Definitive guide to configuring the Ryzen 3900X/3950X and all other 3000 Series CPUs

[mtcn77]

I would like OP's or anyone else's commends on the following screenshots from my 3600 during different workloads
FYI, PBO settings are not just auto or enabled

1. Idle/browsing/watching videos
View attachment 148055

2. CB-R20 single thread
View attachment 148056

3. CB-R20 multi thread
View attachment 148057

4. FarCry NewDawn gameplay
View attachment 148058

You can see the thread counts on differentiating which loads execute on each cpu core.
For instance,

1. Cinebench Multi is the heaviest load, hence C6 is the least and C1 is also the least,
2. Far Cry is in second, C6 inactivity is only minutely present for core 6 and all cores are achieving their Pb state at 100% C1,
3. Cinebench single - 2 cores are off/C6 and 4 are fully active/C1,
4. Browsing/videos - 3 active, 5 boosting time to time and 3 at part sleep.

 

[moproblems99]

I would like OP's or anyone else's commends on the following screenshots from my 3600 during different workloads
FYI, PBO settings are not just auto or enabled

What are you looking for? Maybe ask for specific comments?

 

[TheoneandonlyMrK]

You can see the thread counts on differentiating which loads execute on each cpu core.
For instance,

1. Cinebench Multi is the heaviest load, hence C6 is the least and C1 is also the least,
2. Far Cry is in second, C6 inactivity is only minutely present for core 6 and all cores are achieving their Pb state at 100% C1,
3. Cinebench single - 2 cores are off/C6 and 4 are fully active/C1,
4. Browsing/videos - 3 active, 5 boosting time to time and 3 at part sleep.

Perfect then.

 

[Michael Nager]

I would like OP's or anyone else's commends on the following screenshots from my 3600 during different workloads
FYI, PBO settings are not just auto or enabled

1. Idle/browsing/watching videos
View attachment 148055

2. CB-R20 single thread
View attachment 148056

3. CB-R20 multi thread
View attachment 148057

4. FarCry NewDawn gameplay
View attachment 148058

This would be my 3600X running prime95 small FFTs after over three hours:

The 3600X running CineBench R20 just running "Stock":

The 3600X running CineBench R20 after configuring it according to the guide I posted:

I did run CineBench with my 3600X at 1.275 Volts, but that wasn't stable, and eventually I found the sweet-spot was at 1.29375 Volts, but here are the temps during a CineBench R20 run at 1.275 and below that the CineBench R20 result:

For S&G here is what the 3600X clocked to with PBO and Auto Overclock (notice the voltages):

First PBO:

The CineBench R20 result:

With Auto Overclocking:

Here is the 3600X running CineBench R20 in Default mode:

And now the Cinebench R20 result:

Yes I do take a hit on the single core score, but that being said, when do you actually ever run single core workloads on your system where you generally have a shed-load more going on than just running CineBench.

The cooler on the CPU was the Noctua NH-U12A.

 

[mtcn77]

Perfect then.

Well, not all is.
In Far Cry Dawn, Core 4-5 are not at sleep, but they sure aren't in full boost either(c1 is restricted). Maybe it is a mid boost state, half the core count is at max while the rest is at base clocks. FarCry seems to be such a medium load. I sure would try my chances at affinity load preferencing to those cores.

 

[Michael Nager]

I

Well, not all is.
In Far Cry Dawn, Core 4-5 are not at sleep, but they sure aren't in full boost either(c1 is restricted). Maybe it is a mid boost state, half the core count is at max while the rest is at base clocks. FarCry seems to be such a medium load. I sure would try my chances at affinity load preferencing to those cores.

If you want the easiest way to set the Affinity of your CPU to configure it for games then I would suggest the free software "Process Lasso":

Bitsum. Real-time CPU Optimization and Automation

Real-Time CPU Optimization and Automation. Keep your PC responsive during high CPU loads and automate process settings with rules. Apps run YOUR WAY!

bitsum.com

 

[moproblems99]

@Michael Nager , curious question, why didn't you put these values in the UEFI and make them permanent?

 

[Michael Nager]

@Michael Nager , curious question, why didn't you put these values in the UEFI and make them permanent?

It might be a GigaByte BIOS thing, but one of the first things I tried to do was enter the values from Ryzen Master into the BIOS in different combinations (BIOS, the AGESA portion, various combinations) and every time I did so the system went to shit.

It would reboot a number of times, do the memory training, and then, when I went into the BIOS again invariably the CMOS had been cleared.

I also couldn't find a way of limiting the voltage (but keeping the voltage dynamic and not static) without the system behaving very erratically.

