.0.0.4: Can't finish cinemabench on 4.425Ghz all core
Updated Bios 1.0.0.6: Can run with 4.500/4.425 @75°c and faster bios time (from 17.5s to 13s)


I have 4 stick rams and cannot overclock, don't know why but maybe due to different Die or maybe can't with 4 sticks. I'm time passing newbie.
 

Attachments

  • 7883 - 4.500 and 4.425 1.3625v.png
    7883 - 4.500 and 4.425 1.3625v.png
    635.9 KB · Views: 0
With 4 sticks of ram you may need to change your NB/SOC Voltages by just a small amount. Also may need to bump VDDG and VDDP voltages by a .05 or .1 or so. I have never had any luck pushing one chiplet higher than the other, it never seems to amount to any real world gains.
 
try adding .1v or and/or lower its speed one step.
Thanks for the baseline, its works for me in 16-21-21-42 tRC:64, tRFC:580 1.38v for 4 sticks and got improvement.

Another question, which one mostly affects the speed of ryzen cpu in benching? Is it read/write/copy or latency in aida64?
 
Thanks for the baseline, its works for me in 16-21-21-42 tRC:64, tRFC:580 1.38v for 4 sticks and got improvement.

Another question, which one mostly affects the speed of ryzen cpu in benching? Is it read/write/copy or latency in aida64?
no idea. i dont dig that deep into tweaking ram anymore. i never notice a difference and only see it in benchmarks.
 
Thanks for the baseline, its works for me in 16-21-21-42 tRC:64, tRFC:580 1.38v for 4 sticks and got improvement.

Another question, which one mostly affects the speed of ryzen cpu in benching? Is it read/write/copy or latency in aida64?

The lower the CL the Better... but ultimately... I don't notice THAT much difference highly tuning my memory. I have sat and played with them for hours and hours using the DRAM Calc and Samsung B-Die and this new Hynix NDJR and really only notice scores with a 1~3% difference. In memory benchmarks specifically around a 5% difference. Maybe I suck at memory tuning and that is a high possibility lol, But with the amount of effort it requires, id rather just get the lowest CL timing I can that keeps my fabric around 1800 to 1900 and call it a day. You will notice in my original post that I say Tighter timings yield better results... Which was truer then than now, I attribute a lot of that BIOS maturity.
 
Last edited:
Was playing around with TimeSpy a bit today too, I dunno if people use these still. Was comparing it to my old scores which were around 15300 at 4500mhz

1598648855885.png
 
The lower the CL the Better... but ultimately... I don't notice THAT much difference highly tuning my memory. I have sat and played with them for hours and hours using the DRAM Calc and Samsung B-Die and this new Hynix NDJR and really only notice scores with a 1~3% difference. In memory benchmarks specifically around a 5% difference.
Which app you are using to benchmark ram? im using only userbench which from 116% (CL18-22-22-42) become 126% (CL16-21-19-19-39).
I found out that with my 4 rams I cannot lower tRCDRD below 21, but can go until 19 if using ony 2 rams.
 
Which app you are using to benchmark ram? im using only userbench which from 116% (CL18-22-22-42) become 126% (CL16-21-19-19-39).
I found out that with my 4 rams I cannot lower tRCDRD below 21, but can go until 19 if using ony 2 rams.

I am just using MEMBench that is built into DRAM Calculator.
 
After playing with ram timings I tried benching and got 8k at 4.5/4.4Ghz, 1.3625v @74°c, PBO Off+Auto+RM.

I notice that better ram timings more cinemabench score.

Before @XMP: 4.500/4.425Ghz Cb=7883
Better timing: 4.500/4.400Ghz Cb=8003.
with same 1.3625v

With my 4 ram sticks 16-21-21-42-64, i notice that if i cannot lower my tRCDRD, ill just lossen the primary timings and focused on subtimings and i think its better now running 1.37v no erros in TM5 and Memtest64. I dont want to run it more than 1.37v for 4sticks for daily use as XMP running at 1.37v so i stop timings at some point.

I found out also that my XMP have so many errors.
I found out also with extra VSOC gives error in TM5 so i made it vsoc=1.05 and become stable.
 

