Question PBO going over maximum stated clock speed ?

[LeVzi]

I was under the assumption that Precision Boost Overdrive would never take the core speed over the chips stated max.

The 5900X has a max boost speed of 4.8

Yet HWinfo shows max core clock of 4.950

I have disabled PBO in the BIOS anyway, so how am I getting this speed ? Seems to only happen at startup.

I noticed this because idle temps are extremely high at idle, but under load less than idle, so I am completely confused.

 

[tennis2]

All core is a lower voltage, the clock speed isnt consistent Prime95 gives lower than say cinebench

That's to be expected since you're running against the power draw limits. A test that produces a higher power draw will result in a lower frequency.

 

[LeVzi]

I think tbh I am going to leave it at stock, its more than enough with power to manage everything I throw at it.

I am having thermal issues using PBO tbh, Hitting 80 using Prime95 small FFT's with a -15 All core PBO is still 10 under the Tjmax but for me, that's not doing the silicon any good.

I have idles of 60 and boost of 70 , I can live with that. Max thermals on full load (Handbrake) 74 over 2 hours rendering. It's all good tbh.

I am disappointed with the Kraken X63 tbh, I wish I'd gone with a 3 fan cooler now, or at least a much more powerful 2 fan AIO. Plus NZXT are really slow in updating their controller software, and the fan curve in the BIOS doesn't seem to do anything...

 

[drea.drechsler]

...
Hitting 80 using Prime95 small FFT's with a -15 All core PBO is still 10 under the Tjmax but for me, that's not doing the silicon any good.
...

Just turning a computer on isn't doing the silicon inside any good, it's intrinsic to the nature of it. But it will still give you a long life, certainly way longer than you'll find it useful.

FYI what AMD has to say of Ryzen 5000 temperatures:

Max Safe Temperature for AMD Ryzen 5000 CPUs? | Hardware Times

 

[LeVzi]

Interesting, so even using PBO and running nearer the TJmax , it wont really make much difference to the life expectancy of the chip.

Thats interesting. And makes for easier OC's with less issues.

 

[tennis2]

The stock cooler and a Kraken X63 (280mm AIO) should* have wildly different thermal capabilities. This thread is getting quite muddled on what PBO settings you're using when. Simply put, I'd recommend monitoring all of the following together:

• Normal CPU core frequency
• Effective CPU core frequencies (ensure clock stretching is avoided, needs a "power virus" test like Prime95 to be accurate)
• Core voltage
• Power Draw
• CPU fan speed
• CPU temp

In doing so, you can at least determine what setting is providing the lowest stable* core voltage for a given frequency. Then you can adjust for the capabilities of whatever cooler you slap on top.

Everyone has their own 'comfort zone' for temps, power draw, and noise. I personally stick near the "stock voltage" or lower (via undervolting) for CPUs in the first few years. Keeping or slightly exceeding the stock frequency (as undervolting allows). Of course, that hinges on what you define as "stock" voltage. Is that at the single-core boost frequency? The all-core boost frequency? The "base clock" frequency? Clear as mud.
With my 5600G, I set a -20/-30 (depending on the core) curve optimizer voltage offset and a +200MHz PBO limit, no increase to the max power draw. Not chasing all-core = single-core or anything.

Keep an eye on the % or FPS performance uplift you're getting from overclocking. A 20% overclock doesn't always translate into a 20% performance increase. And that % increase will be different for different games. Also, do you NEED x% more performance in the first place? And how many FPS is that? That will help you decide what's worth the 'risk'.

 

[drea.drechsler]

Interesting, so even using PBO and running nearer the TJmax , it wont really make much difference to the life expectancy of the chip.
...

Just don't take that line of thinking too far....

Temperature really is a silicon killer...so is core current; high temperature at the same time as high core currents are the worst and both are directly affected by voltage. The good thing about using PBO is it leaves the algorithm in place so it's not overclocked with fixed clocks and voltages. That means when not working hard it's going to lower voltage to .6V or less as often as it can, even turning core clocks completely off in C6 deep sleep states. When it is working hard it will lower core clocks from the very high max boost clocks so it can also run at a lower voltage with stability. That all tends to make for a lot of clock dithering as it transitions from one process to another.

AMD is comfortable running up to 90C only because they know the algorithm is going to save it. Also because they have the reliability data needed and have run out the math for lifetime projections to their satisfaction.

