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

[OneMoar]

This CPU is Rev B2 not sure if that makes a difference. Under Vcore the min is 1.032V and CPU Core Voltage min is 1.013V.
Will have to look at this again tomorrow. Enabling +200 overide did boost the core ratings in hydra from 150 to 158 but boost seems to be still capped at 4800/4850.

5.2 is the max and its exceptionally rare to get a sample that will do that without lots of voltage
4.8 seems to be the bar on the most recent bioss

if you can hold 4.6 thoughout cinebench run without throttling into the ground you are in pretty good shape
that should net you a score of 1530-1580k
if you get to 16k you got a good sample and its likely not going to get the extremely high peak boost clocks

 

[The King]

5.2 is the max and its exceptionally rare to get a sample that will do that without lots of voltage
4.8 seems to be the bar on the most recent bioss

if you can hold 4.6 thoughout cinebench run without throttling into the ground you are in pretty good shape
that should net you a score of 1530-1580k
if you get to 16k you got a good sample and its likely not going to get the extremely high peak boost clocks

Ok that wasn't hard to get im sure this can do more. Thanks for the help and your time will play around with this tomorrow its midnight here.

 

[freeagent]

a 5800x wil and will go right to 95c and throttle into the ground at those settings

My 5600X is ok with it.. 130w with PBO lowest all core is 4625 during Linpack

 

[Zach_01]

@The King those scores are fine whats the vcore doing 0.296 is awful low for idle
which may explain why pbo is crashing if its dropping back to idle and not ramping voltage fast enough
edit:thats a b450 board

I tried this but with +100 and it seems ok now in windows at the moment. Will do some stability testing.


HWINFO also shows idle voltage down to 0.200V is that not normal for this CPU? Yes B450M Mortar MAX
View attachment 262784

When we are talking voltage on a Ryzen3000/5000 at least, we only talking about SVI2 TFN sensor. Everything else is not what the CPU is getting.

CPU-Z only uses the board's sensor (inaccurate) to display voltage.
HWiNFO "Core VID(s)" also is not the CPU/core voltage. These are just a request from the cores.

The final and only voltage request is the "CPU Core VID (effective)" and the true (closer to reality) Core Voltage is the "CPU Core Voltage (SVI2 TFN)". Both polled straight from CPU and most accurate when Snapshot CPU Polling is enabled.

 

[OneMoar]

in my setup vcore =[ svii2 tfn I must have renamed the sensor lol
ops

 

[tabascosauz]

@The King are you still running -25 all core? You're really not guaranteed any amount of undervolt offset. There's not really much correlation between the undervolt offset you can run and the quality of the CPU, it's only relative to what AMD has set the V-F table at for each core.

I wouldn't use the CPPC metric either. On my 5900X the actual SP of each core (esp. after CO) is not the same as the CPPC ranking or ratings that AMD has given.

-25 may be stable at stock because the global limit is 4850, but if you open up override above +0 be prepared to dial back your undervolt if you worked on your CO at 4850 only. At 4850 you are only using a portion of your V-F curve so it can mask instabilities. Closer you get to +200 (5050MHz) V-F curve gets steeper.

Like I've said before, if you want to know your actual idle Vcore look at either Ryzen Master, hope you get lucky watching the Vcore box in CPU-Z, or buy a ROG Crosshair board. VID isn't Vcore, and SVI2 Vcore in HWInfo doesn't care for idle.

 

[Zach_01]

@The King are you still running -25 all core? You're really not guaranteed any amount of undervolt offset. There's not really much correlation between the undervolt offset you can run and the quality of the CPU, it's only relative to what AMD has set the V-F table at for each core.

I wouldn't use the CPPC metric either. On my 5900X the actual SP of each core (esp. after CO) is not the same as the CPPC ranking or ratings that AMD has given.

-25 may be stable at stock because the global limit is 4850, but if you open up override above +0 be prepared to dial back your undervolt if you worked on your CO at 4850 only. At 4850 you are only using a portion of your V-F curve so it can mask instabilities. Closer you get to +200 (5050MHz) V-F curve gets steeper.

Like I've said before, if you want to know your actual idle Vcore look at either Ryzen Master, hope you get lucky watching the Vcore box in CPU-Z, or buy a ROG Crosshair board. VID isn't Vcore, and SVI2 Vcore in HWInfo doesn't care for idle.

I dont understand perfectly what you mean about SVI2 Vcore not caring about idle.

HWiNFO by default has a 2000ms sampling rate. A lot of things happen in 2000ms (2sec) on a CPU. At least 100times (if not more) changing voltage/frequency. From my observations idle voltage can vary a lot from session to session depending on what is happening on background and/or on Systray. I tend to run a lot of things in systray. Even if the CPU is by average using 1core total, frequency and voltage stays high. 1.3~1.4V and 4.2~4.3GHz avg freq for most active cores (peak at 4.95~5GHz). If I close them (apps) almost all and wait enough avg voltage drops to 1.0V and avg (discrete) freq around 4GHz if not lower. Avg active cores are at 0.5 then. These voltage readings on SVI2.