Attachments

  • 8003 - 4.5-4.4ghz - 74c.png
    8003 - 4.5-4.4ghz - 74c.png
    851.9 KB · Views: 0
  • 16-21-21-42-64.PNG
    16-21-21-42-64.PNG
    171.3 KB · Views: 0
Last edited:
Ok, So here comes the BIOS settings... They are... Pretty basic as not much is needed.

First though, I am using the newest BETA Bios found here - https://us.msi.com/Motherboard/support/MEG-X570-ACE

Leave PBO on Auto everything -

View attachment 245258View attachment 245260

Nothing fancy, no real secrets. Now, I will say if I am going for broke and looking for the BEST posible benchmarks... Ill go through manually edit all my RAM Timings, and manually set VDDG, VDDP, and my CLDO voltages... But you are talking about 3 to 5% gains for a LOT of work... So. This is the simple 24/7 set up im running. I like the AMD Overclock voltage option because it gives me smaller increments.

Let me know if you have any other questions, Ill do my best to answer.


Thank you SO MUCH for posting this!

Was fighting with my 3900XT on a MSI X570 Tomahawk for a while and that first pic is like magic!

Was trying to get auto boost to work, but it just wasn't going very well. Tried the MSI "Game Mode" in
the BIOS which gives 4.150Ghz but that failed quickly in CB20.

Used your settings from that first pic to go to 4.2, then 4.3, then 4.4, all rock stable so far!
Picked up 500 points in CB20. Last one was 7844 I believe.

I haven't tried any higher yet. I use this computer in my home office for work as well as play so it needs
to be stable. Will leave it at 4.4Ghz for a while and see how it goes. May try higher at some point.

Would be great if you started a new thread with those pics as the first post and maybe a mod could make
it a sticky. I think this info is that important to anybody trying to get the most from their AMD CPU.

Nice going on finding such dead simple settings that work so well!

THANKS! 👍


ETA: Just wanted to add that this is with air cooling, no water. NH-D15 Black cooler.

ETA#2: Same settings worked on my 3600X/B450 Tomahawk. It's at 4.4Ghz so far when before it would not
do more than 4.2Ghz all core.

.
 
Last edited:
Thank you SO MUCH for posting this!

Was fighting with my 3900XT on a MSI X570 Tomahawk for a while and that first pic is like magic!

Was trying to get auto boost to work, but it just wasn't going very well. Tried the MSI "Game Mode" in
the BIOS which gives 4.150Ghz but that failed quickly in CB20.

Used your settings from that first pic to go to 4.2, then 4.3, then 4.4, all rock stable so far!
Picked up 500 points in CB20. Last one was 7844 I believe.

.
Nice. Your score is high for 4.4Ghz, what is you CPU voltage?Ram?
 
You're right... I was going from memory it was 7834.
Just tested again now and it's 7885. :)
I can't run all core 4.4Ghz with 1.32v, I need 1.337v to run that with AMD Balance Powerplan. Maybe your board doing it.
What is you chipset driver version? Did you update from AMD or Dragon Center?

I notice that I'm having lower score when ryzen master is open, now I check with RM closed and got get 7803-7874 at 4.4Ghz.
 
Last edited:
Not sure on chipset driver version, I think whatever the newest is. DL'ed the newest from AMD.
I don't use Ryzen Master or Dragon Center on this system, never installed either of them.

The voltage difference is probably 3900X vs. 3900XT.

I'm using the Ryzen High Performance power plan.

.
 
Thank you SO MUCH for posting this!

Not a problem at all! I tinker quite a bit and did a LOT of testing in the beginning. Tuning on these AMD chips has been a ton of fun, playing with PBO has been a huge let down though lol.

I have a theory that the new chips are going to hit 4.8 all core... and Ill be CRAZY stoked. ( this theory is just based on chiplet refinement and what I have seen in terms of stable clocks between X and XT versions. )

I am doing some testing based on the reporting from Mxking035 showing that his scores got better when he dropped his SOC Voltage. I am wonder if this is all based on the memory controller on the chip it self, OR if this is chipset quality related.