 

[LeVzi]

The stock cooler and a Kraken X63 (280mm AIO) should* have wildly different thermal capabilities. This thread is getting quite muddled on what PBO settings you're using when. Simply put, I'd recommend monitoring all of the following together:

• Normal CPU core frequency
• Effective CPU core frequencies (ensure clock stretching is avoided, needs a "power virus" test like Prime95 to be accurate)
• Core voltage
• Power Draw
• CPU fan speed
• CPU temp

In doing so, you can at least determine what setting is providing the lowest stable* core voltage for a given frequency. Then you can adjust for the capabilities of whatever cooler you slap on top.

Everyone has their own 'comfort zone' for temps, power draw, and noise. I personally stick near the "stock voltage" or lower (via undervolting) for CPUs in the first few years. Keeping or slightly exceeding the stock frequency (as undervolting allows). Of course, that hinges on what you define as "stock" voltage. Is that at the single-core boost frequency? The all-core boost frequency? The "base clock" frequency? Clear as mud.
With my 5600G, I set a -20/-30 (depending on the core) curve optimizer voltage offset and a +200MHz PBO limit, no increase to the max power draw. Not chasing all-core = single-core or anything.

Keep an eye on the % or FPS performance uplift you're getting from overclocking. A 20% overclock doesn't always translate into a 20% performance increase. And that % increase will be different for different games. Also, do you NEED x% more performance in the first place? And how many FPS is that? That will help you decide what's worth the 'risk'.

You dont set PBO as motherboard limits ?

 

[LeVzi]

Just don't take that line of thinking too far....

Temperature really is a silicon killer...so is core current; high temperature at the same time as high core currents are the worst and both are directly affected by voltage. The good thing about using PBO is it leaves the algorithm in place so it's not overclocked with fixed clocks and voltages. That means when not working hard it's going to lower voltage to .6V or less as often as it can, even turning core clocks completely off in C6 deep sleep states. When it is working hard it will lower core clocks from the very high max boost clocks so it can also run at a lower voltage with stability. That all tends to make for a lot of clock dithering as it transitions from one process to another.

AMD is comfortable running up to 90C only because they know the algorithm is going to save it. Also because they have the reliability data needed and have run out the math for lifetime projections to their satisfaction.

With that said I get higher thermals with a negative offset , ANY negative offset. So the chip is hotter at lower voltages ?

 

[drea.drechsler]

With that said I get higher thermals with a negative offset , ANY negative offset. So the chip is hotter at lower voltages ?

I saw that same effect when I started using curve optimizer with my 5800X. It starts to make sense when you look at what CO is doing: it's lowering core voltages mostly, if not only, at the high frequency end of the V-F performance curve, each core individually if doing it that way. CO (all alone) does not change the algorithm's performance parameters (PPT, TDC, EDC and Scalar especially).

Based on Robert Halleck's statements I think we can assume the boosting algorithm is very much temperature seeking. Lowering core voltage at the high end with CO WILL help lower heat generation during boosts but the boost algorithm is temp seeking so it boosts so long as it sees sufficient temperature (and power) margins, i.e., it will still try to boost to the 90C limit.

Low to mid frequencies are going to be dominated by heavier processing and will heat up the cores much as before, especially since they are also less affected by CO.

You'll have to alter some other PBO parameters to affect temperature: some people lower PPT from stock with very little effect on performance with their tweaked CO settings. I lower EDC from 142A to 120A for my 5800X: with tweaked individual CO curves it makes for higher CB20 and 23 scores although it boosts to max clocks less eagerly. I've also set a platform thermal limit of 85C.

All that means getting really good cooling is like overclocking for Ryzen. The improved heat removal of a great cooler means it can keep boosting longer/higher until reaching the max temp ranges (80-90C) where it starts pulling clocks. But it doesn't mean the CPU will necessarily run cooler under extreme work loads with one because Ryzen CPU's are significantly affected by the tiny 7nm geometry that results in such a small surface area to transfer heat across.

 

[tennis2]

With that said I get higher thermals with a negative offset , ANY negative offset. So the chip is hotter at lower voltages ?

As I said above, you can't just measure one variable. Did the negative offset change the frequency? (the frequency should have increased). Also, HWinfo64/RyzenMaster (or your choice of monitoring software) will tell you what the core voltage is. They'll also usually tell you what the performance cap/limitation is so you don't have to guess.

Also, I'm getting the feeling like you haven't yet watched the videos linked in posts 4 & 12 in this thread yet....?

You dont set PBO as motherboard limits ?