You can set HWiNFO sampling at 1000ms which makes it a little more accurate for shorter times, but not lower like 500ms because this keep the CPU too awake not letting it fall into lower C-States.

 

[tabascosauz]

I dont understand perfectly what you mean about SVI2 Vcore not caring about idle.

HWiNFO by default has a 2000ms sampling rate. A lot of things happen in 2000ms (2sec) on a CPU. At least 100times (if not more) changing voltage/frequency. From my observations idle voltage can vary a lot from session to session depending on what is happening on background and/or on Systray. I tend to run a lot of things in systray. Even if the CPU is by average using 1core total, frequency and voltage stays high. 1.3~1.4V and 4.2~4.3GHz avg freq for most active cores (peak at 4.95~5GHz). If I close them (apps) almost all and wait enough avg voltage drops to 1.0V and avg (discrete) freq around 4GHz if not lower. Avg active cores are at 0.5 then. These voltage readings on SVI2.

You can set HWiNFO sampling at 1000ms which makes it a little more accurate for shorter times, but not lower like 500ms because this keep the CPU too awake not letting it fall into lower C-States.

Sometimes it falls to 0.9V sometimes (maybe lower), but HWInfo is the only software that's way off and thinks that Ryzen regularly stays at 1.4V for 90% of its life. It's really not that hard to get Ryzen to fall to lower (higher) C-states, but it is if you only use HWInfo as a reference.

Deep idle (~0.2V) is harder to get to with any applications open, but it's easy enough to check simulnateously both Ryzen Master and board-based die sense Vcore circuit and see that the CPU is usually hanging out around the 0.9-1.1V mark at "normal" idle (die sense is obviously slower because reliant on HWinfo polling rate, but the two are generally close outside of spikes)...............unless you're about to claim that both of those are wrong and only HWInfo is right.

You can easily see what the CPU waking cores looks like (spikes) in both die sense and RM, they are still rare unless you're actively doing something and don't change the overall average for idle, although RM still catches more of the spikes due to faster polling.

Of course, put some load on it and SVI2 quickly gets in line.

 

[Zach_01]

If it was true the SVI2 wouldn't avg below 1.4V ever. But it does if nothing major is happening, or if any app opened does not cause spikes regularly.

Right now (100min session) I have SVI2 1.31V avg but look at task bar...

If I close half of them SVI2 avg is falling at 1.0V = cant think that 90% of time is 1.4V if avg is 1.0V

5900X is relatively new to me (2+ months) and I haven't see any core request 0.2V (VID) so far like my previous R5 3600. That CPU had much easier lower SVI2 avg. Seems like AGESA, windows or something else is treating the 5900X more aggressively.
Reading, calculation and reporting of CPU parameters (v, f, T) is a top secret in RM and none really knows how it does all that. Few that may know are on life NDA for the AMD proprietary method and probably easy for AMD to spot them if ever disclose anything.

 

[The King]

I did this run in R23 this morning ECO mode 95W custom -CO.

Spoiler: R23

Using Hydra got 16031 but its definitely not optimal a 4650 all core run should be doable.

Spoiler: R23 Hydra OC

 

[OneMoar]

All core is extremely dangerous if you don't know what you are doing and your edc/tdc is set high you can insta-cook a core

 

[The King]

All core is extremely dangerous if you don't know what you are doing and your edc/tdc is set high you can insta-cook a core

The 16031 R23 run was done with 4625 all core temps we around 92 degrees still some headroom left.
I am reaching the limits of my Dual tower Air cooler may make a switch to AIO.

 

[OneMoar]

The 16031 R23 run was done with 4625 all core temps we around 92 degrees still some headroom left.
I am reaching the limits of my Dual tower Air cooler may make a switch to AIO.

the problem with manual/all core is pbo isn't there to keep you from spiking 200A though the core and cooking it
a AIO may or may not help the 5800x is a single CCD, at a certain point you just can't get the heat out of the core fast enough the die is physically to small to transmit enough heat to the IHS

above 12 or 15w per core its just not worth pushing it harder there is no gain except excessive temps and power consumption

 

[freeagent]

the problem with manual/all core is pbo isn't there to keep you from spiking 200A though the core and cooking it
a AIO may or may not help the 5800x is a single CCD,

That will not happen with 1.35v or under.

I can’t do 200a, but 190a is pretty routine on my 5900X, and not a big deal.

 

[Zach_01]

That will not happen with 1.35v or under.

I can’t do 200a, but 190a is pretty routine on my 5900X, and not a big deal.

Only time will/can tell.

