Ryzen 9 3900x Voltage and Temperature

[Max(IT)]

No typo
The offset got 3 settings
AUTO
+
-

But yeah cloud change it shouldn't hurt

so your motherboard is actually giving less voltage than intended ?

You are in an opposite situation then

 

[Shou Miko]

so your motherboard is actually giving less voltage than intended ?

You are in an opposite situation then

HWInfo64 shows max CPU 1.312V and minimum 1.294V when while I do a CB20 run it's 1.294-1.300V so that's not bad, but it's not here it's 31c and I cannot get it cooler in my apartment I hate the summers in a 2 room apartment.

 

[Max(IT)]

HWInfo64 shows max CPU 1.312V and minimum 1.294V when while I do a CB20 run it's 1.294-1.300V so that's not bad, but it's not here it's 31c and I cannot get it cooler in my apartment I hate the summers in a 2 room apartment.

quite different than my situation, where with an offset of -0.100 V, HWInfo is reporting max CPU 1.437 V and minimum 1.044 V (which I think is fine).

 

[Shou Miko]

quite different than my situation, where with an offset of -0.100 V, HWInfo is reporting max CPU 1.437 V and minimum 1.044 V (which I think is fine).

Under the Offset I have set the CPU Voltage to 1.300V this can be set to AUTO or a manual value. I trust AMD Ryzen Master it maxes at 1.3 volt.

 

[Max(IT)]

Under the Offset I have set the CPU Voltage to 1.300V this can be set to AUTO or a manual value. I trust AMD Ryzen Master it maxes at 1.3 volt.

Understood. You set a manual voltage of 1.3 V and then added an offset of +0.125 V.
May I ask you why you did that way ?

 

[Shou Miko]

Understood. You set a manual voltage of 1.3 V and then added an offset of +0.125 V.
May I ask you why you did that way ?

I felt a bit of a stutter and slow opening applications running 4.0GHz all-core but now at 4.1GHz all-core with 1.3V and the +0.125V it runs as normal.

 

[Max(IT)]

Anyways, all that needs to be said has been said a number of times. Up to OP to decide whether he wants to figure things out or continue down this road of "Vcore is set to 1.440V". Which wouldn't be all that surprising if this was an early B550 BIOS we're talking about, as I've mentioned, but OP still hasn't provided info as to what BIOS he's on, so I'm out.

If you just actually READ it is written several posts above your:

I‘m using the latest BIOS released a few days ago

 

[lexluthermiester]

The BIOS suffers from the same monitoring-software-syndrome that plagues every software that isn't Ryzen Master. It simply takes an instantaneous reading of Vcore when it starts, and that's what you see.

This is not true. HWinfo and Open Hardware Monitor both do real-time sensor polling. If you haven't seen either one, go take a look as they both are very useful.

Open Hardware Monitor - Core temp, fan speed and voltages in a free software gadget

openhardwaremonitor.org

HWiNFO - Free System Information, Monitoring and Diagnostics

Free Hardware Analysis, Monitoring and Reporting. In-depth Hardware Information, Real-Time System Monitoring, Reporting & more

www.hwinfo.com

Both are free to use.

 

[Max(IT)]

This is not true. HWinfo and Open Hardware Monitor both do real-time sensor polling. If you haven't seen either one, go take a look as they both are very useful.

Open Hardware Monitor - Core temp, fan speed and voltages in a free software gadget

openhardwaremonitor.org

HWiNFO - Free System Information, Monitoring and Diagnostics

Free Hardware Analysis, Monitoring and Reporting. In-depth Hardware Information, Real-Time System Monitoring, Reporting & more

www.hwinfo.com

Both are free to use.

According to an AMD engineer on Reddit, CPU Z also reports real-time sensor polling regarding the Vcore.

 

[tabascosauz]

This is not true. HWinfo and Open Hardware Monitor both do real-time sensor polling. If you haven't seen either one, go take a look as they both are very useful.

Open Hardware Monitor - Core temp, fan speed and voltages in a free software gadget

openhardwaremonitor.org

HWiNFO - Free System Information, Monitoring and Diagnostics

Free Hardware Analysis, Monitoring and Reporting. In-depth Hardware Information, Real-Time System Monitoring, Reporting & more

www.hwinfo.com

Both are free to use.