I've only chosen to use Curve Optimizer and increase the maximum boost of PBO to +200MHz. I haven't felt the need to dabble in increasing power/current/voltage limits yet. Life is busy, and the more settings I change, the more stability testing I have to do. Which means the less time I get to spend using my free time for gaming/etc. Some day, sure, I'll see how much I can squeeze out for SCIENCE!! But again, if I felt like I needed that extra few % CPU performance, I would've bought a better CPU. Keeping in mind that not all games are CPU-bound anyway. So the a double-digit % CPU frequency difference can scale to zero % FPS benefit depending on the game.

What I don't like about mobo limit PBO (on my AsRock BIOS) is that it doesn't expose the set values of PPT/EDC/TDC/Scalar/etc like "manual" setting does. It just says "Motherboard" and all those settings are hidden (for me to discover via testing apparently?) If my knowledge of the industry is any indication, the mobo limit PBO settings are probably effectively infinite, leaving PBO to boost on thermals primarily (with a cap of Fmax+offset obviously), which I don't prefer.

I'm very power consumption driven since that has a direct impact on my PC noise output and the temperature of my room (with the door closed). Inasmuch, my CPU and GPU settings are conservatively set to minimize power/heat/temp with minimal/no effect to performance. For this reason, my use of curve optimizer +200MHz PBO boost limit increases my CPU frequency without exceeding the stock ~65W power draw. Likewise, I typically run my RTX3060Ti at ~75% power limit whilst undervolted to achieve at/near the same ~2100MHz frequency as stock. I also routinely set frame rate caps to minimize power consumption in lightweight games. When I had my previous AMD GPU, I absolutely LOVED the AMD Chill and FRTC power saving features!

Keep in mind that the slope of the frequency-vs-voltage curve increases the further you go. You may be able to eke out an extra 10% frequency on a component, but that may come at a >>>10% increase in power consumption, and perhaps <<<10% actual PERFORMANCE (FPS) improvement. If all you care about is max performance with acceptable lifespan risk at the cost to all other factors, then life is simple. OC the snot out of everything.

 

[Unolocogringo]

I alway just overclock the snot out of everything.
But I cool it very well and it must be stable at the same time.
Folding does not like an unstable computer It takes weeks to find each components stable max with reasonable voltage. You gotta keep the heat down. sometimes less voltage is better as it lets you boost a little higher.
Folding loads one core per video card.
With "stock" settings I was getting 1.38-1.41 with peaks to 1.45 for 4.65ghz. Way too high for 24/7/365 use.
I now get 1.22-1.24 for 2 core 4.65 ghz with slight spikes when it bounces to less capable cores.
Less voltage at same clock speeds and able to do 4.5 all core boost instead of 3.8-4.1 boost with default voltages.
My burn in test is a week of prime 95 with windowed furmark on a loop and looping 3dmark.
If it can handle that it is ready for folding.
Win/Win

 

[tennis2]

Folding loads one core per video card.

Does OCing the CPU improve GPU folding? I always dialed my CPU wayyy back (like 2GHz) when GPU folding (CPU folding is pointless waste of power IMO).

 

[Unolocogringo]

It depends on processor and video card.
My 960/1060 folding box gets almost 100,000 point more per day since I upgraded from a Phenom 2X6@3.6 to a 3600@4.4

 

[tennis2]

It depends on processor and video card.
My 960/1060 folding box gets almost 100,000 point more per day since I upgraded from a Phenom 2X6@3.6 to a 3600@4.4

ah, ok. Details are important!
So you're folding with your CPU IN ADDITION TO your GPUs. The Ryzen 3600 is ~111k PPD whereas the Phenom 2x6 is ~11kPPD, that's the +100kPPD you're seeing. Your GPUs contribute ~600kPPD combined. So adding in your CPU raised your PPD by 18%. OCing your Ryzen 3600 likely added less than 2% extra. My earlier point stands, CPU folding is a waste of power/heat and risk of component failure due to prolonged stress.
I don't think this side convo has much validity in relation to the OP topic.

PS: Hey! my 3060Ti could do 2.8M PPD!!

 

[Unolocogringo]

I do not CPU fold.
Only GPU.
box1
GTX 960FTW and GTX 1060 6GB gamer about 1-1.2 million per day
on a primex570 with 3600
box 2
GTX 1070 SC about 1- 1.4 million per day
Prime z170 with I5 6600
Box 3
RTX 3060 Ti varies wildly from 2.5-4.0 million points per day depending on work unit.
5600x on prime x570
All fold 24/7/365
All overclocked to their max stable
Just retired the Phenom2 x6 on a m5a97 about 2 months ago.
It folded GPU/CPU about half its life and GPU the rest. Had dual GTX460s on it at one time.
Overclocking when done safely does not decrease longevity of parts much.