I also also use the AMD Ryzen High Performance setting in my power profile.

Also, when i go for a top end benchmark in something like CB or Timespy, let your machine boot all the way up, close all extra junk like Steam or any keyboard software ect. Then run a benchmark with your thermal monitor up, just to ensure you are not peeking temps, then CLOSE everything. Close it all. Then run ONLY the benchmark. Dont worry about temps or anything, if temps were going to be bad you'd have noticed in the first run. Software bloat will cost you as little as 3% and as much as 15% in terms of benchmarks.
 
I am doing some testing based on the reporting from Mxking035 showing that his scores got better when he dropped his SOC Voltage. I am wonder if this is all based on the memory controller on the chip it self, OR if this is chipset quality related.
I think so, I tried 3733Mhz 65.9ns (Read 57Gb/s) with 1.1v vSoC cannot reach 8k no matter what with my air cooler, Only settings stable at vSoC=1.05v 16-21 3600mzh 69.9:playful: .

I tested all Power Plan: locked at 4.4Ghz 1.337v(minimum I can run) in 4 runs each, I make sure all are cool to be fair. listed are in sequence in test, i started backward from 3rd run with restart.

AMD Performance: 7833-earlycrash-7803-earlycrash
1usmus Powe : 7863-midcrash-midcrash-7862
AMD Balance : 7819-7832-7820-7832
1usmus Universal : all crashed - tried more
Windows Ultra : all crashed - tried more
Windows High : 7855-earlycrash-reboot-7865
Windows Save : all crashed - tried more

crash=cinemabench stopped
 
Damn. That's a pretty impressive improvement. I usually only get about 7200 cinebench with my 3900X
Ram is at 3200, and processor is undervolted -0.1v

I'm curious, you said you're doing that on air - what are your temps like? I didn't like how hot my processor got when OCing, even though I knew it was safe. I undervolt and it gets to 75c during stress tests and transcoding but only about 65c during normal usage.
 
Damn. That's a pretty impressive improvement. I usually only get about 7200 cinebench with my 3900X
Ram is at 3200, and processor is undervolted -0.1v

I'm curious, you said you're doing that on air - what are your temps like? I didn't like how hot my processor got when OCing, even though I knew it was safe. I undervolt and it gets to 75c during stress tests and transcoding but only about 65c during normal usage.

Not sure which one of us you are responding to.....

My CPU temps bounce around from 34.4 - 40.5 at idle.
Stressed with IBT at default settings I get 77.3 max temp.

Temps monitored in HWInfo.

.
 
Damn. That's a pretty impressive improvement. I usually only get about 7200 cinebench with my 3900X
Ram is at 3200, and processor is undervolted -0.1v

I'm curious, you said you're doing that on air - what are your temps like? I didn't like how hot my processor got when OCing, even though I knew it was safe. I undervolt and it gets to 75c during stress tests and transcoding but only about 65c during normal usage.
Here is mine:
When gaming around 50-60°c max set 1080p dota2, ambient 22-24°c in AC. Auto-0.05v offset

When running 4.5/4.4Ghz cinemabench: around 74-78°c
IMG_20200918_053852.png

IMG_20200914_191128.jpg
 
Last edited:
Is there a general consensus out now what a "safe daily use" on the vcore of the ryzens? I know this is a 3900X thread (I have a 3900XT) but I have been searching for that answer since I had this 3900XT for about 3 weeks now. I've ready anywhere from 1.25 to 1.35 is a daily safe vcore setting. Although I do remember reading about Buildzoids 3700X degrading at 1.37v (i think that voltage) pretty quickly. I'm just a tad gun shy on doing a manual overclock without a clear number to look out for since all im doing is gaming and the pc isn't on 24/7.
 
Is there a general consensus out now what a "safe daily use" on the vcore of the ryzens? I know this is a 3900X thread (I have a 3900XT) but I have been searching for that answer since I had this 3900XT for about 3 weeks now. I've ready anywhere from 1.25 to 1.35 is a daily safe vcore setting. Although I do remember reading about Buildzoids 3700X degrading at 1.37v (i think that voltage) pretty quickly. I'm just a tad gun shy on doing a manual overclock without a clear number to look out for since all im doing is gaming and the pc isn't on 24/7.

