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

[outpt]

i THINK it is the cpu going bad!????
it passes all test trying to isolate the problem ie. cc,memtest,occt,sofroth and so on.
have gone back to a earlier bios version
set bios to opimized settings bsod/black screen reset. the cpu still boost to 4.85ghz. disable pbo, it will still boost to 4.85ghz.,but it is rock stable. what the hell is going on?

 

[mtcn77]

what should be the voltage on the 12v rail?

More than 11.4v.

have gone back to a earlier bios version
set bios to opimized settings bsod/black screen reset.

As you can tell, much of the cpu's drivers come on board the bios in ryzen series. You get a uniform performance across various platforms without having to sort out motherboard OEM brand differences, so don't expect different bios versions to be the same default as before. They do change the motherboard characteristics significantly.

 

[outpt]

is it not strange that pbo enabled gives 4.85ghz and bsod, but disabling pbo still gives 4.85ghz with no bsod. this is what i do not understand.

 

[mtcn77]

is it not strange that pbo enabled gives 4.85ghz and bsod, but disabling pbo still gives 4.85ghz with no bsod. this is what i do not understand.

Yes, pbo sets hundreds of profiles and profile switches which one of them is more likely to crash then your oversupplied one profile. I'm not even beginning about the curve crap...

 

[tabascosauz]

is it not strange that pbo enabled gives 4.85ghz and bsod, but disabling pbo still gives 4.85ghz with no bsod. this is what i do not understand.

It's not strange at all. I had it on certain early BIOSes on the B550M TUF and the B550M Steel Legend with the 3700X, PBO was unusable due to instability. Vendors often don't do a very good job of implementing AGESA so various features are wonky or broken. Well, it's a bit strange that it happens this late in the lifecycle, but even the current AGESA 1.2.0.3b hasn't been so kind to me either, poor CPU performance on the B550I Gaming Edge Wifi, and piss poor iGPU performance on the B550I Aorus AX.

Yes, PBO shouldn't be unstable, but there are no guarantees.

PBO isn't guaranteed to give you more boost. Boost Override is what gives you more frequency over the default global freq limit, but it still depends on what your cores are capable of hitting. eg. 5900X has a ceiling of 5150MHz at +200, but mine can only do 4900-4950ish, so I don't run PBO since it already achieves those speeds at +0; 5700G has a ceiling of 4850MHz at +200, but mine top out at about 4775.

 

[outpt]

going to leave settings as they are. it seems stable of course i have said that before. was hoping for 5ghz just not going to happen so its no big deal. might get a 5900x to play with.

 

[mama]

I am surprised that you don't have more instability with an edc that low.

Maybe I have it all wrong but I've settled on PPT 185 TDC 125 EDC 170 for my 5900x. I've also slapped -25 on all the cores with PBO2 except for the first and second cores for each CCD which are at -2. +50MHz on the clock and it runs super stable and cool. Clocks boost to 4950MHz.

 

[mtcn77]

I am surprised that you don't have more instability with an edc that low.

Maybe I have it all wrong but I've settled on PPT 185 TDC 125 EDC 170 for my 5900x. I've also slapped -25 on all the cores with PBO2 except for the first and second cores for each CCD which are at -2. +50MHz on the clock and it runs super stable and cool. Clocks boost to 4950MHz.

If you don't post it while running, the residency won't be reported properly.

PS: I also don't see any core temperature measurements and too much information hiding the fact.
If you think 70°C is good, think what 60°C will become. You gain a lot of performance if you limit the temperature ramp by allowing only enough EDC for a single ccx to turbo. You can reduce quite a lot of temperature fluctuations and voltage discrepancies if you decrease EDC down to TDC.

 

[mama]

If you don't post it while running, the residency won't be reported properly.

PS: I also don't see any core temperature measurements and too much information hiding the fact.
If you think 70°C is good, think what 60°C will become. You gain a lot of performance if you limit the temperature ramp by allowing only enough EDC for a single ccx to turbo. You can reduce quite a lot of temperature fluctuations and voltage discrepancies if you decrease EDC down to TDC.