I started calling my BIOS "GigaByte Master" as opposed to "Ryzen Master"

I had a good (well actually bad) few months, always with high hopes for either the newest AGESA version or the newest BIOS version, where the games would start all over again.

It's not that the way I described in the original post is the best, it's just that all other ways I tried ended up sucking in one way or another.

As I say I have been experimenting with the system for about eight months now.

I have found the path of least suck

 

[TheoneandonlyMrK]

It might be a GigaByte BIOS thing, but one of the first things I tried to do was enter the values from Ryzen Master into the BIOS in different combinations (BIOS, the AGESA portion, various combinations) and every time I did so the system went to shit.

It would reboot a number of times, do the memory training, and then, when I went into the BIOS again invariably the CMOS had been cleared.

I also couldn't find a way of limiting the voltage (but keeping the voltage dynamic and not static) without the system behaving very erratically.

I started calling my BIOS "GigaByte Master" as opposed to "Ryzen Master"

I had a good (well actually bad) few months, always with high hopes for either the newest AGESA version or the newest BIOS version, where the games would start all over again.

Ahhh, I see where the bios phobia comes from now, please believe not all board bios and chip combos work out like that.

 

[bug]

This would be my 3600X running prime95 small FFTs after over three hours:
View attachment 148062

The 3600X running CineBench R20 just running "Stock":
View attachment 148063

The 3600X running CineBench R20 after configuring it according to the guide I posted:

View attachment 148064

I did run CineBench with my 3600X at 1.275 Volts, but that wasn't stable, and eventually I found the sweet-spot was at 1.29375 Volts, but here are the temps during a CineBench R20 run at 1.275 and below that the CineBench R20 result:

View attachment 148066

View attachment 148067

For S&G here is what the 3600X clocked to with PBO and Auto Overclock (notice the voltages):

First PBO:


View attachment 148071

The CineBench R20 result:

View attachment 148072

With Auto Overclocking:

View attachment 148073

View attachment 148074

Here is the 3600X running CineBench R20 in Default mode:

View attachment 148075

And now the Cinebench R20 result:

View attachment 148076

Yes I do take a hit on the single core score, but that being said, when do you actually ever run single core workloads on your system where you generally have a shed-load more going on than just running CineBench.

The cooler on the CPU was the Noctua NH-U12A.

So basically, differences within the margin of error?
But at least it seems you managed to lower the voltage a bit.

 

[Michael Nager]

Ahhh, I see where the bios phobia comes from now, please believe not all board bios and chip combos work out like that.

Well I spent so much time kissing frogs that, although I may not have met my prince, at least it is no longer a frog.

 

[moproblems99]

It might be a GigaByte BIOS thing, but one of the first things I tried to do was enter the values from Ryzen Master into the BIOS in different combinations (BIOS, the AGESA portion, various combinations) and every time I did so the system went to shit.

It would reboot a number of times, do the memory training, and then, when I went into the BIOS again invariably the CMOS had been cleared.

I also couldn't find a way of limiting the voltage (but keeping the voltage dynamic and not static) without the system behaving very erratically.

I started calling my BIOS "GigaByte Master" as opposed to "Ryzen Master"

I had a good (well actually bad) few months, always with high hopes for either the newest AGESA version or the newest BIOS version, where the games would start all over again.

It's not that the way I described in the original post is the best, it's just that all other ways I tried ended up sucking in one way or another.

As I say I have been experimenting with the system for about eight months now.

I have found the path of least suck

Interesting. Sorry to hear that.

 

[mtcn77]

I

If you want the easiest way to set the Affinity of your CPU to configure it for games then I would suggest the free software "Process Lasso":

Bitsum. Real-time CPU Optimization and Automation

Real-Time CPU Optimization and Automation. Keep your PC responsive during high CPU loads and automate process settings with rules. Apps run YOUR WAY!

bitsum.com

I wonder if assigning 'real-time' access to individual processes gives them private access to specific cores because that would help in other ways, mostly on security.
One alternative reason is data latency is kept private and not shuffled between threads.
I haven't concocted my recipe, but assimilating from the previous far cry dawn example, I'd wager delimiting a few less threads than active(say, 84.6% per 5-6 cores ~ 9/10 full threads - 5 threads are idling on average) to see if frafs bench viewer logs would improve.

