Trying to understand Ryzen 3000 series boost speed variations

[lorry]

messing with PBO and its limits is useless (outside fixing broken bios), as multiple sites have tested even with the top boards from msi/asus/gigabyte,
all more or less limited by voltage/heat transfer (away from chip).
major reason why it should be on auto (and bios sees it as OFF but still uses PB).

@lorry
i had looked into bios settings for my board and came across posts saying some of the latest updates
wouldnt wipe settings with bios update, even when removing bios battery/clearing cmos.

backup your bios settings (pics with ur phone?!),
put the latest bios on a usb stick (no matter if u r already using it), reboot and load defaults then shutdown.

use Qflash to flash the bios again, reboot and load the default/optimized bios settings, save them, reboot again.
now change settings to your liking...

for what reason though?
As that is pretty much what has already been done
latest bios was flashed by myself (F50), optimized settings used and then edited to the current settings

 

[Waldorf]

because my bios related probs went away after i did it that way.

 

[lorry]

Totally unrelated to any of this, just something I saw in a video

Windows Sysinternals

• 12/11/2019
• 5 minutes to read

The Sysinternals web site was created in 1996 by Mark Russinovich to host his advanced system utilities and technical information. Whether you’re an IT Pro or a developer, you’ll find Sysinternals utilities to help you manage, troubleshoot and diagnose your Windows systems and applications.

• Read the official guide to the Sysinternals tools, Troubleshooting with the Windows Sysinternals Tools
• Read the Sysinternals Blog for a detailed change feed of tool updates
• Watch Mark’s top-rated Case-of-the-Unexplained troubleshooting presentations and other webcasts
• Read Mark’s Blog which highlight use of the tools to solve real problems
• Check out the Sysinternals Learning Resources page
• Post your questions in the Sysinternals Forum

Sysinternals Utilities - Sysinternals

Evaluate and find out how to install, deploy, and maintain Windows with Sysinternals utilities.

docs.microsoft.com

 

[Zach_01]

messing with PBO and its limits is useless (outside fixing broken bios), as multiple sites have tested even with the top boards from msi/asus/gigabyte,
all more or less limited by voltage/heat transfer (away from chip).
major reason why it should be on auto (and bios sees it as OFF but still uses PB).

I guess you didn’t experience it or you didn’t notice it or you board/CPU responded in different way than others.... and please let me explain

When I was on AGESA 1003ABBA I had PBO in advanced mode +200MHz and the CPU was boosting to the rated 4200MHz for the high perf# cores. When AGESA 1004B initially released, my CPU has lost about 50-100MHz single core boost clock. Note that I was with win v1903.
Setting the PBO to Auto from advanced has got me back to proper boost of 4200MHz.
After v1909 and later versions of 1004B UEFIs I’m back to PBO advanced +200MHz and 4200MHz single core boosting.

That I was talking about and nothing else...

 

[Waldorf]

Might work in your case.
I meant there is proof it doesn't always help performance on 3xxx cpus compared to pbo auto/off, and most of the time overall perf is actually slightly less.

 

[tabascosauz]

It looks to me like AMD use a lot of proprietary means of recording that they aren't overly interested in making public, just how accurate they are is also difficult to determine.
As an aside I watched a video last night from BZ on ram temps and case temps, what affects them (GPU temps and radiant heat, CPU dies and infinity fabric links with one end being red hot). It was interesting but most of it only really seemed to apply to those that build really bad setups with next to no airflow, cooling etc

Yeah, I saw that video. Pretty interesting. Memtest86 is ice cold because the CPU is under no load and it runs outside of Windows, but P95 Large stresses the mem controller hard. In about 15 minutes, I was up to 49° on the DIMMs. Didn't get any errors on my DJR, but it seems B-die can be much more sensitive to temperature.

Mine might be small, but I'm still running a U9S and A12x25 intakes with one almost directly over the DIMMs. I have a feeling that one would start running into problems with shaky timings/freq on the edge of stability, coupled with a split SFF case with zero airflow like the A4 or Ghost S1 (one of the reasons why I'm not fond of those), a low profile cooler like the L9x65 or Blackridge, and a 290X dumping heat everywhere.

Might work in your case.
I meant there is proof it doesn't always help performance on 3xxx cpus compared to pbo auto/off, and most of the time overall perf is actually slightly less.

...which is why I suggested lorry poke around in that PBO submenu to see if anything will change. Everyone's experience varies. Mine very clearly clocks higher, but it lowers my CB R20 score slightly, presumably because in order to realize those PBO gains I have to return my Vcore to stock, which is quite hot and seems detrimental under heavy sustained load on my U9S.

This time around, reviews literally don't mean anything beyond VRM temps. There are so many factors influencing the performance you get; even having the same board is no guarantee, as some identical RAM kits use different DRAM and the wild silicon lottery takes care of the rest.