I'm not hiding anything. Just played Doom Eternal for an hour at 4K:

 

[Mussels]

i THINK it is the cpu going bad!????
it passes all test trying to isolate the problem ie. cc,memtest,occt,sofroth and so on.
have gone back to a earlier bios version
set bios to opimized settings bsod/black screen reset. the cpu still boost to 4.85ghz. disable pbo, it will still boost to 4.85ghz.,but it is rock stable. what the hell is going on?

BIOS could be the issue with its PBO settings messing with your LLC settings or something similar?

 

[outpt]

BIOS could be the issue with its PBO settings messing with your LLC settings or something similar?

Any type pbo, co,ctr and the system becomes unstable or bios. Just going to leave as is. That 5900x beginning to look better.

 

[Mussels]

Any type pbo, co,ctr and the system becomes unstable or bios. Just going to leave as is. That 5900x beginning to look better.

I've heard of other people having that with various boards

Are you using the +200MHz (or lower) offset? I can imagine higher clocks causing issues, PBO itself should not break anything unless the BIOS has problems

 

[mtcn77]

I'm not hiding anything. Just played Doom Eternal for an hour at 4K:

I cannot offer you any more help, Core 3 and 5 seems to be the heaviest duty runners. If they run on the same ccx(I believe), you already run with the lightest pbo setup and there is no further suggestion for me.

Had it been otherwise, I would suggest you run with a lower edc until either ccx ran with cpb and this might help with decreasing the temperature load availing better pbo activity.

Always iterate on that: temperature ramp is slower with a lower EDC limit. You might play around with a higher PPT to insure steady supply. You have a real high ratio of inactive versus active thread count. Big chips could take a lower edc to ppt ratio for this instance. I would go as low as 70% in order to test it out. Try 140A EDC, 200w PPT and slowly approach them to your regular setting, imo. If you use shortcuts while testing, you might get a result free of idle zone and this might help you figure which setting operates the cores at the highest average frequency. Even minute differences at the average cpu clock might be significant.

 

[outpt]

I've heard of other people having that with various boards

Are you using the +200MHz (or lower) offset? I can imagine higher clocks causing issues, PBO itself should not break anything unless the BIOS has problems

As long as pbo is disabled there are no issues. pbo enabled is a crash fest. I just don’t know what the problem is. Have gone to amd forums and read all kinds of theory’s some are out there. bios changes do not resolve the issue. Really considering a rma and getting a 5900x.
These crashes have gotten much worse over time. When I first installed the processor I could use curve optimizer without problems, not so anymore. Maybe the cpu was never stable to begin. Gaming is where all of the problems are so far.

 

[mtcn77]

These crashes have gotten much worse over time.

You only need to explain what the settings are. LLC, scalar and what not... you definitely have something wrong.

 

[jayjr1105]

Man what a beast PBO is if you put the time into dialing it it. Spent about 3 days tinkering and I think I'm finally happy with performance and thermals. For those interested I'm using a 5600X with a ID Cooling SE-225-XT cooler in push pull config. PBO enabled with an undervolt in curve optimizer set to -5 in BIOS. PPT 100, TDC 70, EDC 90

This is a load with OCCT (large data set, avx2 and normal mode)

 

[Mussels]

Yup! It's all about finding the right point to limit the chip, so you get 95% of all that performance, at half the damn wattage!

 

[outpt]

BIOS could be the issue with its PBO settings messing with your LLC settings or something similar?

i have update to the latest beta bios for my msi mag b550 tomahawk and it "seems" stable so far. i have set pbo limits to ppt=115,tdc=90,edc=110. no CO or added pbo oc. there have been no bsod,lockups etc. gaming is the acid test so far so good.

this chip has been out a while why are there just figuring out these bios bugs. this is crazy.

 

[mtcn77]

i have set pbo limits to ppt=115,tdc=90,edc=110

I think we have told a couple of times not to do that. EDC should be 90% of PPT at most. TDC does not have to be more than 70 since it is an emergency setting. What you are trying to do is push record highs while running into issues.