That's not what I meant. To be fair, HWInfo is technically "real-time" compared to static BIOS readings as it is polling according to whatever interval is set, but not nearly with the polling resolution or ability to take average values like Ryzen Master, which will show you under Average Core Voltage what the CPU is effectively idling at. HWInfo will just tell you every 1-2 seconds that it's idling at 1.3-1.5V, and CPU-Z will occasionally show idle Vcore dip down to 0.2V. HWInfo's SVI2 TFN Vcore sensor then becomes accurate under full load, because Vcore is constant in that scenario.

I use Openhardwaremon all the time on Intel platforms, but along with HWmonitor it's not nearly as useful/accurate on Matisse compared to the wealth of sensors available to recent HWInfo versions.

 

[Leonoid007]

When use all cores manual overcklocking - does CPU looses idle state? Will it result in significantly more power consumption when idling?

 

[Max(IT)]

That's not what I meant. To be fair, HWInfo is technically "real-time" compared to static BIOS readings as it is polling according to whatever interval is set, but not nearly with the polling resolution or ability to take average values like Ryzen Master, which will show you under Average Core Voltage what the CPU is effectively idling at. HWInfo will just tell you every 1-2 seconds that it's idling at 1.3-1.5V, and CPU-Z will occasionally show idle Vcore dip down to 0.2V. HWInfo's SVI2 TFN Vcore sensor then becomes accurate under full load, because Vcore is constant in that scenario.

I use Openhardwaremon all the time on Intel platforms, but along with HWmonitor it's not nearly as useful/accurate on Matisse compared to the wealth of sensors available to recent HWInfo versions.

I don’t know which motherboard are you using but on mine bios readings aren’t static. They are real time.

 

[tabascosauz]

I don’t know which motherboard are you using but on mine bios readings aren’t static. They are real time.

The exact same 0608/0805/1002 BIOSes that you're looking at, minus the red ROG colour scheme?

Just because they change hardly means they're "real time". I don't know how many times I have to repeat myself. My BIOS only ever shows me 1.450 and 1.472 Vcore, both of which are horseshit because no software other than Ryzen Master shows the actual 0.2-0.3V low current idle that these CPUs sit at when they're idling along at 38C as they are in BIOS. We've already been over all this. If Auto is in the box next to Vcore, it tells you all you need to know.

You know you could, with the help of the search box, read into the wealth of information on TPU and elsewhere regarding how Ryzen behaves at idle, why monitoring programs think Matisse idles at "1.5V", how monitoring programs negatively affect that idle to varying degrees by constantly "waking" it from an idle state when they poll the CPU, and how the CPU automatically decides how to scale voltage with load.

As an aside, perhaps a more productive problem to be addressing, Be Quiet coolers aren't known to be the absolute best performing SKUs and the high core count chips do require at least high end air coolers to perform to their fullest. Even the Dark Rock Pro 4 which I have is consistently a step behind its D15 competitor, and the Dark Rock 4 is another step behind, and getting close to the edge of what I'd put on a 3900X. That's not to say that it'll struggle or overheat, far from it since you could run the 3900X with the Wraith Prism if you wanted, but Matisse's stock boosting is highly sensitive to temperatures approaching even 75C, and combined with high ambient temperatures can easily account for 50-100 points in Cinebench. I can crank my AC down 5C and gain 50 points without changing anything else.

 

[Max(IT)]

The exact same 0608/0805/1002 BIOSes that you're looking at, minus the red ROG colour scheme?

I'm on BIOS 1002.

Just because they change hardly means they're "real time". I don't know how many times I have to repeat myself. My BIOS only ever shows me 1.450 and 1.472 Vcore, both of which are horseshit because no software other than Ryzen Master shows the actual 0.2-0.3V low current idle that these CPUs sit at when they're idling along at 38C as they are in BIOS. We've already been over all this. If Auto is in the box next to Vcore, it tells you all you need to know.

I'll take your words for true.

You know you could, with the help of the search box, read into the wealth of information on TPU and elsewhere regarding how Ryzen behaves at idle, why monitoring programs think Matisse idles at "1.5V", how monitoring programs negatively affect that idle to varying degrees by constantly "waking" it from an idle state when they poll the CPU, and how the CPU automatically decides how to scale voltage with load.

I don't know why you are making these assumptions
I'm using CPU Z and/or HWInfo64 as monitoring programs (CPUZ for Voltage was indicated by an AMD engineer as a good application to see voltage and clock speed without affecting idle state) and those programs are indicating me that the CPU is staying around those voltages, with spikes at 1.5 V.
It is not only the idle state that worries me, it is also the voltage under load.
Maybe I'm overthinking it, but temperatures are surely high on default AUTO setting.

