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.

 

[TerryLaze]

Would need pics from all of your bios pages, my assumption would be that your mobo has some sort of overclock/all core enhancement enabled.

 

[LeVzi]

Ah right, that would be Core Performance Boost. I'd forgotten about that.

 

[CountMike]

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.

PBO (Precision Boots OVERDRIVE) is not doing that but Precision Boost or in old parlance good ole Turbo function.

View: https://www.youtube.com/watch?v=B7NzNi1xX_4

 

[Karadjgne]

4.8GHz is max default turbo. PBO can (if set) bump that upto 200MHz higher if cooling and voltages allow. There's usually 2 places (don't ask me why, it's AMD stupidity) where PBO can be enabled, in the tweeker settings And in the AMD overclock section. Which can get confusing. In my Asus bios, enabling/disabling one doesn't change the other.

 

[LeVzi]

Yeah PBO is in the AMD Overclocking area and the misc, XFR section ... lol

I have just left them at AUTO which I understand is disabled.

I have a problem with the CPU not down volting at all. Always about 1.3V going to 1.4 too often.

I have discovered that NZXT CAM software is NOT compatible with Windows 11 at all. Which I Find laughable tbh .

 

[TerryLaze]

I have just left them at AUTO which I understand is disabled.

Auto is whatever the mobo maker decided, it doesn't always mean off.

 

[tennis2]

I have a problem with the CPU not down volting at all. Always about 1.3V going to 1.4 too often.

This is starting to sound like Ryzen Master type stuff. Have you used Ryzen Master to tweak anything?

What mobo do you have?

For reference, we can only assume that you're working with "out-of-the-box" settings on everything from mobo settings to windows settings, and no 3rd party software. If this isn't the case, you need to tell us so we're not wasting time getting the full story.

 

[LeVzi]

I hadn't made any changes to the stock settings at all, was running the F36 BIOS from Gigabyte for the X570 Elite Rev 1 board. Using a Kraken X63 cooler .

I haven't used Ryzen Master to do anything other than watch temperatures and core speeds . I use HWInfo for temps as well.

It seems to be boosting non stop even at idle. Under full load temps are fine at 71 degrees. Idle after gaming or a video render is over 60 constantly. Cool boot idles are 50 degrees.

 

[Karadjgne]

I believe that board has 3 mode settings in bios. Eco, standard and performance. To follow Amd Ryzen guidelines for power limits, timing, voltages etc, you should be on Standard. Performance mode raises that, basically Gigabytes version of PBO. All the major vendors have such, it's how they can claim 'our board outperforms the competition' and get away with it.

 

[LeVzi]

I set PBO Limits to Motherboard, and Negative all cores 30 , its stable from what I can see.

Thermals are WAY better. and no dip in performance.

Instead of idling at 60 after gaming, it drops back to 50.

 

[tennis2]

 

[Karadjgne]

Have to be careful with idle temps, they can be misleading with Ryzens.

With Intel, at idle the cpu drops speeds and voltages, but all cores are active. So any background tasks, services, processes etc are split up amongst all the cores, getting a very minor spike with each startup. So you'll generally see idle at @ 10°C ish above room ambient as idle is read as hottest core, which individually isn't doing all that much. Most might be 31° and one be 33° for a second, drops back to 31° and another process starts and a different core jumps to 33°. You'd just see the constant switching of 33° cores as hottest.

Ryzen run 1 core. The rest are truly idle, asleep. So the entire workload of all those processes and services is dumped on a single core. It's going to be the hottest by far as a result. Every time a new process starts, that just gets added, and others drop off, so you'll see small-large spikes depending on the load. But that results in idle temps that can easily range from 20°ish to over 40° ish above ambient temp. It's not uncommon to have a 40°C idle that jumps sporadically to 60° ish, and depending on which software is used, if any, that can also show funky behavior, depending on read times. Ryzen Master averages the last 3 seconds worth of loads, so if you had 40,40,70, you'd see a idle of 50°. Some software reads only 1x a second, if it happens to read the 70° temp, you get a reported temp of 70°C for idle, and ppl freak out. It's not truly 70°, nowhere near, but the software read a spike, not a nominal temp. If it happens to read the 40, it'd make you think you had awesome idle temps, but in reality, most of the cores are much closer to 30ish° and just the one is hitting the 40+.

And what's crazy is as soon as you open a small load, like moving the mouse or a web page, the rest of the cores wake up, and temps can drop from the higher idle as the load gets bounced around, background tasks get put on hold etc.

So take idle temps with a grain of salt, figure what's normal for your particular cooling, cpu, airflow and only get worried when they deviate drastically.

 

[tennis2]

Temps also need to be associated with cooler fan speeds. My GPU "idles" at 50C but the fans are also completely off.

 

[CountMike]

For general use, Ryzen Master temps are most usable, you don't get details but it displays "Effective temps" as if it was a single core CPU.

 

[LeVzi]

Thanks everyone, I just needed to know if the higher idle temps were actually damaging the chip, but everyone here and finally AMD said no, its all normal Ryzen behaviour. Its the under load temp that really matters, and with PBO i have a nice stable 4.5 across all cores at a cool 72 max load.

I could push it harder, but there is no real need.

Idle temps are lower too.

I just wish AMD would fix the chipset installer because getting chipset drivers updated is a nightmare.

 

[LeVzi]

One thing I noticed , with PBO set to Advanced, negative curve optimiser of all cores 15 , when you test with Prime95 small FFT's for thermal testing, the temp is 85 , which is too close to TJmax for me.

