AMD PBO Bios Settings

So Im aware that I have limitations with my cooling and motherboard VRM's, so enabling PBO will probably make no difference from where Im at currently. But if a person has a top line board like the Gigabyte Master or Extreme, combined with a excellent custom water cooling solution, in theory if they enable PBO in those two locations, they should be able to go beyond the 4600ghz of the 3900X up to 4800ghz?

Wrong. I mean, your theory sounds correct even by AMD's documentation. However, as I said, these CPU's are limited elsewhere. The motherboard's TDC, EDC and PPT values aren't what hold them back. Using the motherboard values or even inputting your own makes no difference. I've tried this on the MSI MEG X570 GODLIKE, which while not quite as good as a GIGABYTE Aorus Master in the VRM department, is more than capable of pushing enough voltage to these CPU's to handle significantly higher clock speeds and power than we need even overclocked. In other words, it's not VRM temps that are the problem.

If you watch these CPU's in real time through Ryzen Master, what you'll see is that in order to boost to 4.6GHz, the CPU will pull over 1.5v just to hit those clocks on a single core. Keep in mind that generally speaking, you only see one core that's capable of hitting the advertised boost clock. So, what's limiting these CPU's? I think its a matter of thermals. Essentially, these CPU's are at the very edge of what the silicon can do in its current form. To hit the advertised boost clock requires a great deal of voltage. It probably needs even more voltage to go higher. Voltage needs as you raise clock speeds aren't necessarily linear. For example: You get all the way to 4.3GHz on these CPU's at 1.30-1.35v. Yet, a single core reaching 4.6GHz requires as much as 1.50v+. You can observe this behavior for yourself in Ryzen Master. Load up Cinebench R20, and choose the single CPU test. Run it, and watch that one core with the gold star hit 4.6GHz (3900X) or near it and then look at your voltages. They'll spike quite high to achieve those clocks. It's likely that a great deal more voltage is required to push the CPU even a little bit further. At that point, even on water cooling it's likely that the heat becomes an issue.

You can put all the radiator space in your rig you want to. It won't matter. The thing is, you are limited to the single patch of space provided by the IHS and the conductivity of the TIM inside the CPU. Your thermal paste comes into play as well, but even the range of crappy thermal paste to the best stuff on the market is rather small, so there isn't much we can do there. So we are limited by the IHS's ability to dissipate heat rather than our cooling systems. This is why a high end 360 AIO or even an entry level custom loop is probably enough for taking Ryzen 3000 series CPU's to their limits and why guys like me running custom loops can't break this 4.6GHz barrier. LN2 allows you to go further as we've seen clocks over 5.0GHz using LN2.

Essentially, going to higher clocks isn't possible using ambient cooling. It has nothing to do with the motherboard's VRM's and therefore PBO won't help us. The only solution is more voltage, which an ambient cooling solution using air or water simply can't deal with.

No, because even with a decent water cooling setup you'll still be stuck due to the way it scales clock with thermals as you won't be able to bring the temps down low enough. With how little total TDP these chips actually are capable of drawing and the thermal scaling limitations - You're wasting money on some expensive cooling setup if it's for anything more than just the looks of the setup.

There is literally no reason to mess around with overclocking these chips unless you want a specific value forced via manual overclocking, but if you do this you are sacrificing single core speeds...

PB2 is pushing these chips as far as they can safely go, and there is nothing you can do to modify the values that PB2 scales off of.

The only tweaking you can do that will net you real performance gains with a 3000 series build is just with memory.

Well, actually we can alter the values used by PB2. That's what PBO does. It just doesn't help us because those values aren't the limiting factor. On your point about manual overclocking, I agree and said as much in my review of the 3900X. There is some value to manual overclocking, but only in specific circumstances where greater multi-threaded performance is required and that sacrifice of single-threaded performance is considered a worth while trade off. The types of applications that would benefit from this are going to be the types used often by those using content creation applications almost exclusively. Those people would probably be better served by Threadripper with a higher core count. Essentially, the Ryzen 9 series is in a crossover realm between a gaming / desktop PC and a workstation / HEDT system.

