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

[Mussels]

The 5800x is well known for having heat issues on stock settings, because it's single 8 core CCX design shares the same power limits as the 2x6 (dual 6 core CCX) 5900x - so it's got a lot less space to cool itself with.

A big part of the problem is that motherboards will raise the maximum TDC, EDC and PPT values to effectively infinite values the moment you enable PBO, such as when you're trying out curve undervolting for the first time. I made a lengthy post on the issue over here.
Make sure you manually enter your CPU's default PBO limits as a starting point, if you're new to this.

A Zen 3 chiplet (whether it has 4 cores or 8) is 80.7mm square -modern intels are over 200mm square. Four times the surface area to cool with.
It doesn't matter how big a heatsink you slap on top, what matters most is the quality of the contact, and the quality of heatsink base.

All the dual CCX Zen 3 chips look like this, with three sources of heat (lower is the IO die, upper two are the CPU chiplets)
A 5950x with two 8 core chipsets, a 5900x with two 6 core chiplets and a 5800x with one 8 core chiplet all have the same 140W PPT target - but your motherboard might have that limited to 105W, 142W, or 400W depending on your PBO settings.

It's simple to imagine that a 5800x having the wattage of both CCX's shoved into just one is going to be harder to cool and fussier about the heatsink contact on top.
You can either fight with a really high-quality cooler to get the contact exactly right... or tweak the PBO values and stop the chips from sending so much power out. You could throw 30W more for less than a 1% boost, so why not trim the excess?




When compared to intel, you can see they have a MUCH larger surface area to spread the heat out over, which is why they can report lower temps despite having much higher wattages



Skipping the explanation, heres some recommended settings from our various forum users and myself!
Regardless of which chip you have, you can simply do some minor tweaks to match your chip better: if these settings work super cold (like they do for me on custom water) - you can simply up the values for more performance if you want it. Just make sure to scale them together

Single threaded temperatures are often higher than multi threaded, because Ryzen 5000 reports the peak temperature of the hottest core - not an average of all of them. It's key to note that these temperatures are updated every 1 millisecond, so those spikes are VERY brief.

Due to a current AGESA bug, keep 65W CPU's EDC below 90A and all higher wattage CPU's below 140A, to prevent a bug that downclocks the CPU.

5600x

Spoiler

5600x settings per Zach_01
Default limits of 5600X are:
PPT: 76W
TDC: 60A
EDC: 90A


And some higher performance settings:
PPT: 90W
TDC: 75A
EDC: 100A

5800x
Exclusively with this chip due to heat density you'll find very diminishing returns past 100w PPT - expect lapped heatsinks and liquid metal for high values.
100-115W seems to be where people have the best results, curve offsets allowing higher clocks at the lower wattages

Spoiler

GerKNG's highly efficiet 5800x settings are as follows:

Max efficiency:
PPT: 95W
TDC:60A
EDC: 90A
56C, 4.2GHz all core load, 4.9GHz boost
Note how the PBO values all match at 100% all together - Gherkin King here really tuned these in!




And then I tested some higher values, since i have custom water cooling on mine:
You'll notice that getting even 100Mhz more, adds 10C


5800x custom water
+200Mhz
Curve optimiser: -10 all core
PPT: 110W
TDC: 90A
EDC: 100A
R23: 4.45GHz all core, 63C (4.925GHz 1T)



5800x custom water
+200Mhz
Curve optimiser: -10 all core
PPT: 120W
TDC: 75A
EDC: 110A
These were tuned in like Gerkngs so they all hit 100% together to please my OCD
R23: 4.5-4.55Ghz all core, 69.5C - (5.025GHz 1T)

Spoiler: Matt Ferris 5800x

For each setup, curve optimizer was: -23 on best cores; -30 on "bad" cores
Initial Setup

• PPT: 142
• TDC: 95
• EDC: 125
• Max Temp Multi-Core: 82C
• CB23: 15134

Second Test

• PPT: 124
• TDC: 85
• EDC: 127
• Max Temp Multi-Core: 76C
• CB23: 14957

Third Test

• PPT: 120
• TDC: 75
• EDC: 110
• Max Temp Multi-Core: 69.8C
• CB23: 14653

Fourth Test

• PPT: 117
• TDC: 80
• EDC: 120
• Max Temp Multi-Core: 71.3C
• CB23: 14862

5900x

Spoiler

Stock is:
PPT 142
TDC 95
EDC 140

Older BIOS without the 140A EDC bug:
(Thanks FreeAgent)

PPT: 240
TDC: 160
EDC: 190

5900x and 5950x with AGESA past 1.2.0.3c need EDC at 140A or less, but you can still raise PPT (to what your VRMs and cooler can handle) and TDC to less than 140A.