 

[jayjr1105]

So I messed with the curve optimizer some more. Set to -15 and started getting some errors in OCCT. Set down to -10 and still a few errors but less. Is this the only way to clear these errors up? Does adjusting EDC/TDC/PPT do nothing for these? Can I force PBO to just not boost as high? I'd be happy with a 4.5 all core tubo instead of 4.6.

I guess if I bring PPT down it will naturally downclock on all core loads correct?

 

[mtcn77]

So I messed with the curve optimizer some more. Set to -15 and started getting some errors in OCCT. Set down to -10 and still a few errors but less. Is this the only way to clear these errors up? Does adjusting EDC/TDC/PPT do nothing for these? Can I force PBO to just not boost as high? I'd be happy with a 4.5 all core tubo instead of 4.6.

I guess if I bring PPT down it will naturally downclock on all core loads correct?

No. Curve optimizer pushes all bins to a lower voltage afaik? You are trying to run the chip with a huge vdroop as it practically won't.
If you want to decrease fluctuation play with EDC-PPT, not with the curve. You'll make the chip hit adaptive clock vdroop correction more frequently if you play with the curve.

 

[jayjr1105]

No. Curve optimizer pushes all bins to a lower voltage afaik? You are trying to run the chip with a huge vdroop as it practically won't.
If you want to decrease fluctuation play with EDC-PPT, not with the curve. You'll make the chip hit adaptive clock vdroop correction more frequently if you play with the curve.

I'm a little confused by your reply. So set the curve back to -5 like I had it a couple days ago and then play with PPT/EDC?

 

[mtcn77]

Can I force PBO to just not boost as high?

You want it to slow in ST or MT? ST:EDC, MT: PPT, but readjust EDC under the threshold too if you tune PPT because you will lose its throttling along with PPT if you lower the PPT margin(EDC won't respond if PPT is the limit, will try to overshoot ST boost). Basically, watts are 2 orders less restrictive than voltage w=i²*r. (I know I said voltage when EDC controls ampere).

 

[outpt]

So I messed with the curve optimizer some more. Set to -15 and started getting some errors in OCCT. Set down to -10 and still a few errors but less. Is this the only way to clear these errors up? Does adjusting EDC/TDC/PPT do nothing for these? Can I force PBO to just not boost as high? I'd be happy with a 4.5 all core tubo instead of 4.6.

I guess if I bring PPT down it will naturally downclock on all core loads correct?

Do not go all core you will lose gaming performance. I have mine set to: ppt-120,tdc-70,edc-103. These should lower temps. but keep turbo to 4.85ghz. There is no need to add +200 to your pbo oc. unless the cpu can do more. this works well for me as my chip does not like co plus 200 to pbo
Turn off CO.

 

[mama]

I'm a little confused by your reply. So set the curve back to -5 like I had it a couple days ago and then play with PPT/EDC?

I like your fox avatar by the way.

I suggest you reset CO2 settings to 0. Set the PPT, TDC and EDC to the default settings for the chip. Ideally start with a fresh BIOS. Try to find the curve in PBO2 for your chip to begin with. Remember that the best performing cores will need more voltage so less of an undervolt. You will know you have gone too far when you get WHEA errors on certain cores. You can identify the core creating the error by using Eventviewer and CPUZ report. Pull back the undervolt on those cores which create errors until you get a stable system. Cinebench is okay for testing for errors but ideally you want to run the system in general use for a while. After some testing, you might want to look at adding 25-200MHz to the clocks to see if a higher clock will hold. Once you are happy then move on to PPT, TDC and EDC. Step up the values but try maintaining the default ratios. Use Cinebench to see if the changes you make are beneficial until you reach a point where you are not comfortable with the temperatures. As another member has pointed out to me, controlling temperature is the key.

This process is not a quick fix and I'm sure others do it differently but this is the way I approach it if it's of any help.