Lastly, in addition to the memory settings, you can also overclock the Infinity Fabric clock which can yield some gains.
 
Wrong. I mean, your theory sounds correct even by AMD's documentation. However, as I said, these CPU's are limited elsewhere. The motherboard's TDC, EDC and PPT values aren't what hold them back. Using the motherboard values or even inputting your own makes no difference. I've tried this on the MSI MEG X570 GODLIKE, which while not quite as good as a GIGABYTE Aorus Master in the VRM department, is more than capable of pushing enough voltage to these CPU's to handle significantly higher clock speeds and power than we need even overclocked. In other words, it's not VRM temps that are the problem.

If you watch these CPU's in real time through Ryzen Master, what you'll see is that in order to boost to 4.6GHz, the CPU will pull over 1.5v just to hit those clocks on a single core. Keep in mind that generally speaking, you only see one core that's capable of hitting the advertised boost clock. So, what's limiting these CPU's? I think its a matter of thermals. Essentially, these CPU's are at the very edge of what the silicon can do in its current form. To hit the advertised boost clock requires a great deal of voltage. It probably needs even more voltage to go higher. Voltage needs as you raise clock speeds aren't necessarily linear. For example: You get all the way to 4.3GHz on these CPU's at 1.30-1.35v. Yet, a single core reaching 4.6GHz requires as much as 1.50v+. You can observe this behavior for yourself in Ryzen Master. Load up Cinebench R20, and choose the single CPU test. Run it, and watch that one core with the gold star hit 4.6GHz (3900X) or near it and then look at your voltages. They'll spike quite high to achieve those clocks. It's likely that a great deal more voltage is required to push the CPU even a little bit further. At that point, even on water cooling it's likely that the heat becomes an issue.

You can put all the radiator space in your rig you want to. It won't matter. The thing is, you are limited to the single patch of space provided by the IHS and the conductivity of the TIM inside the CPU. Your thermal paste comes into play as well, but even the range of crappy thermal paste to the best stuff on the market is rather small, so there isn't much we can do there. So we are limited by the IHS's ability to dissipate heat rather than our cooling systems. This is why a high end 360 AIO or even an entry level custom loop is probably enough for taking Ryzen 3000 series CPU's to their limits and why guys like me running custom loops can't break this 4.6GHz barrier. LN2 allows you to go further as we've seen clocks over 5.0GHz using LN2.

Essentially, going to higher clocks isn't possible using ambient cooling. It has nothing to do with the motherboard's VRM's and therefore PBO won't help us. The only solution is more voltage, which an ambient cooling solution using air or water simply can't deal with.



Well, actually we can alter the values used by PB2. That's what PBO does. It just doesn't help us because those values aren't the limiting factor. On your point about manual overclocking, I agree and said as much in my review of the 3900X. There is some value to manual overclocking, but only in specific circumstances where greater multi-threaded performance is required and that sacrifice of single-threaded performance is considered a worth while trade off. The types of applications that would benefit from this are going to be the types used often by those using content creation applications almost exclusively. Those people would probably be better served by Threadripper with a higher core count. Essentially, the Ryzen 9 series is in a crossover realm between a gaming / desktop PC and a workstation / HEDT system.

Lastly, in addition to the memory settings, you can also overclock the Infinity Fabric clock which can yield some gains.

Wow that is some serious quality information, I learned far more in this thread from you guys then all my time on Google or watching videos of annoying people like buildzoid. HARDOCP!!
 
  • Like
Reactions: Dan_D
like this
The problem with most of these 'video makers' is that they ramble for hours about shit, and don't seem to understand effective communication to a wider audience. Although this is largely intentional because they make more money the longer you watch the video and get more ads.

