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

[tabascosauz]

I'm not sure what AMD were thinking with the default voltage curve on these. I was also hitting 90c out of the box on stock. MSI has a 'Kombo Strike' option in their latest BIOS that provides three all core CO options. I now get 80c under full load using level 3. Supposedly that's a -30.

I learned about the Kombo Strike thing earlier today and kinda wish Asus had it too. Though, the convenience of tweaking CO and limits on-the-go in Windows is just too good to pass up.

Something to try before using CO is to try negative offset on the CPU in the BIOS then test if R23 score are going higher or lower.
Once you get the highest score with my 5800X its offset 0.0500V this gets the highest score any lower or higher performance drops.

Then use CO on top of that and you should have better performance.

I will definitely try that. Just still trying to figure out how this CPU behaves - it seems to behave differently in a lot of ways than any other Zen2 or Zen3 CPU I've come across. Some people on reddit and OCN were speculating that the cache is calling the shots on how boost behaves. Makes some sense. ST boost is very lazy compared to other Zen3. Whereas 5900X might spike close to Fmax just opening an application (lightning fast), 5800X3D just kinda sits around and so the difference is definitely felt.

I have not succeeded in hitting 1900Mhz Fabric yet. Tried a lot of things, just won't train. Funny because at 1800 the VSOC and VDDG characteristics are pretty much identical to my 5900X but it just refuses to train 3800.

 

[The King]

I learned about the Kombo Strike thing earlier today and kinda wish Asus had it too. Though, the convenience of tweaking CO and limits on-the-go in Windows is just too good to pass up.



I will definitely try that. Just still trying to figure out how this CPU behaves - it seems to behave differently in a lot of ways than any other Zen2 or Zen3 CPU I've come across. Some people on reddit and OCN were speculating that the cache is calling the shots on how boost behaves. Makes some sense. ST boost is very lazy compared to other Zen3. Whereas 5900X might spike close to Fmax just opening an application (lightning fast), 5800X3D just kinda sits around and so the difference is definitely felt.

I have not succeeded in hitting 1900Mhz Fabric yet. Tried a lot of things, just won't train. Funny because at 1800 the VSOC and VDDG characteristics are pretty much identical to my 5900X but it just refuses to train 3800.

I am also having an issue getting 1900 FCLK boot consistently with 4 dimms. With 2 DR DIMMS no problems even 2 SR DIMMS it boots fine. You may need to adjust ProcODT

 

[tabascosauz]

I am also having an issue getting 1900 FCLK boot consistently with 4 dimms. With 2 DR DIMMS no problems even 2 SR DIMMS it boots fine. You may need to adjust ProcODT
View attachment 263542

Yeah, no bueno I think. Tried most of the procODT options. I'm out of tricks at the moment - I am up to 3733 though. No problem at all, no WHEAs, VSOC is still low and a logical increase, just don't know why it refuses to train 3800.

The static Vcore offset needs a bit more testing - I was already kinda losing perf at -0.025. Temps were pretty good, with no need for setting limits really.

 

[QuietBob]

Ran it today in a 22c room, with CO -30 on all cores:

With 25c ambient maximum Tctl/Tdie was 83c stock and 75c with CO during 30min stability test.

 

[The King]

So apparently PB02 tuner thinks my 5800X has 16 cores.

 

[Mussels]

@Mussels I need some help here please.

Leave it to me to buy a 5800X the week before ZEN 4 launches.

So the issue I am having is that this CPU wont do PB0 +50 without bluescreening. Both my 5600 and 5600X can do PB0 +200 and -CO on top of that.
Stock perofmance seems slighly lower than the TPU review. So far I am not that happy with this CPU but it could be because it behaves differently from my 5600 and 5600X.

Here are the core ratings in HYDRA. With PBO disabled. So is it normal for a 5800X to not be able to do PB0 +50? I have tried +25 it can boot into windows but +50/+75 etc mostly unstable crashed before windows loads or after I log in. I am considering requesting a return and exchange on Amazon but before I do that I wanted to check with someone who has more experience with this CPU.