 

[lorry]

Yeah, I saw that video. Pretty interesting. Memtest86 is ice cold because the CPU is under no load and it runs outside of Windows, but P95 Large stresses the mem controller hard. In about 15 minutes, I was up to 49° on the DIMMs. Didn't get any errors on my DJR, but it seems B-die can be much more sensitive to temperature.

Mine might be small, but I'm still running a U9S and A12x25 intakes with one almost directly over the DIMMs. I have a feeling that one would start running into problems with shaky timings/freq on the edge of stability, coupled with a split SFF case with zero airflow like the A4 or Ghost S1 (one of the reasons why I'm not fond of those), a low profile cooler like the L9x65 or Blackridge, and a 290X dumping heat everywhere.



...which is why I suggested lorry poke around in that PBO submenu to see if anything will change. Everyone's experience varies. Mine very clearly clocks higher, but it lowers my CB R20 score slightly, presumably because in order to realize those PBO gains I have to return my Vcore to stock, which is quite hot and seems detrimental under heavy sustained load on my U9S.

This time around, reviews literally don't mean anything beyond VRM temps. There are so many factors influencing the performance you get; even having the same board is no guarantee, as some identical RAM kits use different DRAM and the wild silicon lottery takes care of the rest.

must admit that I've not checked temps on ram when running P95, but I presume that as it ran for over an hour that nothing was much wrong

Showing my absolute zero knowledge here - my gigabyte aorus rtx 2070 super (the 1905 factory OC one), does that dump a lot of its heat outside the case or inside?
Asking as mine is vertically mounted (proper vertical mount, actually sits where the PCI case brackets are). I mean, I've not seen the ram temps go above 31c (but like I said, didn't check whilst P95).

You say poke around in PBO, but not mention what can be adjusted, nor by how much. That's what is of interest to me, how much does say 1 unit on any particular setting alter things?

 

[Waldorf]

one reason i went LC for cpu and now gpu.
lowers case/vrm and many other temps by about 30*C, cpu stays below 70C and gpu below 55C under full load.
and that with 8dba fans throttled (full rpm reached at 40-60*C, depending what the fan is for.)

 

[tabascosauz]

must admit that I've not checked temps on ram when running P95, but I presume that as it ran for over an hour that nothing was much wrong

Showing my absolute zero knowledge here - my gigabyte aorus rtx 2070 super (the 1905 factory OC one), does that dump a lot of its heat outside the case or inside?
Asking as mine is vertically mounted (proper vertical mount, actually sits where the PCI case brackets are). I mean, I've not seen the ram temps go above 31c (but like I said, didn't check whilst P95).

You say poke around in PBO, but not mention what can be adjusted, nor by how much. That's what is of interest to me, how much does say 1 unit on any particular setting alter things?

It'll only heat up the RAM if you're running Large FFTs, as it's the IMC and DIMMs being stressed in that one (it'll fit pretty much as large a workload as it can into whatever amount RAM you have, for maximum RAM usage). Small and Smallest both use small enough tests to fit entirely in the CPU's L3. I don't think there's much else that can heat up those sticks aside from P95 Large.

I have a feeling single rank sticks could also run cooler than dual rank, but that depends on where the temp sensor is placed on the DIMM PCB, and I've no clue.

All axial coolers dump heat into the case, but I mentioned the 290X because Hawaii was a particularly hot and hungry chip and was up there with the likes of the GTX 480 and heavily OCed 7970 GHz in being space heaters. You have a big case; you'll be fine with good airflow.

This is the PBO submenu under AMD Overclocking if you set it to Advanced:

Because no one really knows how many of them make a difference, if any at all, just set PBO limits to Disable, then yeet PBO scalar and boost clock override to their maximum settings. For thermal limit, I put 95, but I'm pretty certain that setting does nothing. And if they don't make any observable difference at all, just turn PBO back to whatever setting you had before.

 

[lorry]

It'll only heat up the RAM if you're running Large FFTs, as it's the IMC and DIMMs being stressed in that one (it'll fit pretty much as large a workload as it can into whatever amount RAM you have, for maximum RAM usage). Small and Smallest both use small enough tests to fit entirely in the CPU's L3. I don't think there's much else that can heat up those sticks aside from P95 Large.

I have a feeling single rank sticks could also run cooler than dual rank, but that depends on where the temp sensor is placed on the DIMM PCB, and I've no clue.

All axial coolers dump heat into the case, but I mentioned the 290X because Hawaii was a particularly hot and hungry chip and was up there with the likes of the GTX 480 and heavily OCed 7970 GHz in being space heaters. You have a big case; you'll be fine with good airflow.