There is no such thing as a standard safe voltage although maximum safe FIT voltages usually cap out around 1.325v. Max FIT voltage rarely goes past that. What that means is your max FIT voltage could be lower. Everyone running higher than that is on borrowed time longterm or just benching for fun, again not longterm.
 
There is no such thing as a standard safe voltage although maximum safe FIT voltages usually cap out around 1.325v. Max FIT voltage rarely goes past that. What that means is your max FIT voltage could be lower. Everyone running higher than that is on borrowed time longterm or just benching for fun, again not longterm.

To find max FIT you run p95 on small FFT and check with hwinfo64 (sorry kinda new this)?
 
To find max FIT you run p95 on small FFT and check with hwinfo64 (sorry kinda new this)?

Not quite. You have to set PBO to run to achieve exactly the target clock frequency that you are looking to run, that's under the assumption you wanna run an allcore or whatever. Then you run P95 or whatever load that is representative of your load then sit back and watch PBO do its thing via hwinfo. Again, whatever you find is only applicable to the load you generated during the test, so if you ran R20 the voltage reading you'd get won't be applicable to a heavier P95 load. In other words the lower the stress load, the higher the voltage will be since PBO will scale more in that sitaution. That higher voltage will be too higher in a high load situation, since what PBO will do is lower the voltage to stay safe. I hope I'm conveying the voltage vs load relationship.
 
Yeah, I struggled with this for a while and found that essentially 1.32 to 1.35 under FULL load, is about the tops. What we have to understand with these new chips is that the damage is done when its High Volts and High Draw. So if you are running a stressful workload like P95 at 1.4 with no vdrop .... You are going to roast that chip. I actually do not use an agressive load line calibration when I OC for scores. I tend to use like level 3 where it's more of a flat line, I want my voltage to stay the same at load, or to even drop a bit, and then calibrate my 24/7 stable voltage that way. Again, any time you overclock something or push it outside of spec... you run the risk of degrading the chip. Maybe it's only 10% over 5 years. Either way that is a risk we take, and if you are here on HARD, chances are pretty good its a risk you are willing to take.
 
Though as a side bar... I would like to talk about how much I miss the HardOCP site... Gamers Nexus is doing ok... But I miss the write up and the origins of the "Real World" benchmark... Anyways. Nostalgia mode deactivated.
 
This is a great thread. Can someone a lot wiser and with lots more OC experience than me summarize the "best practices" here. Maybe it becomes a sticky.
 
I am Hyped about the 5950x and the 5900x! When i get my hands on one i'll do another new post.
 
Chips sold out instantly... even having one in the cart did not matter... So... Time to wait I suppose.
 
Chips sold out instantly... even having one in the cart did not matter... So... Time to wait I suppose.

I took a ride up to the MC in Westmont, IL this morning to try to get a 5900X.
Got there 30 minutes before open and there had to be 150 people there wrapped all around the building.

Just turned around in the lot and went home. lol

.
 
I took a ride up to the MC in Westmont, IL this morning to try to get a 5900X.
Got there 30 minutes before open and there had to be 150 people there wrapped all around the building.

Just turned around in the lot and went home. lol

.

Yea.... yea. Been there done that. Those 150 bros all thinking the same. They restock Mondays at my local MC...
 
I wrote a massive reply but decided to deleted it.

The safe voltage you should run is no more than 1.4v with a conservative Load Line.

or

1.35v with an aggressive load line.

These are observations I have made on my own platform since day 1.

I get the best thermals and performance @ 1.32 volts and I am fully watercooled.

Remember your custom loop doesn't allow one to be retarded with their OC's. It just lets you be more aggressive. Dont go full retard and you will be good!

Lastly keep in mind that a idle chip can safely run 1.5 or more volts as long as you dont load it. Just becuase something appears stable at 1.4+ v doesn't mean it needs that much.

We have to abandon Intel voltage thought here. Were not dealing with 14 +++++++++ nm here anymore lmao.

Pretty much spot on.
 
Back
Top