View attachment 262780View attachment 262781

Don't forget that +anything is overclocking and may not mix with any undervolting at all, curve or static

Hydra never worked great for me either, it's optimised for certain boards and glitchy on the rest

Just wondering, but why do all of you bother overclocking CPU's? I imagine most of you are gamers, and 100-300 mhz extra doesn't change anything fps wise for games. So why waste time navigating it all?
I used to find overclocking fun, but that was on the i5-2500k when I could see actual gains in real world usage.

Because we get more for our money?

Theres something to be said for finetuning what you own, it's the same reason i grease the chain on my garage roller door - it's perfectly usable while squeaking but why not spend time and make it better at what it does?

You're right that overclocking isn't what it once was, these days it's more about getting the most performance within your wattage limit rather than sheer clockspeed.
I've got my 4.6GHz overclock locked in, shaving a good 50W off the PBO alternative, 30C less and still gaming away easily at 4k 144


At stock settings, i'm the same (or worse) performance for over 150W more power consumption

 

[The King]

Don't forget that +anything is overclocking and may not mix with any undervolting at all, curve or static

Hydra never worked great for me either, it's optimised for certain boards and glitchy on the rest

First day I got the CPU I wanted to see what Boost override would work at stock. (CPU AUTO) Only +25Mhz would go into windows and run stable without BSOD.
I was abit worried about that when I posted that because both my 5600 and 5600X can do +200MHz overide no problems.

After a few days of tinkering I can get +100Mhz to boot into windows and run stable by having CPU voltage -.0.0500V offset.

Now I run PBO disabled in the BIOS so no Boost overide no CO in the BIOS only -0.0500V. Everything else is set with Hydra and I have no issues. +16K in R23. CPUZ ST 680 MT 6811 on Air.

Overall Im very happy with this CPU. Gaming performance is also really great, Games do seem to run abit smoother compared to my 5600/X

 

[Zach_01]

First day I got the CPU I wanted to see what Boost override would work at stock. (CPU AUTO) Only +25Mhz would go into windows and run stable without BSOD.
I was abit worried about that when I posted that because both my 5600 and 5600X can do +200MHz overide no problems.

After a few days of tinkering I can get +100Mhz to boot into windows and run stable by having CPU voltage -.0.0500V offset.

Now I run PBO disabled in the BIOS so no Boost overide no CO in the BIOS only -0.0500V. Everything else is set with Hydra and I have no issues. +16K in R23. CPUZ ST 680 MT 6811 on Air.

Overall Im very happy with this CPU. Gaming performance is also really great, Games do seem to run abit smoother compared to my 5600/X

Did you check for any WHEA?

 

[The King]

Did you check for any WHEA?

WHEA is fatal when it comes to CPU OC. I immediately get BSOD if clocks are unstable or CO.

I managed to get +200Mhz Override to run eventually it was just one core that was boosting way too high. I adjusted that and now its fine.

 

[tabascosauz]

WHEA is fatal when it comes to CPU OC. I immediately get BSOD if clocks are unstable or CO.

I managed to get +200Mhz Override to run eventually it was just one core that was boosting way too high. I adjust that and now its fine.

4.9 effective is already pretty good, man. You're pretty close to my 5900X (~4.93 for 687 CPU-Z and 1640 R23). Sub 4.7GHz Zen 3 is crazy consistent now but I find once you get to about 4.8GHz it's all a crapshoot, 5800X/5900X/5950X all beating or losing to each other on 1T clock left and right, SP is still too random.

My 5800X3D seems to be making it easy for me, -25 is stable and -30 seems alright too before testing. 5900X took me like 2.5 weeks to fully test because the distribution is like -2/5/30/10/20/15/20/20/20/20/20/20. Wack.

 

[Andrea87]

Following on this thread, I've experimented a little bit more with my 5800x.

Reduced the total power allowed to 100W, TDC 75A and EDC 90A. Max Boost +100MHz, offset voltage -0.0625V. Coming from 110W/80A/100A, I've seen almost no drop in cinebench - 15248 vs 15300ish, while temps fell a solid 5-6°C. Went from 67-68°C to 62.5 max in a 20°C Room.