 

[Zach_01]

This would be my 3600X running prime95 small FFTs after over three hours:
View attachment 148062

The 3600X running CineBench R20 just running "Stock":
View attachment 148063

The 3600X running CineBench R20 after configuring it according to the guide I posted:

View attachment 148064

I did run CineBench with my 3600X at 1.275 Volts, but that wasn't stable, and eventually I found the sweet-spot was at 1.29375 Volts, but here are the temps during a CineBench R20 run at 1.275 and below that the CineBench R20 result:

View attachment 148066

View attachment 148067

For S&G here is what the 3600X clocked to with PBO and Auto Overclock (notice the voltages):

First PBO:


View attachment 148071

The CineBench R20 result:

View attachment 148072

With Auto Overclocking:

View attachment 148073

View attachment 148074

Here is the 3600X running CineBench R20 in Default mode:

View attachment 148075

And now the Cinebench R20 result:

View attachment 148076

Yes I do take a hit on the single core score, but that being said, when do you actually ever run single core workloads on your system where you generally have a shed-load more going on than just running CineBench.

The cooler on the CPU was the Noctua NH-U12A.

Well, this is not commending but ok! I expected at least some boost related commends on the 3600, because I read a lot about it in previous posts that its not entirely true. The 3600X is not for direct comparison as there's a big difference in binning and rated limits.
Anyway, I dont like in-windows settings and I only do that from UEFI. And for me single thread clock is most important. I dont do any real life all core workloads other than benches.

What i've done with it is to increase clock through the entire boost clock curve and not only for all-core loads. It all comes down to preference I guess.

 

[Waldorf]

@Michael Nager
GB seems to have issues with not clearing any values when you update bios, no matter if cleared it by loading defaults or even removed cmos battery.

if you have flash back on ur board:
copy latest bios file to a usb stick like u do for update, rename the same file in different locatio n into gigabyte.bin (all caps?!) and copy to the same stick.
turn off pc,
flash back the the gigabyte file, turn pc on and update bios again thru qflash in bios.
reboot and load defaults.

 

[Michael Nager]

@Michael Nager
GB seems to have issues with not clearing any values when you update bios, no matter if cleared it by loading defaults or even removed cmos battery.

if you have flash back on ur board:
copy latest bios file to a usb stick like u do for update, rename the same file in different locatio n into gigabyte.bin (all caps?!) and copy to the same stick.
turn off pc,
flash back the the gigabyte file, turn pc on and update bios again thru qflash in bios.
reboot and load defaults.

A new BIOS for the GigaByte board I have came out yesterday (F12e) and I was back to doing what I have done many times before and that is to try to achieve what I am now doing with a combination of the BIOS and Ryzen Master.

Changing the values to what I know they should be and getting the usual pushback from the "Save&Exit, Reboot, Three Beeps, Reboot, Memory Retraining, Cleared CMOS, Rinse&Repeat, finally getting a clean boot, Windows loading with garbage on the screen, Reboot, Windows goes into Repair Mode, Reboot, Windows again in Repair Mode, finally booting into Windows and things not running correctly, grabbing my backup, kicking off Acronis and doing a complete restore, going back into the BIOS, setting it back to what I know works there, loading Windows and going back to using Ryzen Master.

That was six effing hours of my life I am never going to get back.

 

[Waldorf]

then why not try it, as it prevents that cycle,
and the reason i posted it, as it fixed the mem/rettain/boot cycle (cause of old bios values).

 

[Michael Nager]

The problem

then why not try it, as it prevents that cycle,
and the reason i posted it, as it fixed the mem/rettain/boot cycle (cause of old bios values).

I have found that the incompatibility appears to be between the GigaByte part of the BIOS and the AGESA binary from AMD.

I should tell you that I am autistic (Asperger's) and that has advantages and disadvantages for me.

The advantage is that I can really focus on a problem and stick to it until I get a solution - or find that no solution exists.

The disadvantage is that I want it the way I want it, and I will not compromise that result.

I could sorta get to where I want to go, but would have to suck some lemons to get there which I am just not willing to do.

I have the advantage of having a dual BIOS on the motherboard which I can address individually, so when doing what I do, I make changes to one BIOS and if that works, switch to the other BIOS to go further etc.

There are points where it falls down and I can replicate those points time and again - including Windows needing to go into Repair Mode.

 

[Waldorf]

if u do flash-back when pc off, and Qflash update for BOTH (incl backup bios) after that, it fixes it.
at least give it a try

 

[Michael Nager]

There is nothing to fix in the BIOS.

The problem is that what I want to do is not achievable in the BIOS and requires the use of Ryzen Master.

I wrote in my original post:

So what you see displayed as "The BIOS" is in effect just a configuration menu for the AGESA. The problem about editing the AGESA portion found under "Settings" under the headings "AMD CBS" and "AMD Overclocking" directly is that with some of the options, if you enter a wrong value, then your system will not boot. What is worse however is that some of the settings cannot be removed with a "Clear CMOS" and your mobo is effectively bricked.