These should work on any Zen 3 chips, but watch out for differences in the CCX and core counts.
2x6 cores on a 5900x has far more surface area to cool with, vs a single 8 core CCX on a 5800x - so it can handle higher values.


Two more tips follow:
1. Watch out for two seperate PBO settings on some motherboards, on my Asus board i have one in 'Ai Tweaker' and one in advanced. Only the AI tweaker settings work and apply, the Advanced ones are ignored (except for curve optimiser, which is only visible there)

2. Curve optimiser is great, but watch out for random crashes at idle or moderate load situations. Some of my cores go to -30, some will only go to -5 without issues. Be patient, and only test this when you're 100% sure your system is already stable, or you could waste a lot of time with testing it out. I ended up settling on -13 all core, instead of tweaking them all individually.
I m about to follow some advice here and start using core cycler to test out the offsets in an automated way
GitHub - sp00n/corecycler: Stability test script for PBO & Curve Optimizer stability testing on AMD Ryzen processors

 

[freeagent]

on my Asus board i have one in 'Ai Tweaker' and one in advanced. Only the AI tweaker settings work and apply, the Advanced ones are ignored (except for curve optimiser, which is only visible there)

Sweet man!

I just wanted to note that I do my overclocking from the Advanced menu, not AI tweaker..

They fixed my board a few flashes ago

Edit:

If I use both menus I get a high clock but lower effective by a couple hundred MHz

 

[birdie]

PPT/TDC/EDC
105/70/95: A little faster than stock, about 70C on my 5800x

This absolutely tanks performance in CB R23.

By default: ~15500, with these settings ~14100.

I've settled on PPT 125W and that's it. Now CB R23 shows ~15100 and I'm OK with that. Temps are lower.

There's a discussion about the best PPT/TDC/EDC settings on Reddit

Spoiler: reddit

https://www.reddit.com/r/Amd/comments/kfpele

but I've tried their best settings (120/75/110) and they are not very good.

 

[Mussels]

You can simply tune the settings to suit your needs and temp preferences

In my case, i get my full 5.05GHz single core boost, and lost only 75MHz off my multi threaded - this DOES seem to vary board by board, so you gotta adjust.

 

[mtcn77]

I think I have told on occasion that PPT has to be 10% higher than EDC which is what you have done.

Ryzen Master Performance Boost Overdrive EDC vs. PPT showdown

Hi, https://www.reddit.com/r/Amd/comments/cnykxh/3700x_frequencyvoltajepower_on_pbo_custom/ I just found this on Reddit which I happened to consider to be useful for comparison's sake between EDC vs. PPT settings. Normally, you would think power regulation via PPT should fare better. That is...

www.techpowerup.com

Spoiler: 10%

On r7, the option to decrease edc 10% less than ppt is found to work best.

Spoiler: 10%

Previous observations developed the understanding that 10% lower EDC than PPT is a good threshold to start.

Unless you want single threaded performance, you don't need EDC all that much either and even if you are overclocking you should only make PPT 10% higher than EDC, no more. Again, these are best practice measures.

First thing is first, ppt has to be 10% higher than edc.

If you are looking for ST scores you clamp PPT, and otherwise for MT you keep PPT just 10% above EDC.

I suggest decreasing EDC from its 140 default to 130 such that there is a 10% gap from PPT which was a key point in former Ryzen generations.

 

[freeagent]

For fast single core stuff like superpi 32m I use 200/130/130. It will ignore those settings and take what it wants when it comes to multi core. For instance yes 200/130/130 is set, but it will take 215/140/180. Stock is 143/140/90 I think. I might have some of those numbers mixed up tdc and edc always trip me up.

 

[mtcn77]

For fast single core stuff like superpi 32m I use 200/130/130. It will ignore those settings and take what it wants when it comes to multi core. For instance yes 200/130/130 is set, but it will take 215/140/180. Stock is 143/140/90 I think. I might have some of those numbers mixed up tdc and edc always trip me up.

It is all correct since you said single thread in the beginning.

 

[GerKNG]

This absolutely tanks performance in CB R23.

By default: ~15500, with these settings ~14100.

weird i have almost your stock 15500 with the 95W "as efficient as possible" Powerlimits.

 

[birdie]

weird i have almost your stock 15500 with the 95W "as efficient as possible" Powerlimits.

I'm quite sure you didn't just set PPT to 95W, you have a lot more optimizations.

 

[GerKNG]

I'm quite sure you didn't just set PPT to 95W, you have a lot more optimizations.

95/60/90 (and +50Mhz which has no influence on the score (it hurted single core))
1X Scalar, no other changes.
the only thing that the user should do is playing with the curve optimizer.
but i did not tweaked anything else.