I dropped to 8 and it was 82 which is better but hardly worth it.

But when I run Prime95 small FFTS with PBO set to auto, the temps are 60 and there are not any thermal issues., in fact the idle temps are 58 full load under Prime95 and it doesnt raise.

Is that to be expected ?

 

[Unolocogringo]

Different tests in prime 95 produce different loads and different temps.If you want to check max thermal load run the torture test.
It will run different tests. The first 5 or so are lighter, then the heavy load kicks in.
Make sure you watch temps, if your cooling is not good enough it will over heat and throttle.

 

[LeVzi]

Different tests in prime 95 produce different loads and different temps.If you want to check max thermal load run the torture test.
It will run different tests. The first 5 or so are lighter, then the heavy load kicks in.
Make sure you watch temps, if your cooling is not good enough it will over heat and throttle.

I thought that the test with small FFT's was the best for thermal testing and stability ?

I run that with stock 5900X settings and the thermals are fine. I run it with even a small -5 offset with PBO an d the thermals jump nearly 20 degrees.

It's running it at stock that confuses me, as it really doesn't effect thermals at all. Not to what i'd expect.

 

[Unolocogringo]

Small FFTs do load the processor part of the CPU hardest.
But that is only part of an AMD CPU.
The other part , The I/O die has the memory controller,PCI-E ,SATA,USB etc.....
The blend torture test stresses all of the parts of a CPU and memory subsystem.
A better test for overall system stability.

 

[drea.drechsler]

I set PBO Limits to Motherboard, and Negative all cores 30 , its stable from what I can see.
...

That's really quite surprising as there are usually a couple of cores at least that require higher voltage (less negative offset) during boosts. They are the gold star or best cores. Some people have even reported using a positive curve offset for those cores actually improved boost performance for their CPU's.

How are you testing for stability? It usually won't crash in heavy threaded, all-core, stress testing as the cores won't be trying to boost to max clocks then. It shows up in light threaded, bursty type processing. I can get mine to crash in a Defender quick scan. One of the best way is to use a utility called Core Cycler. It uses a single threaded Prime95 huge FFT to test one core at a time. It also tells you precisely which core crashed (or returned the erroneous result) so you can reduce the offset a little on that core and try again.

 

[LeVzi]

That's really quite surprising as there are usually a couple of cores at least that require higher voltage (less negative offset) during boosts. They are the gold star or best cores. Some people have even reported using a positive curve offset for those cores actually improved boost performance for their CPU's.

How are you testing for stability? It usually won't crash in heavy threaded, all-core, stress testing as the cores won't be trying to boost to max clocks then. It shows up in light threaded, bursty type processing. I can get mine to crash in a Defender quick scan. One of the best way is to use a utility called Core Cycler. It uses a single threaded Prime95 huge FFT to test one core at a time. It also tells you precisely which core crashed (or returned the erroneous result) so you can reduce the offset a little on that core and try again.

I thermal test with Prime95 and Cinebench, but the stability, as you rightly said, when doing lighter tests wasn't there. Just updating a piece of software caused a system hang.

For now until I can sort out this cooler, i've gone back to stock.

Incidentally Ryzen master says Cores 3 and 4 are the most popular ones. Once I get the cooler sorted i'll consider trying to OC again.

 

[Unolocogringo]

Evey chip and motherboard is different.
My 5600x and Prime x570 runs with
PBO set to manual.
PPT 142
TDC95
EDC 140
Curve optomizer set to +28.
LLC set high.
ALL other core boost etc... turned off. They use too much voltage.
4.65 ghz for 3 core or 4 core depending on load. 1.21- 1.26v
4.5ghz all core boost 1.28-1.32v
6 core 4.65 used 1.36-1.41 all core boost and way too hot
every piece of silicone is different, and every core is different.
All core boost and voltage depends on your weakest core and the voltage it needs to reach desired boost speed.
My 3600 on a prime x570 does 4.4 all core boost @1.28v
All PBO etc turned off. pure overclock.

 

[LeVzi]

Mine is hitting 4.9 with 1.4V for the single core boost (or 2 core boost)

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

 

[drea.drechsler]

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

Inconsistent clock speed is pretty much to be expected since the algorithm adjusts individual core clocks up to 100 times a second based on load and temperature and whatever else it uses. Clocks will also constantly dither up and down, and lower on average, in a heavy all-core workload like Cinebench. An unrealistically heavy workload, such as Prime95 small FFT's, will make the algorithm pull clocks even further to keep core temps under control. The major advantage from using PBO and Curve Optimizer with Ryzen is the boost algorithm is still working and protecting the CPU as it gets hot under load so that's what you're seeing.

All that dithering does make clock-watching pretty much worthless as a way to infer performance though: you have to run a timed benchmark all the way through to effectively gage performance changes. The other effect of it is improved cooling in and of itself is like overclocking since the algorithm won't pull back as far if the cores aren't heating up as much.

When using Curve Optimizer I've found I get the best over-all performance by actually reducing the EDC below stock: for a 5800X EDC stock is 142A, I'm running at 120A. The effect is to make it a bit less eager to show a lot of cores hitting maximum clocks in light threaded boosting but Cinebench BM scores are higher. I think what's happening is it reduces core heating a bit early on by trading some early high clocks for a higher average sustained clock through the entire run. But that's just a guess.