As an aside, perhaps a more productive problem to be addressing, Be Quiet coolers aren't known to be the absolute best performing SKUs and the high core count chips do require at least high end air coolers to perform to their fullest. Even the Dark Rock Pro 4 which I have is consistently a step behind its D15 competitor, and the Dark Rock 4 is another step behind, and getting close to the edge of what I'd put on a 3900X. That's not to say that it'll struggle or overheat, far from it since you could run the 3900X with the Wraith Prism if you wanted, but Matisse's stock boosting is highly sensitive to temperatures approaching even 75C, and combined with high ambient temperatures can easily account for 50-100 points in Cinebench. I can crank my AC down 5C and gain 50 points without changing anything else.

Yep, that's a good point.
I was aware of that when I choose the Dark Rock 4, but reading a lot of reviews I was under the impression it still was a good choice as an air cooler. Maybe the extremely hot temperatures of these days are affecting boost performance, and surely I will test it again in one week, when the weather here in the Netherlands will return to "normality".
It could be that the undervolting that I'm using now (-0.075 V) won't be necessary in a colder weather because I'm operating in the range of 70/85º CPU temperature that, as you said, is affecting boost.

 

[tabascosauz]

It is not only the idle state that worries me, it is also the voltage under load.

Yep, that's a good point.
I was aware of that when I choose the Dark Rock 4, but reading a lot of reviews I was under the impression it still was a good choice as an air cooler. Maybe the extremely hot temperatures of these days are affecting boost performance, and surely I will test it again in one week, when the weather here in the Netherlands will return to "normality".
It could be that the undervolting that I'm using now (-0.075 V) won't be necessary in a colder weather because I'm operating in the range of 70/85º CPU temperature that, as you said, is affecting boost.

85C when? At all-core load? That's a bit high, but not out of the realm of possibility for a cooler that lies somewhere in the middle between a 212 EVO and U12A in performance. Especially considering your room is 28C. Up to 75C, it can still be debatable whether your BIOS/firmware or CPU internal voltage limits are the factors holding you back, but past 80C you definitely lose speed on account of temperature.

It's not a bad cooler; it's just that Ryzen behaves differently in a thermal sense. Regardless of what people say about running 3900X on the Wraith Prism and 3950X on air coolers, can =! should, so if you want to get all the performance you paid for then water is the only way. A top-end air cooler can definitely get you very close on a 3900X, though.

As long as Vcore is somewhere in the 1.3-1.35V range under full load with default settings, there doesn't look to be anything out of the ordinary. 3900X really isn't a cool chip. If your case is as restrictive as it looks + you haven't added fans + your room is 28C = nothing looks too off.

Manually overclocking is always a surefire way to reduce idle temperatures and noise, but you should probably find out your chip's safe voltage before you start, which you can do through Prime95 Small FFTs with all AVX checkboxes disabled. At that point, it's completely up to the quality of your CPU the kinds of all-core speeds you can get out of that safe voltage or less.

Some have experience with overclocking in RM instead, some say that they can Manual OC set a max never-exceed Vcore in RM and have the system respect that while still otherwise boosting as normal. I haven't been able to replicate that "feature" with many different boards and firmware, so ymmv.

 

[Max(IT)]

85C when? At all-core load? That's a bit high, but not out of the realm of possibility for a cooler that lies somewhere in the middle between a 212 EVO and U12A in performance. Especially considering your room is 28C. Up to 75C, it can still be debatable whether your BIOS/firmware or CPU internal voltage limits are the factors holding you back, but past 80C you definitely lose speed.

85º is the maximum spike I saw, during a CB20 all cores at default AUTO voltage, but temperature was around 75º for the whole test. While gaming the CPU stays around 60/63º.

It's not a bad cooler; it's just that Ryzen behaves differently in a thermal sense. Regardless of what people say about running 3900X on the Wraith Prism and 3950X on air coolers, can =! should, so if you want to get all the performance you paid for then water is the only way. A top-end air cooler can definitely get you very close on a 3900X, though.

According to my research the Dark Rock 4 is enough for a 3900X and much better than a 212 EVO! I had a 212 EVO on a i5-9600K and you cannot compare the twos. The be Quiet! is much bigger than the 212 Evo and it weights more than twice !

According to the tests I saw it is comparable to a U12A (especially at max fan speed)

As long as Vcore is somewhere in the 1.3-1.35V range under full load with default settings, there doesn't look to be anything out of the ordinary. 3900X really isn't a cool chip. If your case is as restrictive as it looks + you haven't added fans + your room is 28C = nothing looks too off.