 

[birdie]

the only thing that the user should do is playing with the curve optimizer.

This is like crazy important and decides everything and you're glossing over it as if it's nothing.

Again, even with "95/60/90 (and +50Mhz)" the multithreaded benchmarks results will be abysmal and the performance loss will be extreme.

 

[GerKNG]

This is like crazy important and decides everything and you're glossing over it as if it's nothing.

Again, even with "95/60/90 (and +50Mhz)" the multithreaded benchmarks results will be abysmal and the performance loss will be extreme.

the curve optimizer is mentioned in mussels post.

but in my case even an unstable -20 across all cores makes a difference of 1-2 percent.

stock vs stock and Curve optimizer vs curve optimizer with these limits results in the same gains.

 

[freeagent]

Honestly most of the time I just run @ stock clocks with -30 set and no pbo. Runs great even with plain old 3200c14. 3D runs at the top of the clock scale, heavy duty stuff drops the clocks a bit but it’s still very productive.

 

[mtcn77]

3200c14

That's overclocker grade expansion.

 

[freeagent]

That's overclocker grade expansion.

They can boogie

They are almost wasted on AMD..

CO rox

 

[OolongGoat]

I don't have much issues with CB23..... My first run was in the 19200 range and my 2nd one attached is in the 18800 range. Seems good to me.
I am simply using the "a little faster than stock" settings with a Negative 15 on my PBO Curve Optimizer on all cores. ** I think when I opened and used the Snip & Sketch tool, might have played a role in the drop of my score, but close enough. **

 

[tabascosauz]

I don't have much issues with CB23..... My first run was in the 19200 range and my 2nd one attached is in the 18800 range. Seems good to me.
I am simply using the "a little faster than stock" settings with a Negative 15 on my PBO Curve Optimizer on all cores. ** I think when I opened and used the Snip & Sketch tool, might have played a role in the drop of my score, but close enough. **

That's quite a bit of a drop for a 5900X, why limit it to 105W? Bone stock 5900X @ 142W is in the 21000 range, MT loads run pretty cool even on a bog standard air cooler. I'm sitting at a hair under 22000, all on air, with a variable per-core CO. It's a bit different for them 5800X owners, but we don't really have their heat problems here on the 12-core, single-thread is what runs hotter at stock.

You don't have to run 10 min worth of CB R23 just to get a score. You can go to File > Advanced Benchmark, turn off minimum test duration.

 

[Tomgang]

I can´t really talk much about temp on zen 3 yet. As i have yet to complete my build based on a Ryzen 5 5600X and a Ryzen 9 5950X. Bofh CPU´s are air cooled. So i have only had a little time to test them at stock and with a prime 95 test with the most heat generating test. As said this is with all at stock settings. i have only activated momory XPM profile. Else nothing is touched, so i have yet to fine tune. The below temps are right out of boks with stock voltage.

Cooling setup:
5600X is cooled by a low profile noctua nh-l9x65 se-am4 cooler. With fan a round 1900 RPM it hits 70-71 degress celcius and with the fan at 2500 RPM it comes down to 67 degree celsius with a prime 95 test in the most heat generating test. The little fan is even covered a little by the GPU. Ilde at 45 degree celsius with the fan at 1000 RPM. Some peeple even told me this little cooler, cut not handle a 5600X, But at stock it handles it just fine. All throw this little cooler do limit the overclock potential. But overclock whas never the idea with the 5600X any way.

5950X is cooled by a Noctua NH-D15 chromax black and stock fans has been replaced by Noctua IPPC industrial 3000 RPM fan. One 140 MM and one 120 mm. Here i have so far only testet with the fans a 1600 RPM and not the maximum 3000 RPM. But with a prime 95 test it stabilized around 55 degree celsius and my gues with 3000 RPM temp can get down to around 50 degree celsius. Ilde sits at 35 degree celsius with fans a 1100 RPM.

This is the exsperience i have had so far with Zen 3. So i really dont have a heat issue. Again i so also have the CPU with the lowest TDP and the CPU with the best binned chiplets and most cores to spread the wattage out on. So that will off cause help on the heat.

Some screenshots and a image of my nearly completed system, to give an idea of what´s going on at how airflow is in my system.

 

[outpt]

I have not played with optimizing each core,but as little as -5 in CO causes system to become unstable. so i have turned it off and set pbo2 to +100mhz. it now boosts to 4.95ghz with 4.9ghz being more common

edit: doesn't like pbo2 either guess i have a bum chip!