I'll stability test the system properly with a few OCCT runs, then start tuning the memory. My brother has the same kit, should be able to hit 3600MHz without too much hassle.

 

[Mussels]

First day I got the CPU I wanted to see what Boost override would work at stock. (CPU AUTO) Only +25Mhz would go into windows and run stable without BSOD.
I was abit worried about that when I posted that because both my 5600 and 5600X can do +200MHz overide no problems.

After a few days of tinkering I can get +100Mhz to boot into windows and run stable by having CPU voltage -.0.0500V offset.

Now I run PBO disabled in the BIOS so no Boost overide no CO in the BIOS only -0.0500V. Everything else is set with Hydra and I have no issues. +16K in R23. CPUZ ST 680 MT 6811 on Air.

Overall Im very happy with this CPU. Gaming performance is also really great, Games do seem to run abit smoother compared to my 5600/X

As a sidenote to this, my RAM overclock is unstable when my CPU passes 75C
Raising my SoC voltage made the RAM OC higher, but also raised the CPU temps bringing that instability back in


Part of the reason you're struggling could be higher ram clocks, higher ram rank counts, or the IMC's temps when you had PBO on

You can do +200Mhz, but limit the PPT and cTDP (hard temp limit)

 

[The King]

As a sidenote to this, my RAM overclock is unstable when my CPU passes 75C
Raising my SoC voltage made the RAM OC higher, but also raised the CPU temps bringing that instability back in


Part of the reason you're struggling could be higher ram clocks, higher ram rank counts, or the IMC's temps when you had PBO on

You can do +200Mhz, but limit the PPT and cTDP (hard temp limit)

It was just one core that was causing windows to bluescreen with +200mhz. I had to give it a slight + to the CO on that 1 core.

In windows the other cores were boosting 4800 but core 2 was boosting 4925Mhz with the same voltages. so 125mhz higher than the other cores. when CO is all 0

 

[Mussels]

I bought an x3D, so i'll be having all sorts of fun soon

 

[Falcon216]

I've spent a large amount of my last two weeks' free time delving into how my particular new-in-box 5800X works with my particular mobo (Asus X570 Prime-Pro) and AIO (Rev.6 420 no RGB).
Coming from a quite capable 3600 non-X which I understood very well, it's been a bit of a journey getting used to Zen3's behavior. Some of it I like, some of it could be better managed, and I believe a lot of it comes from how the motherboard has been programmed.

My particular parts combo and environment gets ~80C @ 28Cambient with a 135PPT/100TDC/125EDC.
What you'll notice in the R23 pic is I have AOC + 125Mhz -and- all-core CO -30.
I've read an All-CO -30 is guaranteed unstable, but it passes P95 blend test for 12 hours, and it passes hours of complex CPU heavy game tests like 2077, SOTR, and Civ

Which leads to my main point:
Some mobos must have implemented truly awful volt/clock behavior, like mine apparently, which is why a CPU in one board can overheat and run the same speed at 75C in another.
I've found the self reported core VIDs to be accurate for testing stability under load (they change based on how hard the core is hammered, so 1.12v for 4.4Ghz in R23 might need to be 1.18v in Prime95, like my 5800X), but my vcore out and thus SVI2 are always defaulting way higher than expected.
I've found Ryzen Master reports one voltage under load and the telemetry shows something completely different.
I've tried setting a negative offset to make SVI2 match requested core VID and it wrecked performance. I tried a positive offset and it throttled the chip.
None of my telemetry matches what the CPU indicates it wants from it's VF curve, and nothing I do or try in the BIOS, which is not a manual voltage, can match the Core VID tables/VF curves to actual supplied voltage.
It's baffling, astounding, and downright frustrating.

So if you've eliminated all variables like cooling, paste, mounting, and your BIOS options, any problem with temps and power for your expected clockspeed is likely the motherboard was programmed like trash and is the source of your hard times.
I can essentially only rely on PBO2+Boost offset with power limits, with CO maxed out -30 on all cores, to get heavy MT workload settings anywhere close to what I spent literal days checking and stability testing with p95 and R23.