This is the PBO submenu under AMD Overclocking if you set it to Advanced:

View attachment 140682

Because no one really knows how many of them make a difference, if any at all, just set PBO limits to Disable, then yeet PBO scalar and boost clock override to their maximum settings. For thermal limit, I put 95, but I'm pretty certain that setting does nothing. And if they don't make any observable difference at all, just turn PBO back to whatever setting you had before.

At the moment I have PBO set to advanced & max CPU Boost Clock Override set to 200MHz but everything else in that section is untouched

I see that in your example you have PBO manual, PPT to 105, TDC to100 & EDC to 0 ??? What reasonable limits should they be?
Also scaler set to 10x
You say set PBO limits to disable yet show it as manual ?
Also you show platform thermal limit set to 0, I presume that is just an example and that it should be set to 95?

 

[tabascosauz]

At the moment I have PBO set to advanced & max CPU Boost Clock Override set to 200MHz but everything else in that section is untouched

I see that in your example you have PBO manual, PPT to 105, TDC to100 & EDC to 0 ??? What reasonable limits should they be?
Also scaler set to 10x
You say set PBO limits to disable yet show it as manual ?
Also you show platform thermal limit set to 0, I presume that is just an example and that it should be set to 95?

You can just set PBO Limits to Disable, it comes up as an option in the menu that pops up; it takes away all of PPT/TDC/EDC. I usually have all of PBO as Disabled for better performance and thermals, so I just switched it to Advanced to take a screenshot. You can set all three to an arbitrarily high number too, as it won't matter since the voltage limiter is still in place and it won't be drawing inane amounts of current and power.

No, I was just fooling around with 95 as that's the AMD-set throttle limit. You can set it to whatever you want, and I can't remember if there's also a Disable option for that. I don't think it makes much of a difference.

Again, don't sweat the details. I'm not convinced all of this has very much an effect on performance unless your chip is binned so amazingly well that it's capable of far more than the stock limiters allow.

 

[lorry]

You can just set PBO Limits to Disable, it comes up as an option in the menu that pops up; it takes away all of PPT/TDC/EDC. I usually have all of PBO as Disabled for better performance and thermals, so I just switched it to Advanced to take a screenshot. You can set all three to an arbitrarily high number too, as it won't matter since the voltage limiter is still in place and it won't be drawing inane amounts of current and power.

No, I was just fooling around with 95 as that's the AMD-set throttle limit. You can set it to whatever you want, and I can't remember if there's also a Disable option for that. I don't think it makes much of a difference.

Again, don't sweat the details. I'm not convinced all of this has very much an effect on performance unless your chip is binned so amazingly well that it's capable of far more than the stock limiters allow.

You have PBO set to disabled? Doesn't that then loose the boosting for a longer period? I thought the idea was to set it to advanced and then set 200MHz?
or did you mean setting PPT/TDC/EDC to disabled and then let the auto limiters take effect?

 

[tabascosauz]

You have PBO set to disabled? Doesn't that then loose the boosting for a longer period? I thought the idea was to set it to advanced and then set 200MHz?
or did you mean setting PPT/TDC/EDC to disabled and then let the auto limiters take effect?

PBO kills my performance. It varies for everybody, some will find that PBO boosts performance, and others will find it a detriment. Probably down to silicon quality variation, again. My chip hits something of a voltage wall after 4.2GHz; trying to manually run P95 Smallest after that just turns my rig into Reactor #4.

I was referring to the FIT voltage limiter, which will stop your CPU from doing stupid things even if you take all the other limits off regarding PPT (socket power) and temperature. It puts a ceiling on what Vcore can be safely given to you over the SVI2 bus, depending on the circumstances in the given moment.

The only testing I did was setting PPT/TDC/EDC all to 105, thermal to 95, scalar and override to max, and stock Vcore. It boosted one step past the maximum rated boost clock for the 3700X (4430MHz), but my gains were limited to CPU-Z bench, and the chip was quite a bit hotter.

Side note, if you use P95 Small or Smallest for testing conventional CPUs, it appears that in a recent AGESA version, AMD has set some sort of multiplier limit when running Small FFTs. Smallest usually caps out at 41x or so, which is usual for a heavy all-core load, but now Small is limited to about 39x, being the more intensive of the two tests. It wasn't like this, at least in 1.0.0.3AB, where Small used to pull the same 40.5x multipliers before dropping gradually. Either that, or the FIT has been updated to accommodate for Small. Or AMD might have implemented a power limiter for that.

 

[Zach_01]

I have tried the PBO limits up until 900 for all PPT/EDC/TDC and does nothing. The internal self regulator/manager of the CPU is setting the limits as programmed by AMD. This is only for board bypass restrictions.
The temp limit (thermal throttle) of the CPU by AMD is 95C and I never tried it to like 100C or more to see if anything changes.
CPU temp never exceeds 64C full load with low pump speed and medium fan rpm. Ambient 22C.