 

[Karadjgne]

Voltage is a multiplier. Of Amperage. It's a balancing act. If you lower voltages, and the cpu decides it has the room to boost more, or more cores, that takes Power. So amperage goes up, and so does the heat.

Lowering voltages doesn't always result in lower temps, only when those voltages are sufficient to max out the power requirements.

PBO is responsible there, the cpu is saying it can do more, and applying a negative offset you drop VID voltage, limiting the voltage the cpu can draw upon. So it bumps the amperage instead, since those limits are no longer existing.

Dumping pbo, allowing the Ryzen to govern its own limits and then undervolting, you'll drop temps.

With Cam, I just set mine on Silent mode and let her rip. It kept a 3770k at 4.9GHz OC to 72°C with p95 small fft.

 

[LeVzi]

OK I have fixed and correct all my thermal and OC issues, well I am not even OC'ing and finally got my thermals in line (Which is why I started fiddiling with undervolting anyway).

I have a max clock boost of 4.5 on normal threads. Idle temps are NOW 32 degrees down from 60.

I can introduce PBO should I want to but tbh, I have a stable and quiet system after weeks of checking.

 

[drea.drechsler]

...
finally got my thermals in line
...
I have a max clock boost of 4.5 on normal threads. Idle temps are NOW 32 degrees down from 60.
...

You should be seeing max boost clocks of 4.8Ghz for a 5900X. It's not unreasonable to see 4.85 or even 4.9Ghz boosts with stock settings on some motherboards and with good cooling. With temps that low it's possible something's off in your setup. I'd try to get stock performance ironed out before fiddling with PBO.

And thermals in line with what? Temps of 60C are perfectly good, possibly great even for that processor - or most any modern high performance desktop processor for that matter - when being used. 32C is very low and should only be seen when the system is sitting perfectly quiet with cool ambient temperatures.

 

[LeVzi]

You should be seeing max boost clocks of 4.8Ghz for a 5900X. It's not unreasonable to see 4.85 or even 4.9Ghz boosts with stock settings on some motherboards and with good cooling. With temps that low it's possible something's off in your setup. I'd try to get stock performance ironed out before fiddling with PBO.

And thermals in line with what? Temps of 60C are perfectly good, possibly great even for that processor - or most any modern high performance desktop processor for that matter - when being used. 32C is very low and should only be seen when the system is sitting perfectly quiet with cool ambient temperatures.

I was idling at 60 , but now 36 cold boot and 40 after use. Max temps are 75 under full load.

Was rogue programs causing the CPU to boost all the time.

I get good boost clocks tbh, I dont need to use PBO yet if I am honest.

 

[drea.drechsler]

...
Was rogue programs causing the CPU to boost all the time.
...

They aren't "rogue" programs. They are Windows's own background processes. Things like the search indexer and update service. Even your mouse is a part of that crowd: processing the pointer location and reacting as it "rolls" over hot-spots by changing it (from arrow to finger, for instance) doesn't come free.

Open Task Manager and check the CPU section of the Performance tab. There will be several thousand registered threads supporting a couple hundred active processes. Most are quiescent but every few seconds one will wake up and do something...with an accompanying boost of a CPU core to get it done FAST.

 

[LeVzi]

They aren't "rogue" programs. They are Windows's own background processes. Things like the search indexer and update service. Even your mouse is a part of that crowd: processing the pointer location and reacting as it "rolls" over hot-spots by changing it (from arrow to finger, for instance) doesn't come free.

Open Task Manager and check the CPU section of the Performance tab. There will be several thousand registered threads supporting a couple hundred active processes. Most are quiescent but every few seconds one will wake up and do something...with an accompanying boost of a CPU core to get it done FAST.

Let me educate you, there was a rogue program, calling on CPU cycles non stop under Windows 11 , when it didn't on Windows 10.

It was removed, and the problems went away.

Also, disabling driver verifier dropped it further.

So from 10-12% CPU IDLE , (Which isn't idle) down to less than 0% true idle.

 

[Karadjgne]

Also, disabling driver verifier dropped it further.

That'll do it. Those programs are not much more than a gimmick. Windows drivers run most of everything except specific services like nvidia/amd 3d stuff, and nvidia even got together with Microsoft and includes the basic drivers now. Which leaves stuff like specific printers or keyboards or other add on components which have specific driver needs. Which most ppl can update themselves if there's an issue.

The verifyer wouldn't be bad if it ran once at boot, then shut off, but constant checks are absolutely not necessary, drivers don't suddenly go out of date in a second.