My case is ok (it was ok for an overclocked 9700K @ 5 GHz so it surely isn't an issue).
Surely unusual ambient temperatures are playing a role, but I will take a further look to Vcore under full load.

BTW my CB20 results seem to be fine (507 single and 7200 multicore are in line with every review I've read about 3900X) and the clock speeds seem to be fine too: during CB20 multicore I read 4/4.1 GHz and during CB20 single core most of the time is 4.4/4.5 GHz.
Does it sound good ?

Manually overclocking is always a surefire way to reduce idle temperatures and noise, but you should probably find out your chip's safe voltage before you start, which you can do through Prime95 Small FFTs with all AVX checkboxes disabled. At that point, it's completely up to the quality of your CPU the kinds of all-core speeds you can get out of that safe voltage or less.

I'm not planning to manual overclock the CPU.
I just want it to safely runs at default clock using PBO if possible.

Right now PBO is doing nothing, maybe due to the high ambient temperature.

 

[tabascosauz]

Is that 507 score after the undervolt? 3900X chips should be at least in the 510s if not higher. My 3700X is a mediocre bin amongst its peers and does 509 easy at an effective clock of 4.325-4.35GHz on the working core. If it's with the undervolt, that's probably the loss of single thread perf coming into play.

See, the issue about cooler reviews is that every site comes up with different results. Not too much a fan of Tom's. Most reviews I've seen put the U12A competitive with or slightly better than my Dark Rock Pro 4, with a number of them putting it within striking distance of the D15. DR4 is most definitely not beating the DRP4 with 1 less heatpipe, less fin stack area, and one less (larger) fan.

The other issue is that the results you've provided are with a 5930X, a physically large soldered chip with a larger(?) die on a larger process with low thermal density. Same goes for the 9700K, both have "less" silicon than a 3900X. You will find that 1) Intel chips draw a lot of power but run cool for that power draw and are thermally more predictable than Ryzen and 2) on Ryzen the differences between coolers are shrunk considerably because of the issue of getting heat out of the tiny chiplets and into the heatsink in time.

Vcore will fluctuate slightly during a test as the initial high all core speeds gradually come down, but you will get a good idea of what you're pulling at load.

 

[Max(IT)]

Is that 507 score after the undervolt? 3900X chips should be at least in the 510s if not higher. My 3700X is a mediocre bin amongst its peers and does 509 easy at an effective clock of 4.325-4.35GHz on the working core. If it's with the undervolt, that's probably the loss of single thread perf coming into play.

507 was after the undervolt, but even at AUTO voltage the result was about the same...
My 3900X in single core CB20 is keeping a slightly higher clock than yours at 4.419 GHz

Guru3D obtained 502:

OC3D around 500 points:

Extremetech 503 points:

Maybe it's not the best silicon but it seems to be just fine

I will check if a lower ambient temperature would improve things.

See, the issue about cooler reviews is that every site comes up with different results. Not too much a fan of Tom's. Most reviews I've seen put the U12A competitive with or slightly better than my Dark Rock Pro 4, with a number of them putting it within striking distance of the D15. DR4 is most definitely not beating the DRP4 with 1 less heatpipe, less fin stack area, and one less (larger) fan.

Noctua D15 surely is better, bu I wasn't looking for such an huge cooler.
I bought DR4 because of many reviews saying it is one of the best "mid-size" cooler.

Here it is quite close to the U-12A (single fan) at full speed.
Long story short, my cooler isn't the best (I knew that since the beginning ) but it is not holding my CPU back, most probably.

Vcore will fluctuate slightly during a test as the initial high all core speeds gradually come down, but you will get a good idea of what you're pulling at load.

I've noticed that.

 

[tabascosauz]

Mine is a 3700X......if your 3900X can barely match mine in ST and sports a nearly 0.2GHz single core deficit at stock compared to what is advertised, then something's blatantly wrong with your BIOS settings or board itself. Matisse not meeting advertised box speeds was completely fixed about 8 months ago...

Clear CMOS and start fresh, see if anything changes. Make sure your chipset drivers are up to date from AMD's website.

Half of the reviews you linked are launch day reviews. Like I said, early AGESA prior to 1.0.0.3ABBA had serious boosting issues, leading to the whole commotion about "Ryzen not meeting advertised speeds". Today, all reviewers are standardized on 1.0.0.4B or v2 and there is, firmly, no way a 3900X scoring below 510 stock is "normal". Notice the results for all the Matisse CPUs:

As to your cooler, I have no doubt you would see higher all-core clocks under water, but it's not the culprit behind your single core clocks.