Me, you or anybody else can’t determine right now if EDC 180+A is ok in the long term, or if 1.35V is safe.
You can’t just isolate voltage or current or frequency or temperature as operating parameters and say they are ok.
Under what conditions these are safe? How many of them are in excess limits/conditions together? what kind of load? How frequently? For how long?

EMI and silicon degradation has no warning signs unless it’s too late. Keep that in mind before you (all in general) set up those limits too high and push the silicon through benchmarks and stress tests for the bragging rights.
If you need more MT power for real then there are other ways to get it.

Just a friendly reminder.

 

[Space Lynx]

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.

 

[Athlonite]

Just wondering, but why do all of you bother overclocking CPU's?

Because we're all ePeen whores and we like to brag

 

[LifeOnMars]

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.

Indeed, I game at 4k with eco mode on my 5700x

Unofficial answer - We're all after that unattainable perfect level of smooth experience in every game ever created, it feels like that extra 25mhz will give us this. Humans are never happy and strive for more, it's in our DNA.

 

[freeagent]

Me, you or anybody else can’t determine right now if EDC 180+A is ok in the long term, or if 1.35V is safe.

If you have a completely stock CPU, and an Asus board and you enable Asus Performance Enhancement, it will boost your stock EDC to 180, no PBO required., as well as raise your clocks, though not past global limits. So 180 has to be ok for the CPUs that can use it.

or if 1.35V is safe.

I wouldn't run 1.,35 manually 24/7.

Under what conditions these are safe? How many of them are in excess limits/conditions together? what kind of load? How frequently? For how long?

If you use PBO, its all safe. Because PBO wont let you hurt your CPU.

EMI and silicon degradation has no warning signs unless it’s too late. Keep that in mind before you (all in general) set up those limits too high and push the silicon through benchmarks and stress tests for the bragging rights.

Pay attention to how the CPU is operating. If it can use the power, let it have it, if the extra power is a hinderance, and the CPU struggles don't use it.

If you need more MT power for real then there are other ways to get it.

Not by running lower limits..

All I have are my 2 CPU's. My CPU's have not degraded.

 

[Zach_01]

If you have a completely stock CPU, and an Asus board and you enable Asus Performance Enhancement, it will boost your stock EDC to 180, no PBO required., as well as raise your clocks, though not past global limits. So 180 has to be ok for the CPUs that can use it.

I wouldn't run 1.,35 manually 24/7.

If you use PBO, its all safe. Because PBO wont let you hurt your CPU.

Pay attention to how the CPU is operating. If it can use the power, let it have it, if the extra power is a hinderance, and the CPU struggles don't use it.


Not by running lower limits..

All I have are my 2 CPU's. My CPU's have not degraded.

Don't fall into that trap of board vendors. Board limits can work a CPU to what I call a stupid high level so vendors can brag and sell their boards as the best. They are competing in stupidity right there.
PBO the way is implemented by boards (on any auto level) can be potentially unsafe. Especially when users take it for granted. Dont ever trust board vendors that trying to pass their marketing for something you/we need or something that can be done easily. They even set their boards to report wrong telemetry to the CPU to push them further to absolute silicon limit if not more. Who really knows.

I didn't say your CPU(s) are degraded or that they surely be one day. I'm just saying (or trying) that when you play with fire you can get burned. If you or anyone like to risk their hardware for 5~10% more performance its ok. What I wouldn't like to see is that this kind of configuration is the normal, fine, routine and without any risks. Just because ASUS, Gigabyte, MSI, AsRock or any other want to market their products.
Its un-purposely misleading to the unaware user.

You can get more performance by getting a faster CPU too.

 

[OneMoar]

No amperage is Amperage and the only thing that lets you run more amperage is colder temps VASTLY colder temps

I would not run more then 160A through a 5800x unless you could keep it below 75c (which is impossible on anything but subambient or direct die loop) not sustained its different with pbo because you are not sustaining that draw for long

its physics the die can handle X amount of current at a given temperature
thats it thats the only limiting factor

stability doesn't matter
the cpus happiness doesn't matter
what you had for lunch last tuesday doesn't matter

the whole point of this thread is to limit the power consumption because the 5800x being a single die approaches a thermal run-away scenario beyond 140PPT/150EDC this because you are drawing so much power that you can't get the heat out of the die

 

[freeagent]

140PPT is a bit hard on my 5600X, don't think it likes it that high too much. At least not with the cooler it has on right now.

 

[tabascosauz]

way too many watts stock......thank god for PBO2 tuner, seems like this chip runs -30 with no complaints

Close to 100W core power stock is insanity. My 5900X runs 100W cores power across 2 CCDs...

 

[Super Firm Tofu]

way too many watts stock......thank god for PBO2 tuner, seems like this chip runs -30 with no complaints

Close to 100W core power stock is insanity. My 5900X runs 100W cores power across 2 CCDs...

View attachment 263538 View attachment 263537

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.

 

[The King]

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.
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 and efficiency.