If they made a video that basically said - PBO does nothing because even the stock limits aren't near being hit - It would be a 15 second video and they'd not get any income from ads.
 
EDIT:All good information.

AMD will straighten the BIOS blunders out, but maybe they will not because people complain about high temperatures in idle and high CPU voltage.

You need a good working BIOS for your Motherboard.For your CPU to hit Max boost clocks written on the box.

Ryzen 3
How boost is supposed to work with your new CPU

AMD Ryzen processor will increase clock speeds when the CPU has electrical, thermal, and/or utilization headroom to spare. In other words: if the processor’s analysis of its own environment indicates that it can safely go faster… it will!

For those who will not look at video.
Cinebench 20 single run .
All cores hit Max boost on 3600X (4400Mhz Written on The Box)in the run on Cinebench 20 with up to 3CCX(6 cores all running Max boost at same time.)


This CPU Ryzen 3600X has hit 4525Mhz on a couple cores in light gaming load and when cores are not in use they downclock.

Under Heavy load the CPU clocks/Voltage will decrease as much as the CPU needs .

I believe this is how AMD intended the boost to work for all.




Link to screenshot 4K for those who can not see numbers
https://postimg.cc/VSbXGzQx

Desktop-Screenshot-2019-08-09-01-32-19-47.png
 
Last edited:
turn it off and just run 4.2 take off spectrum too...and declock idle with CoolNQuiet at 2200. GOD with even Airrr how cool it runs.
 
EDIT:All good information.

AMD will straighten the BIOS blunders out, but maybe they will not because people complain about high temperatures in idle and high CPU voltage.

You need a good working BIOS for your Motherboard.For your CPU to hit Max boost clocks written on the box.

Ryzen 3
How boost is supposed to work with your new CPU

AMD Ryzen processor will increase clock speeds when the CPU has electrical, thermal, and/or utilization headroom to spare. In other words: if the processor’s analysis of its own environment indicates that it can safely go faster… it will!

For those who will not look at video.
Cinebench 20 single run .
All cores hit Max boost on 3600X (4400Mhz Written on The Box)in the run on Cinebench 20 with up to 3CCX(6 cores all running Max boost at same time.)


This CPU Ryzen 3600X has hit 4525Mhz on a couple cores in light gaming load and when cores are not in use they downclock.

Under Heavy load the CPU clocks/Voltage will decrease as much as the CPU needs .

I believe this is how AMD intended the boost to work for all.




Link to screenshot 4K for those who can not see numbers
https://postimg.cc/VSbXGzQx

View attachment 179622

The advertised boost clocks are for single core and lightly threaded application use only. When using a heavily multi-threaded workload, the cores will start out higher (around 4.3GHz using a 3900X as an example) and then downlclock quickly as heat builds up. On a 3900X you'll see around 4.1GHz on all cores under such a work load.
 
Times I wished I had a water chiller but don't want to foam everything up to keep the dew down.
 
The advertised boost clocks are for single core and lightly threaded application use only. When using a heavily multi-threaded workload, the cores will start out higher (around 4.3GHz using a 3900X as an example) and then downlclock quickly as heat builds up. On a 3900X you'll see around 4.1GHz on all cores under such a work load.

I will quote myself and I wrote this in the video but I do not talk in video ,just show boring shit.So I do not expect anyone to even look.

Under Heavy load the CPU clocks/Voltage will decrease as much as the CPU needs .

Of course the CPU works this way.It is supposed to.

All cores hit Max boost on 3600X (4400Mhz Written on The Box)in the run on Cinebench 20 with up to 3CCX(6 cores all running Max boost at same time.)


This CPU Ryzen 3600X has hit 4525Mhz on a couple cores in light gaming load and when cores are not in use they downclock.

I believe this is how AMD intended the boost to work for all.