 

[Max(IT)]

I'm a PITA I know but I did further testing
I tried a CMOS reset and default settings (only DOCP enabled for XMP profile 3600 CL16).
Cinebench results were much lower: 6853 multi, 489 single.

I stand my position: the motherboard is giving too much voltage and it affects temperature (in my situation of cooler and ambient temperature).
I monitored Voltage and clock speed, and it was around 4.3 GHz single core and 3.9 GHz multi, while temp was hitting 82º (in multicore) before settling down at 80º. Voltage initially was 1.27 V (single) and 1.38 V (multi).

With a -0.075 V undervolt the situation was much better.
Cinebench results: 7188 multi, 507 single (consistent).

Clock speed was 4.5 GHz single (this alone is very important IMHO. I'm almost reaching the max rated freq, which indicates the CPU is working within limits) and 4.1 GHz multi (almost 200 MHz more than default voltage). Temperature never exceeded 80º, settling down at 78º (multicore).
I observed the voltage and it was 1.25 (single) and 1.35 V (multi).

As far as I can see temperature is the culprit here.
Without undervolting I'm hitting temperature limits that starts throttling back voltage and thus frequency in order to keep the CPU below 80º.
With undervolting I'm giving my CPU a little thermal headroom to reach slightly higher clock speeds (about 100/150 MHz).

I'm curious about results with a 8/10º lower ambient temperature.
A better air cooler (Noctua D15) or a custom loop liquid cooler could improve the situation, for sure, but considering the high cost involved and the very little improvement I'm not planning to do that

Mine is a 3700X......if your 3900X can barely match mine in ST and sports a nearly 0.2GHz single core deficit at stock compared to what is advertised, then something's blatantly wrong with your BIOS settings or board itself. Matisse not meeting advertised box speeds was completely fixed about 8 months ago...

Clear CMOS and start fresh, see if anything changes. Make sure your chipset drivers are up to date from AMD's website.

Half of the reviews you linked are launch day reviews. Like I said, early AGESA prior to 1.0.0.3ABBA had serious boosting issues, leading to the whole commotion about "Ryzen not meeting advertised speeds". Today, all reviewers are standardized on 1.0.0.4B or v2 and there is, firmly, no way a 3900X scoring below 510 stock is "normal". Notice the results for all the Matisse CPUs:

View attachment 165278

As to your cooler, I have no doubt you would see higher all-core clocks under water, but it's not the culprit behind your single core clocks.

do you want me to say YOU ARE RIGHT ?


The Techpowerup review you reported above is FROM 1 JULY 2019
Exactly like the others.

BTW I will help you: I found a more recent review , one from July 2020 :

The result is above 510, ok, but we are speaking about a "whole" 11 points, or 2% difference.
I wouldn't speak about "something's blatantly wrong" , especially because they are using a Gigabyte X570 Aorus Master motherboard with 8GB G.Skill FlareX CL14 modules for a 32GB capacity, so the 11 points difference could just be caused by a better system (the cooler was not specified but they usually are using Corsair Hydro H115i RGB Platinum 280mm liquid cooler).

BTW temperature are lower this evening (due to a thunderstorm ) and take a look:



the CPU reached the rated 4.6 GHz during the test



this with a -0.075 V of undervolt.
It was an ambient temperature related "issue".

Tamb is 24º now. It will further improve.


Just to add more data, this review with AGESA 1.0.0.4 has a whole 4 points difference from my CPU in single (and a lower result in multi !). Are you still speaking about "something blatantly wrong" in my setup ?

I just think something is blatantly wrong in the Dutch weather this summer

 

[tabascosauz]

Check yourself before you start going off the rails. That graph is from the 3900XT review, with up-to-date results for the older Matisse parts. XT SKUs came out a month ago.

I don't know what your major malfunction is, but I'm here trying to help you figure out why your 3900X is performing on default settings like a 3900 MT and 2700X ST. The vast majority of users don't need to set a -0.075V offset to achieve expected performance on up to date firmware. No need to throw at me for offering suggestions to help you fix your shit.

Was going to suggest that you physically bridge the CMOS jumper instead of using BIOS to load defaults because I've personally had weird behaviour on old boards that can only be fixed that way, and that CB doesn't care about memory speed, but it looks like you're pretty confident in yourself over the conclusions that others have come to over a year of ownership. Good day to you.

