5800x (and other Zen 3 chips) PBO settings/Temperature fix

[Mussels]

That much i could think of too , higher clocks = higher voltage but the best cores need more voltage for the same speed , i guess they are programmed with lower VIDs from the factory, i thought they all came with the same VID.

I know from blasting my 1400 and 1700x that not all cores would OC the same, or even come close. Some would do 400MHz higher than others at the same voltage, and the highest clocking ones were not always the best undervolting ones either.

It could be they're the closest to the power delivery and have more stable voltages, or the opposite (the furthest away could have smaller transients, smoothed by the distance) - this stuff would need AMD engineers to know and they don't like sharing without lots and lots of alcohol.

(Disabling cores in the BIOS and OCing with just 1-2 cores active - sketchy barely stable stuff on early AM4)

 

[izy]

I know from blasting my 1400 and 1700x that not all cores would OC the same, or even come close. Some would do 400MHz higher than others at the same voltage, and the highest clocking ones were not always the best undervolting ones either.

It could be they're the closest to the power delivery and have more stable voltages, or the opposite (the furthest away could have smaller transients, smoothed by the distance) - this stuff would need AMD engineers to know and they don't like sharing without lots and lots of alcohol.

(Disabling cores in the BIOS and OCing with just 1-2 cores active - sketchy barely stable stuff on early AM4)

I had an 1700x too and it was the same for me , some cores wouldnt go higher so the max all core OC had to be done to what the bad cores could do , i remember i had one bad core that wouldnt go higher even i was pumping more volts but maybe its different for zen3 or zen4 even.
If its not like what this guys said that "The best cores have lower VIDs for each frequency step" and all steps are the same maybe the factory tune for 5700x was made like for a 5800x or something , i just tested my 5700x in OCCT all core load memory test, they all boost to 4725 - 4750 with the same vids , 1,300v - 1,313v - 1,319v - 1,325v - 1,331v, all the cores fluctuating on this values the same way.

(they all go 4850mhz , i started the all core test before all of them got to boost to max)

 

[Thelost41]

Benim de 1700x'im vardı ve benim için de aynıydı, bazı çekirdekler daha yükseğe çıkmıyordu, bu yüzden kötü çekirdeklerin yapabileceği maksimum tüm çekirdek OC'nin yapılması gerekiyordu, ben bile daha yükseğe çıkamayan kötü bir çekirdeğim olduğunu hatırlıyorum. daha fazla volt pompalıyor ama belki zen3 veya zen4 için farklıdır.
Bu adamların söylediği gibi değilse, "En iyi çekirdekler her frekans adımı için daha düşük VID'lere sahiptir" ve tüm adımlar aynıysa, belki 5700x için fabrika ayarı 5800x veya benzeri bir şey için yapılmış olabilir, 5700x'imi az önce OCCT'de test ettim. çekirdek yük hafızası testi, hepsi aynı videolarla 4725 - 4750'ye yükseliyor, 1,300v - 1,313v - 1,319v - 1,325v - 1,331v, tüm çekirdekler bu değerlerde aynı şekilde dalgalanıyor.
View attachment 326583View attachment 326585View attachment 326584
(hepsi 4850mhz'e gidiyor, hepsi maksimuma çıkmadan önce tüm çekirdek testine başladım)

What is your Cinebench R23 score in these settings?

The scores I got in my own system with all cores at 4.8 ghz 1.300v are as follows:

 

[izy]

What is your Cinebench R23 score in these settings?

The scores I got in my own system with all cores at 4.8 ghz 1.300v are as follows:

I dont have CB R23 now but i have a pic from when i tested a time ago, think i had the "bad cores" at -27 back then and only 2 stick of RAM at lower CL.

 

[Waldorf]

Any gains for gaming raising limits?
i have no prob on the hw/power/temp side, so i tried a couple of things to find out max draw (180 PP, 200 PPT, 160 TDC, 200 EDC, +50mhz, 1x.)
dont want to mess with CO (for now), just trying to tweak things a bit, not going for lowest (temps/power),
but not interested in wasting things either (e.g. +20% power for 1% in perf).

once i have a gimbal (phone) or drone again, i will start doing editing/encoding, but for now its mainly gaming.

 

[izy]

Any gains for gaming raising limits?
i have no prob on the hw/power/temp side, so i tried a couple of things to find out max draw (180 PP, 200 PPT, 160 TDC, 200 EDC, +50mhz, 1x.)
dont want to mess with CO (for now), just trying to tweak things a bit, not going for lowest (temps/power),
but not interested in wasting things either (e.g. +20% power for 1% in perf).

once i have a gimbal (phone) or drone again, i will start doing editing/encoding, but for now its mainly gaming.

Gaming can benefit more from CO + AutoOC (higher frequency) , i think you can safely do a -15 all core if not even -20 and you should use AutoOC +200 , i see no reason why not, with only -50 AutoOC you prob can even do -25 -30 and you should leave the scalar on Auto (it reports as 1x anyway) , more draw its not more performance in general , there is a sweet spot, my PBO is 140W PPT - 95A TDC - 120A EDC, it doesnt go more than 80C-84C in any demanding all core stress tests even over night from what ive tested, more like 79-80C.