I am aware of what you are saying, and I just wish that were the problem and the solution.

 

[Zach_01]

So what you see displayed as "The BIOS" is in effect just a configuration menu for the AGESA. The problem about editing the AGESA portion found under "Settings" under the headings "AMD CBS" and "AMD Overclocking" directly is that with some of the options, if you enter a wrong value, then your system will not boot. What is worse however is that some of the settings cannot be removed with a "Clear CMOS" and your mobo is effectively bricked.

This is where BIOS Flashback is worth having... other than initially having an unsupported CPU.

 

[Waldorf]

unless the bios uses/applied settings from previous version.
i could not even get it to post after i flashed update, even that the update was for better ram compatibility.
after flash back, then qflashed it (reg update) the same bios, and then it worked, without it having anything to with "wrong" settings.

not saying it WILL fix it, but not trying means ZERO chance it will. ;-)

 

[BrainMuncher]

You said you did a lot of experimenting, I'm curious as to if you tried a negative voltage offset but with everything else at stock. I haven't played with zen 2 yet but I will be doing so soon, with zen 1 a negative offset results in higher sustained boost since it is already running into internal power limits at stock settings. So benchmark scores (both single & multi) improved without touching the clocks at all, everything at stock except 1 voltage setting in the BIOS.

If you did try it how did it compare to your current method?

 

[Michael Nager]

unless the bios uses/applied settings from previous version.
i could not even get it to post after i flashed update, even that the update was for better ram compatibility.
after flash back, then qflashed it (reg update) the same bios, and then it worked, without it having anything to with "wrong" settings.

not saying it WILL fix it, but not trying means ZERO chance it will. ;-)

The thing is that I have been experimenting with this motherboard for four months or so now.

The other thing is, that when I revert the settings and just do it the way that I know WORKS then there are absolutely no problems or anomalous behaviour.

It has gotten to the stage now that I can take one look at the temp the CPU is reaching when running CineBench and I can tell you what score is going to result, and I won't me more than one or two points off.

You said you did a lot of experimenting, I'm curious as to if you tried a negative voltage offset but with everything else at stock. I haven't played with zen 2 yet but I will be doing so soon, with zen 1 a negative offset results in higher sustained boost since it is already running into internal power limits at stock settings. So benchmark scores (both single & multi) improved without touching the clocks at all, everything at stock except 1 voltage setting in the BIOS.

If you did try it how did it compare to your current method?

Hell yeah I tried that.

It was a shitshow.

Utilities like Ryzen Master or HWMonitor or HWInfo or AIDA were showing me that I was using 0 (Zero) Volts going into the CPU. The temps were higher to the point where I had to stop CineBench runs and Windows would hard crash.

From some of the replies to my original post I get the feeling that people think that I WANT to do it the way I documented and that I am too much of a wimp to do the "Manly man" thing and configure it in the BIOS.

The thing is that the CPU is listening to the AGESA portion of the BIOS and NOT the GigaByte portion of the BIOS.

The other problem I suspect lies in the fact that this time around AMD itself is producing the chipset - which as far as I know will change with the next chipset where it will be manufactured by ASMedia again.

I will be testing out the 3900X my friend loaned me on the GigaByte X470 Gaming 7 WiFi motherboard I have to see what, if any, differences there are, because the chipset for that board was manufactured by ASMedia.

I am watching the Gamers Nexus factory tour video and Steve said something that really made me laugh, he said:

... supply from China has experience slowdowns and limitations due to something that starts with "C" and rhymes with "Demonetized"

 

[Zach_01]

You said you did a lot of experimenting, I'm curious as to if you tried a negative voltage offset but with everything else at stock. I haven't played with zen 2 yet but I will be doing so soon, with zen 1 a negative offset results in higher sustained boost since it is already running into internal power limits at stock settings. So benchmark scores (both single & multi) improved without touching the clocks at all, everything at stock except 1 voltage setting in the BIOS.

If you did try it how did it compare to your current method?

Negative voltage offset is not working with ZEN2 as it does with ZEN+. Not in the same manner anyway. ZEN2 is a whole new story and I've been saying it for months now that it needs to be treated from almost zero point, from scratch. Pretty match forget what we know about past CPUs. Just like static OC/voltage.

@Michael Nager has done a lot of testing, and found a way to setup these chips that suit his needs. Feel free to test it as I dont see anything that can compromise silicon longevity. Is it the best way? Is Buildzoid's best way? Probably not and that depends on what everyone is seeking for. There is no best way IMHO. I'm actually peaking a different approach from both of them.