Dan I do not disagree with you on everything.Also in the video I wrote in light loads gaming/idle you can actually see CPU voltage up to 1.57v AMD intended it this way,to give all the juice to some cores to achieve max boost on some/all cores.
If you look at video Cinebench 20,I start cinebench20 then started HWinfo64 to show the CPU voltage in that kind of workload and the voltage will be around 1.5v max .Same behavior you mention in one of the post.

Also CPU of Ryzen 3 xxx are capable of hitting max boost on every core at different times during light load runs like Cinebench 20 ,just like in the video posted.

I can only show how it works and I know your review the stuff ,For example you are trying to get a point across here generally speaking.If you look at the video I posted, then you will see that all cores are capable of hitting max boost and more on all cores,that's about it.The 3900X is a binned part ,so the 3900X is more than capable of the same performance as a not binned 3600X part.
If you watch these CPU's in real time through Ryzen Master, what you'll see is that in order to boost to 4.6GHz, the CPU will pull over 1.5v just to hit those clocks on a single core. Keep in mind that generally speaking, you only see one core that's capable of hitting the advertised boost clock. So, what's limiting these CPU's? I think its a matter of thermals.

Sorry to much text.Take my post as you will .
 
Times I wished I had a water chiller but don't want to foam everything up to keep the dew down.

Would be a waste of money on electricity anyways. Even if you somehow managed to get 4.6ghz-4.8ghz all core on these things it's such a small performance gain over the stock 4.1-4.2ghz it wouldn't even be worth it.
 
Would be a waste of money on electricity anyways. Even if you somehow managed to get 4.6ghz-4.8ghz all core on these things it's such a small performance gain over the stock 4.1-4.2ghz it wouldn't even be worth it.

I agree with this especially in gaming .Tested 4250Mhz thru 4500Mhz and no difference to talk about.Of course did not test every game.
 
I will quote myself and I wrote this in the video but I do not talk in video ,just show boring shit.So I do not expect anyone to even look.



Of course the CPU works this way.It is supposed to.





Dan I do not disagree with you on everything.Also in the video I wrote in light loads gaming/idle you can actually see CPU voltage up to 1.57v AMD intended it this way,to give all the juice to some cores to achieve max boost on some/all cores.
If you look at video Cinebench 20,I start cinebench20 then started HWinfo64 to show the CPU voltage in that kind of workload and the voltage will be around 1.5v max .Same behavior you mention in one of the post.

Also CPU of Ryzen 3 xxx are capable of hitting max boost on every core at different times during light load runs like Cinebench 20 ,just like in the video posted.

I can only show how it works and I know your review the stuff ,For example you are trying to get a point across here generally speaking.If you look at the video I posted, then you will see that all cores are capable of hitting max boost and more on all cores,that's about it.The 3900X is a binned part ,so the 3900X is more than capable of the same performance as a not binned 3600X part.


Sorry to much text.Take my post as you will .

I understand what you are trying to say and how you arrived at that conclusion. However, I do not agree. First off, I don't know why you brought up the voltage as I already said the same thing. Why did I say that? Because AMD said it first. At the bottom, Robert from AMD states that operating ranges upwards of 1.50v are normal for achieving boost clocks.

As for the rest, I have no reason to think that the Ryzen 9 3900X is capable of achieving its maximum boost clocks on all cores and neither do you. A 3600X is a bad analog for a 3900X. It does not boost to 4.6GHz. A 3600X also has only one CCD and not two. Observing a 3600X achieve 4.4GHz on all cores (briefly) does not in any way, shape or form translate to a Ryzen 9 3900X being capable of achieving a 4.6GHz clock speed on all of its individual cores. We do not, and never have observed this behavior. We haven't even seen any signs that the processors are capable of this. Articles done by reviewers specifically address this topic and show that more than likely, not all Ryzen 3000 series cores are created equal. Even Ryzen Master tells us this by differentiating cores with a symbol to denote which cores are the best ones and which are the worst.