Oh and then there was the time I tried to RAM overclock with DOCP enabled, not once, but three times. And had to reset my BIOS by touching the board jumpers until I figured out what I was doing wrong.
But that's just me

 

[Zach_01]

Individual core VIDs are somewhat useless.
The CPU request is only one and its the "CPU Core VID (effective)" below the SVI2 TFN sensor (by default sensor order)

 

[BetrayerX]

As soon as I installed my 5800X, noticed how bad the temps where. I switched from a single tower cooler to a double tower one and still it hits 95C FAST on Cinebench. With Ryzen Master, I selected the Auto Overclock w/100+ on the speed and a Per Core optimization. LLC is set to 4 and that's all on the BIOS. Damn thing took like 75 mins to run and a few reboots.

After it got done, I kinda got the best of both worlds. Lower voltage and power consumption at idle and low workloads (everyday use) and a scary 140W power draw on high workloads. Speeds top at 4950 with the small detail that I see max temps on the low 80s instead on full workloads, compared to stock 95C (and I wonder if it throttled, didn't check, because it got hot fast). Pretty happy with the results.

 

[freeagent]

Why do you have llc on if you have PBO enabled?

 

[The King]

Even if you can do -30 ALL core CO that does not mean that you should do that and expect the best performance. Some cores actually loose performance going close to -30

VS higher - CO

 

[freeagent]

If you lose performance getting close to -30 then you should increase power limits

 

[The King]

If you lose performance getting close to -30 then you should increase power limits

PBO is on PPT 1000W.
Can't go any higher EDC and TDC at max too. Also LLC3 the secret sauce behind it all.

To break 16K I had to use Hydra. The above R23 scores are not done using Hydra.

 

[DemonicRyzen666]

If you lose performance getting close to -30 then you should increase power limits

I'm already 240 watts on 5600x. I do feel like this cpu could be better from lapper of direct die cooling as two cores are seemingly hotter then the others for no real reason. I already tried remounting 5 times.

 

[freeagent]

I'm already 240 watts on 5600x. I do feel like this cpu could be better from lapper of direct die cooling as two cores are seemingly hotter then the others for no real reason. I already tried remounting 5 times.

Try 200/140/180 all core -29 and see what happens, you should be getting 4850 solid.

 

[Mussels]

I've spent a large amount of my last two weeks' free time delving into how my particular new-in-box 5800X works with my particular mobo (Asus X570 Prime-Pro) and AIO (Rev.6 420 no RGB).
Coming from a quite capable 3600 non-X which I understood very well, it's been a bit of a journey getting used to Zen3's behavior. Some of it I like, some of it could be better managed, and I believe a lot of it comes from how the motherboard has been programmed.

My particular parts combo and environment gets ~80C @ 28Cambient with a 135PPT/100TDC/125EDC.
What you'll notice in the R23 pic is I have AOC + 125Mhz -and- all-core CO -30.
I've read an All-CO -30 is guaranteed unstable, but it passes P95 blend test for 12 hours, and it passes hours of complex CPU heavy game tests like 2077, SOTR, and Civ

Which leads to my main point:
Some mobos must have implemented truly awful volt/clock behavior, like mine apparently, which is why a CPU in one board can overheat and run the same speed at 75C in another.
I've found the self reported core VIDs to be accurate for testing stability under load (they change based on how hard the core is hammered, so 1.12v for 4.4Ghz in R23 might need to be 1.18v in Prime95, like my 5800X), but my vcore out and thus SVI2 are always defaulting way higher than expected.
I've found Ryzen Master reports one voltage under load and the telemetry shows something completely different.
I've tried setting a negative offset to make SVI2 match requested core VID and it wrecked performance. I tried a positive offset and it throttled the chip.
None of my telemetry matches what the CPU indicates it wants from it's VF curve, and nothing I do or try in the BIOS, which is not a manual voltage, can match the Core VID tables/VF curves to actual supplied voltage.
It's baffling, astounding, and downright frustrating.