 

[Mussels]

I have not played with optimizing each core,but as little as -5 in CO causes system to become unstable. so i have turned it off and set pbo2 to +100mhz. it now boosts to 4.95ghz with 4.9ghz being more common

edit: doesn't like pbo2 either guess i have a bum chip!

dont rule out that it could be something else borderline stable, like SoC voltage, DRAM voltage, and so on.

 

[outpt]

ok ppt=125,edc=90,tdc=70. pbo2=+50. 4.9ghz stable so far.
pbo2 @ +75mhz yielded 4.95 ghz but unstable in SoTTR.

 

[turbogear]

@Mussels If one is using custom cooling then one of the problems is that most of the water blocks are optimised for cooling hotspot in the middle of the CPU, but on Ryzen 3000/5000 there is an offset due to chiplet design.

One of the solution is to consider this Der8auer offset kit:

Drop Temperatures on RYZEN 3000 CPUs: der8auer RYZEN 3000 OC Bracket

der8auer OC Bracket: http://bit.ly/38NACVq---------------------------------------------------------Support me on Patreon: https://www.patreon.com/der8auer---...

www.google.com

There are some blocks like TechN that is specifically designed für Ryzen 5000 taking the chiplet design and offset of hotspot into consideration:

TechN CPU Waterblock against Watercool Heatkiller IV Pro and EK Quantum Magnitude - AMD’s Zen3 may come | igor´sLAB

The new CPU waterblock from TechN (here as AM4 version for 99 Euro RRP) has to prove itself in today’s test with the quasi classic cooler in the form of the Watercool Heatkiller IV Pro AMD (from…

www.igorslab.de

I changed from XSPC RAYSTORM to TechN and I was able to reduce about 4°C in Cinebench on multiple thread when I had 5800X Curve Optimized.

I don't have the CPU now anymore in my computer as I went to 5900X but I am still using TechN bock.
The 5800X will go to my son's computer in a few weeks when I have the other components for building it up.

 

[Mussels]

@Mussels If one is using custom cooling then one of the problems is that most of the water blocks are optimised for cooling hotspot in the middle of the CPU, but on Ryzen 3000/5000 there is an offset due to chiplet design.

One of the solution is to consider this Der8auer offset kit:

Drop Temperatures on RYZEN 3000 CPUs: der8auer RYZEN 3000 OC Bracket

der8auer OC Bracket: http://bit.ly/38NACVq---------------------------------------------------------Support me on Patreon: https://www.patreon.com/der8auer---...

www.google.com

There are some blocks like TechN that is specifically designed für Ryzen 5000 taking the chiplet design and offset of hotspot into consideration:

TechN CPU Waterblock against Watercool Heatkiller IV Pro and EK Quantum Magnitude - AMD’s Zen3 may come | igor´sLAB

The new CPU waterblock from TechN (here as AM4 version for 99 Euro RRP) has to prove itself in today’s test with the quasi classic cooler in the form of the Watercool Heatkiller IV Pro AMD (from…

www.igorslab.de

I changed from XSPC RAYSTORM to TechN and I was able to reduce about 4°C in Cinebench on multiple thread when I had 5800X Curve Optimized.

I don't have the CPU now anymore in my computer as I went to 5900X but I am still using TechN bock.
The 5800X will go to my son's computer in a few weeks when I have the other components for building it up.

The derbauer kit doesnt fit with modern EK blocks, and i've got one of the modern AM4 exclusives meant to fix that issue

Trust me, i bought it, drained the loop, disassembled everything and then cried when it wouldnt fit (the screwholes are too small)

 

[outpt]

finally got pbo to work. boost to 4.925ghz. returned all settings to stock except, temp. limit to 80C,VDDP=0.900 and VDDG=0.950,pbo=+75mhz.
the VDDP and VDDG values i got from a thread at guru3d where posters were having random reboots at idle or doing very little when heavy work loads would not crash.
cinebech23 gave me mt=15553 and single thread was about 1600 or so.

 

[mtcn77]

ok ppt=125,edc=90

Wrong setup. If you need these settings your cooling is too slow to respond, or you are pushing your limits.
Again, happy to see you found about undervolt limits, but it still delineates the same issue. You are pushing the chip in opposite directions. First of all, you cannot play with pbo and ppt-edc both. One is an overclocker, the others are undervolter settings. One pushes pb settings up which waste what edc is trying to do, the others alternate between p states which don't overboost to the turbo states.
When I had the puny 6800K I couldn't even apply llc, everything had to be within fixed offset for "maximum vdroop" under load to keep temperature down - how can you guys play around so brazenly, I really have to congratulate AMD engineers for hiding the impact of power output. You should have abstained from "all" scalar, LLC what not, but since you didn't you're constantly troubleshooting, imo.