The reasons you believe that these CPU's are capable of this are all flawed. Tons of review sites have covered this. Tom's Hardware did a great analysis on this very topic. It's also in line with what I've observed. I've literally seen exactly one core ever clock to 4.6GHz. I've seen others that are capable of clocking to 4.4GHz and I know from manual overclocking that all of them can do 4.3GHz but at 4.4GHz, the system wasn't stable. This tells us that not all of the cores on our 3900X review sample are necessarily capable of 4.4GHz. AMD never states that its CPU's are capable of achieving their maximum advertised boost clocks on all cores. Its reviewers guide was quite clear in that the advertised boost clocks are not an "all core" boost clock. In fact, what the frequency ramping chart shows is that as you increase the workload on the processor in a multi-threaded workload, the clock speeds will drop and level off at a point the CPU can sustain. Using the 3900X example, this point is around 4.1GHz on all cores for that processor.

You can see the frequency ramping graph provided by AMD here: https://thefpsreview.s3.amazonaws.com/wp-content/uploads/2019/07/Clock-scaling.png. This data shows us what to expect from a 3900X CPU using PB2 or even PBO in a multi-threaded workload. AMD flat tells us that you cannot and should not expect your Ryzen 3000 series CPU to achieve or sustain its full boost clocks on all its cores under a heavy workload.

Going back to the 3600X vs. 3900X example, there is a big difference in the thermals and power requirements of a 6 core processor and a 12 core processor. The 3600X also only boosts to 4.4GHz and your own video example shows that your 3600X CPU does not sustain those clocks either. It shows a maximum value which is in line with that advertised boost clock, but I never saw it sustain those clocks in the example. The average column shows that the cores 2-6 average only about 3.0GHz. You also need to understand that there is a huge difference between running one of these CPU's at 4.4GHz and 4.6GHz. As your clock speeds increase, the power requirement isn't a linear one. That means the heat generated by these CPU's as you increase the clocks won't be linear either. Again, you can only dissipate so much heat through the heat spreader on the CPU and its thermal interface material. So no matter how much radiator space you add, or how big an air cooler gets you will eventually hit a hard limit where you can't cool the CPU through ambient cooling anymore. That heat spreader is the limiting factor as is the contact area of our thermal solution.

Also, keep in mind that all CPU's are binned parts. That's how it works. If a CPU fails to perform in a given specifications range or doesn't function correctly, it's rejected. Other CPU / cores / dies are assigned their final configuration based on how they perform. It's based on their clocks, TDP numbers, etc. So yes, a 3900X is a binned part, but so it s a 3600X. It just goes in a different bin. A 3600X only needs to achieve a clock speed of 4.4GHz. With the way power demands and thermals go up with clock speed increases on these chips, there is a huge difference between clocking at 4.6GHz and 4.4GHz. The 3900X also has twice as many cores. Again, people who have these CPU's rarely see them hit their boost clocks on a single core, much less see this happening on different cores throughout the CCD's. Ryzen Master even tells you which of its cores are weakest by how they are marked. I have literally never seen any CPU core with a gray dot or no marking at all hit the advertised boost clock speeds. It's the core with the gold star every time. In the Tom's Hardware test, they even disabled the CCX with the better cores and the CPU was unable to achieve its full boost clocks on any core.

This tells us that not all of the cores in a 3900X are created equal. We know from trying to overclock these chips that these CPU's are incapable of achieving their boost clocks on all cores. Every reviewer tried to do that very thing and failed. The only way you'll see that is possibly with LN2 cooling, which is hardly useful for 24/7 operation. No ambient cooling solution will be capable of dissipating the heat load of these CPUs at such speeds. Lastly, CPU manufacturers like AMD and Intel have a habit of cherry picking CPU's to send to reviewers. If there were ever any 3900X's that could hit 4.6GHz on all cores, they'd probably have sent those to us.
 