So if you've eliminated all variables like cooling, paste, mounting, and your BIOS options, any problem with temps and power for your expected clockspeed is likely the motherboard was programmed like trash and is the source of your hard times.
I can essentially only rely on PBO2+Boost offset with power limits, with CO maxed out -30 on all cores, to get heavy MT workload settings anywhere close to what I spent literal days checking and stability testing with p95 and R23.

Oh and then there was the time I tried to RAM overclock with DOCP enabled, not once, but three times. And had to reset my BIOS by touching the board jumpers until I figured out what I was doing wrong.
But that's just me

curve undervolt goes unstable at idle, not at load


and yes some boards overvolt massively, which leads to the crazy variance in thermals

You may have more luck with the LLC controls instead of the offset, as the CPU performance is based on requested voltage (VID) which offsets adjust while LLC does not

 

[tabascosauz]

I've spent a large amount of my last two weeks' free time delving into how my particular new-in-box 5800X works with my particular mobo (Asus X570 Prime-Pro) and AIO (Rev.6 420 no RGB).
Coming from a quite capable 3600 non-X which I understood very well, it's been a bit of a journey getting used to Zen3's behavior. Some of it I like, some of it could be better managed, and I believe a lot of it comes from how the motherboard has been programmed.

My particular parts combo and environment gets ~80C @ 28Cambient with a 135PPT/100TDC/125EDC.
What you'll notice in the R23 pic is I have AOC + 125Mhz -and- all-core CO -30.
I've read an All-CO -30 is guaranteed unstable, but it passes P95 blend test for 12 hours, and it passes hours of complex CPU heavy game tests like 2077, SOTR, and Civ

Which leads to my main point:
Some mobos must have implemented truly awful volt/clock behavior, like mine apparently, which is why a CPU in one board can overheat and run the same speed at 75C in another.
I've found the self reported core VIDs to be accurate for testing stability under load (they change based on how hard the core is hammered, so 1.12v for 4.4Ghz in R23 might need to be 1.18v in Prime95, like my 5800X), but my vcore out and thus SVI2 are always defaulting way higher than expected.
I've found Ryzen Master reports one voltage under load and the telemetry shows something completely different.
I've tried setting a negative offset to make SVI2 match requested core VID and it wrecked performance. I tried a positive offset and it throttled the chip.
None of my telemetry matches what the CPU indicates it wants from it's VF curve, and nothing I do or try in the BIOS, which is not a manual voltage, can match the Core VID tables/VF curves to actual supplied voltage.
It's baffling, astounding, and downright frustrating.

So if you've eliminated all variables like cooling, paste, mounting, and your BIOS options, any problem with temps and power for your expected clockspeed is likely the motherboard was programmed like trash and is the source of your hard times.
I can essentially only rely on PBO2+Boost offset with power limits, with CO maxed out -30 on all cores, to get heavy MT workload settings anywhere close to what I spent literal days checking and stability testing with p95 and R23.

Oh and then there was the time I tried to RAM overclock with DOCP enabled, not once, but three times. And had to reset my BIOS by touching the board jumpers until I figured out what I was doing wrong.
But that's just me

-30 is not "guaranteed unstable", stable CO is different for every CPU sample. You just need to figure yours out for each core through OCCT/corecycler if you really want to get to the bottom of it. For CO testing, you're completely wasting your time with Prime95 Blend.

VID is not useful. SVI2 TFN should only be relied on if you're under significant load (whether single core or otherwise). On the Prime-Pro you don't have another reliable die sense Vcore to rely on ("Vcore" from the Richtek controller is BS).

All DOCP/XMP does for mem OC is set a couple of voltages and major timings for you. It's your choice whether to set it or not set it when manually working on your RAM profiles, if your BIOS is made properly then it should not make a difference.

In general, your Vcore looks kinda weird. But it's hard to tell as you've scrambled the order of all the HWInfo entries. What BIOS are you running?