 

[Thelost41]

Proving to him who the boss is

 

[Waldorf]

@izy
already seen more than 1x and 25-50Mhz offset dont gain much.

just not running the same chip as you, and i expect a little more variation between cores,
at least from the perf spread i have seen during ram (HCI) testing.
most stuff runs on 2K/4K, so less gains there anyway.

temps arent the issue (custom loop), its more that i dont want to have to go back and forth between
changing settings/testing for (possibly) "unnoticeable" gains.

 

[Kursah]

I have some work to do and may need some pointers and guidance.

But first, I wanted to ask if I'm at a loss because I'm on an X370 board (system specs are up-to-date)?

I've never really invested time tuning this build...originally bought it used off of TPU years ago with a 2700X and some G.Skill DDR4 3000 that eventually went bad. Snagged a 5800X and 32GB TEAMGROUP T-Force Vulcan Z DDR4 4000 (currently at 3600 CL18). For what it is, it has been pretty decent and super reliable.

I'm running a Noctua U14S (2013 edition) with a 2nd fan, I've loved this cooler since I purchased it for my 4770K then 4790K back in the 2013/14 timeframe. I do have spare Thermalright Peerless Assassin 120 SE ARGB I may test at some point. On the Noctua, I kick up to 100% fan speed at 60C. Updated my MB's BIOS to 6203 to the latest which includes AGESA 1.2.0.A.

I did adjust the curve to -30. PBO is enabled, which is default, as-is most everything else. I'll be looking into messing with the other settings to keep my temps down (I got up to 83C on Cinebench 2024 and 86C on Cinebench R23, but this is also with the case fans on my Corsair 600C at low speed.).

Memory has been touch and go, can't get DOCP for DDR4 4000 to post, nor when I try manual 3800. 3600 I can get, but not with CL16. Granted I only tried for about an hour last night, and only messed with core timings and DDR voltage up to 1.40v. ATM I'm at 3600 at 18 20 20 20 42 1.35v. Nothing special.

I admit great ignorance in tuning these...but now that my son's both have upgraded Intel-based systems (in sig), it'd be nice to get a little more pep in my PC's step. I'm going to go back through this topic from the OP. I don't expect to keep up in actual benches or gaming performance necessarily, but I'd like to get this thing tuned to do better than it is now if possible. I know leaving PBO enabled for OC makes sense... I'll spend some more time tuning memory.

I ran Cinebench 2024 and R23. Was kinda disappointing to see I couldn't match the 5800X on that, but I know YMMV and running an older board may have me at a disadvantage. Also, this was kind of a baseline run, as I've been running with this general setting for a bit now...most recently updating the BIOS last night and tightening timing from 18 22 22 44 to 18 20 20 42.

 

[Mazer]

I set -10 all core left everything else on auto (scalar, no boost override, default mobo settings) it seems stable, even did time spy run no crashing or instability with -10 to all cores, my other 5600x did -18 all cores but I don't think it's going to work on this 8 core chip on a single tower cooler hitting 83c.

 

[cyx2111]

So my friend. You have good score in CB but your temps still higher. Best way UnderVolting Per Core not All cores for more performance and better temperatures.
On my 5800X i use PBO2 - UnderVolt "PER CORE" not all cores bcs any core behaves different.

Core1 to Core5 -28 - Core6: -5 (sensitivity core) and the 2 best cores Core7-8: -18. I have max 76c in benchmarks and max 67c on any AAA Game with RTX 3070 Ti OC (3440x1440p)
PPT: 120W - TDC: 85A - EDC: 130A using one of the best Top3 mobo B550 platform the Asus RogStrix B550 E-Gaming (not F). Αlso Ι use it on all my computers my favorite brand of AIO the Arctic.de
The Arctic Freezer III 360 (New model - Hot price right now 85€ only via website) - Arctic Freezer II 420 - Artic Freezer II 360.

PBO Boost: Disable
Scalar: Auto or x1 (Better Auto)
Global C-State: Disable
Power Options in Windows: CPU Power min 5% - 100% max (for better idle temps)
AIO Curve: 25C - 40%
50C - 60%
75C - 100%

My Best score in CBR23: 15.748

U can try to UnderVolting the 5800X with some programs to find and select the best undervolt for any core.
Εach processor (5800Χ) it's different, it works and behaves differently...you understand me.

HarrisK.
Tech Writter of X-Treme PC Build
Ookla Analytics

 

[izy]

I am using more or less same settings on my 5700x as @cyx2111 from a long time now and it works great, setting "per core" CO is what you want to use if you have time to play with it.

 

[cyx2111]

I am using more or less same settings on my 5700x as @cyx2111 from a long time now and it works great, setting "per core" CO is what you want to use if you have time to play with it.

Yes my friend need time to make it per core. More Time and patience. For the 5700x with the Arctic Freezer II or III no need UnderVolt. Only PBO Disable. Οf course it assumes good case and good airflow.