Well thanks for the reply and link to toms hardware,which was about the 3600X .
  • Only one core on our Ryzen 5 3600X processor will hit AMD's rated boost frequency. AMD confirmed some cores in Ryzen 3000-series processors are faster than others, which is denoted in Ryzen Master. That means that not all cores on can hit the single-core turbo frequencies. Instead, there are a mix of fast and slow cores.
I will not bother with this anymore.
 
Well thanks for the reply and link to toms hardware,which was about the 3600X .
  • Only one core on our Ryzen 5 3600X processor will hit AMD's rated boost frequency. AMD confirmed some cores in Ryzen 3000-series processors are faster than others, which is denoted in Ryzen Master. That means that not all cores on can hit the single-core turbo frequencies. Instead, there are a mix of fast and slow cores.
I will not bother with this anymore.

Just a note, we have observed the same behavior from the Ryzen 9 3900X. It does not hit 4.6GHz on all cores. As I said, it can't even manage an all core overclock of 4.4GHz.
 
Just a note, we have observed the same behavior from the Ryzen 9 3900X. It does not hit 4.6GHz on all cores. As I said, it can't even manage an all core overclock of 4.4GHz.

My 3900X will run all core at 4.3 on a manual overclock but 4.4 takes 1.5 volts to keep stable and that is to much for me. If I just let it clock on it's own then what you see your 3900X do is what mine will do as well all cores at around 4.1 and boosts just shy of 4.6 on single core. For the most part I just let it run at whatever speed it wants to run at, have seen little benefit to manually overclocking it short of synthetic benchmarks.
 
  • Like
Reactions: Dan_D
like this
My 3900X will run all core at 4.3 on a manual overclock but 4.4 takes 1.5 volts to keep stable and that is to much for me. If I just let it clock on it's own then what you see your 3900X do is what mine will do as well all cores at around 4.1 and boosts just shy of 4.6 on single core. For the most part I just let it run at whatever speed it wants to run at, have seen little benefit to manually overclocking it short of synthetic benchmarks.
I agree. You gain more performance and better stability by just letting it do its thing and tweaking your timings and ram speed. This way you get the better Single core boost when it can and also benifit from proper sleep states, correct voltage idle etc..
 
Just a note, we have observed the same behavior from the Ryzen 9 3900X. It does not hit 4.6GHz on all cores. As I said, it can't even manage an all core overclock of 4.4GHz.

Let me start by saying you correct and I apologize for any misinformation I supplied due to my ignorance. I was aware of Third-Party Monitoring Tools not being as accurate as Ryzen Master and I just ignored that fact all together.
I have started testing with Ryzen Master only now to see what the actual clocks on my Ryzen 3600X really is.Although I still get over Max Boost as what is written on the box so I have no recourse to complain to AMD or refund my CPU'S.Also they are still fast little CPU's Still happy.
I have only witness Cinebench20 hitting Max Boost on 2 cores.
https://i.postimg.cc/HLnQJjLW/3600-X-Ryzen-boost.png
 
precision boost runs all my 3900x cores at 4.2-4.3Ghz with no manual overclocking or playing with p-states. Just letting it do it's thing. Just stays there churning away as long as it's needed.

will hit the 4.5-4.6Ghz pbo on 1-2 cores if only doing single thread loads.

Manual Overclocking really looks like it's time is up. Diminishing returns and improved auto-oc'ing has killed any desire to pursue it for me.

In terms of bragging rights these days, it seems more of a worthwhile brag to undervolt and retain stock performance/stability than to eek out a 2.5% increase in frequency across 8-12 cores and either not be able run 100% cpu for any length of time or have to run in a chilled environment.


i guess the big question is if there's going to be an easy way to expose which cores are the "golden" cores to operating system schedulers so they can schedule high cpu single thread tasks to them directly rather than rely on random chance to schedule them on those cores. Since it's not like the cpu is going to auto-shuffle it for you. Unless AMD always assigns cpu0 and/or cpu1 to the golden cores. I failed to pay attention when i was booted into windows to see what ones ryzen master said mine were.
 
Back
Top