 

[Max(IT)]

Check yourself before you start going off the rails. That graph is from the 3900XT review, with up-to-date results for the older Matisse parts. XT SKUs came out a month ago.

You were right about the graph, my bad, but the result changed 6 points from the initial review to the final results of one months ago, so the point stands: 510 vs 526 is a negligible difference speaking about two different systems with two different motherboards and different RAM.

I don't know what your major malfunction is, but I'm here trying to help you figure out why your 3900X is performing on default settings like a 3900 MT and 2700X ST. The vast majority of users don't need to set a -0.075V offset to achieve expected performance on up to date firmware. No need to throw at me for offering suggestions to help you fix your shit.

Was going to suggest that you physically bridge the CMOS jumper instead of using BIOS to load defaults because I've personally had weird behaviour on old boards that can only be fixed that way, and that CB doesn't care about memory speed, but it looks like you're pretty confident in yourself over the conclusions that others have come to over a year of ownership. Good day to you.

there are a LOT of review of 3900X achieving worst results than mine, and you completely ignore the results I posted above.
quite ridiculous your comment about "2700X single core results". You are just trying to downplay other people. You started with the cooler, then the case ....
510 is a ST results perfectly in line with 3900X CPU.
Not a new record, for sure, but it was quite expected with a Dark Rock 4 single fan cooler.
And the funny part is you wrote "everything below 510 is not normal", and when I reached 510 you completely ignored that and my CPU are offering "2700X ST results"...

And the link above about AGESA 1.0.0.4 results in line with mine ? Ignored again...

And, AGAIN, you didn't even read my post:

I tried a CMOS reset and default settings

English might not be my first language, but I think is quite clear that I bridged the jumper for a CMOS reset.

You are not giving ANY suggestion. Since the beginning you are just standing your position that you are right and everyone else is wrong.

And, by the way, this is a well known website with updated results for cinebench R20 scores:

https://www.cgdirector.com/cinebench-r20-scores-updated-results/

I will help you: Ryzen 9 3900X results are 511 ST - 7100 MT

Or this, even lower:

https://nanoreview.net/en/cpu-list/cinebench-r20-scores


So much for a "2700X single core results" ...

 

[moproblems99]

you could run the 3900X with the Wraith Prism if you wanted

I can assure anyone who has questions, you don't want to.

 

[Dos101]

Been following this thread since I have been struggling with my 3900XT, I can't seem to hit anywhere close to the CB R20 scores that the 3900XT and 3900X are getting in reviews. With everything stock in the BIOS except XMP enabled, and Ryzen High Performance power plan set, I can get 6963 in CB R20 multi-core, but with temps hovering around 80c (my ambient temp is ~22c). Single core score is 485 and never hits 4.7Ghz on any core, highest I see it go is 4.5Ghz. Setting PBO from auto to enabled drops my multi-core score to 6866 but with temps only around 73c, but single core score bumps up to 496 and I do actually hit 4.7Ghz on one of the cores. I have played around with negative offsets but I end up with better temps, slightly lower multi core score, and much lower single core scores. These results have been consistent through the last 3 BIOS revisions on my X570 Aorus Master.

Is there a gold standard of what BIOS settings should be used with 3900X/3900XT?

 

[Max(IT)]

Been following this thread since I have been struggling with my 3900XT, I can't seem to hit anywhere close to the CB R20 scores that the 3900XT and 3900X are getting in reviews. With everything stock in the BIOS except XMP enabled, and Ryzen High Performance power plan set, I can get 6963 in CB R20 multi-core, but with temps hovering around 80c (my ambient temp is ~22c). Single core score is 485 and never hits 4.7Ghz on any core, highest I see it go is 4.5Ghz. Setting PBO from auto to enabled drops my multi-core score to 6866 but with temps only around 73c, but single core score bumps up to 496 and I do actually hit 4.7Ghz on one of the cores. I have played around with negative offsets but I end up with better temps, slightly lower multi core score, and much lower single core scores. These results have been consistent through the last 3 BIOS revisions on my X570 Aorus Master.

Is there a gold standard of what BIOS settings should be used with 3900X/3900XT?

did you try 1usmus Power Plan ?
I don't know how much it improves performance because I'm using it since the beginning, but everyone told me to use it.
From my multiple tests I can say that undervolt can help if you are somehow temperature limited (approaching 80º the CPU starts dropping voltage and clock speed) but I found the "sweet spot" between -0.075 and -0.0625 V.
More and you are slightly affecting single core performance, less and there is no visible results.
In my case PBO isn't helping at all.