 

[lorry]

PBO kills my performance. It varies for everybody, some will find that PBO boosts performance, and others will find it a detriment. Probably down to silicon quality variation, again. My chip hits something of a voltage wall after 4.2GHz; trying to manually run P95 Smallest after that just turns my rig into Reactor #4.

I was referring to the FIT voltage limiter, which will stop your CPU from doing stupid things even if you take all the other limits off regarding PPT (socket power) and temperature. It puts a ceiling on what Vcore can be safely given to you over the SVI2 bus, depending on the circumstances in the given moment.

The only testing I did was setting PPT/TDC/EDC all to 105, thermal to 95, scalar and override to max, and stock Vcore. It boosted one step past the maximum rated boost clock for the 3700X (4430MHz), but my gains were limited to CPU-Z bench, and the chip was quite a bit hotter.

Side note, if you use P95 Small or Smallest for testing conventional CPUs, it appears that in a recent AGESA version, AMD has set some sort of multiplier limit when running Small FFTs. Smallest usually caps out at 41x or so, which is usual for a heavy all-core load, but now Small is limited to about 39x, being the more intensive of the two tests. It wasn't like this, at least in 1.0.0.3AB, where Small used to pull the same 40.5x multipliers before dropping gradually. Either that, or the FIT has been updated to accommodate for Small. Or AMD might have implemented a power limiter for that.

No idea What, or even Where, FIT voltage limiter is, or even if it is configurable ?

I have to ask all this because the MB manual doesn't even have a PBO section in the BIOS chapter!

When I run/ran P95 i hyst started the auto configured stress test, which i believe has a mixture of everything in it?

I have tried the PBO limits up until 900 for all PPT/EDC/TDC and does nothing. The internal self regulator/manager of the CPU is setting the limits as programmed by AMD. This is only for board bypass restrictions.
The temp limit (thermal throttle) of the CPU by AMD is 95C and I never tried it to like 100C or more to see if anything changes.
CPU temp never exceeds 64C full load with low pump speed and medium fan rpm. Ambient 22C.

I guess that I will have to go into BIOS to see exactly what mine does?
Or would HWINFO show me the max settings for PPT/EDC/TDC ?

 

[Zach_01]

I guess that I will have to go into BIOS to see exactly what mine does?
Or would HWINFO show me the max settings for PPT/EDC/TDC ?

HWiNFO simply read and report UEFI limits of PPT/EDC/TDC. If those are set to auto then it reports the default values of the specific CPU type as the 100%. If you set them manually for example to 900 then HW will take and report 900 as the 100%. RM also report the same.

 

[lorry]

HWiNFO simply read and report UEFI limits of PPT/EDC/TDC. If those are set to auto then it reports the default values of the specific CPU type as the 100%. If you set them manually for example to 900 then HW will take and report 900 as the 100%. RM also report the same.

So where would I find those figures online then please?
Are they Cpu specific or MB, or both?

 

[Zach_01]

You mean the default figures, right? Yes they are CPU specific by model. 3600, 3600X, 3700X... etc...

If now you have them auto, then you can see the max allowed values in RM, above histogram.

yours are these:
Temp throttle limit 95C
PPT 142W
TDC 95A
EDC 140A

 

[lorry]

You mean the default figures, right? Yes they are CPU specific by model. 3600, 3600X, 3700X... etc...

If now you have them auto, then you can see the max allowed values in RM, above histogram.

yours are these:
Temp throttle limit 95C
PPT 142W
TDC 95A
EDC 140A

Thanks, is that available online? Guess it has to be, but where please?

And also, is there any suggestions as to what you might increase those figures by? 5%, 10%? etc

 

[Zach_01]

I found that those numbers are 2 tier.
65W TDP CPUs. (like mine)
105W TDP CPUs (like yours)

This review of GamerNexus explains a lot for, PrecisionBoost, PBO and other factors of ZEN2 CPUs.
It’s very confusing...

Explaining AMD Ryzen Precision Boost Overdrive (PBO), AutoOC, & Benchmarks | GamersNexus

stub With the launch of the Ryzen 3000 series processors, we’ve noticed a distinct confusion among readers and viewers when it comes to the phrases “Precision Boost 2,” “XFR,” “Precision Boost Overdrive,” which is different from Precision Boost, and “AutoOC.” There is also a lot of confusion...

www.gamersnexus.net

 

[lorry]

I found that those numbers are 2 tier.
65W TDP CPUs. (like mine)
105W TDP CPUs (like yours)

This review of GamerNexus explains a lot for, PrecisionBoost, PBO and other factors of ZEN2 CPUs.
It’s very confusing...

Explaining AMD Ryzen Precision Boost Overdrive (PBO), AutoOC, & Benchmarks | GamersNexus

stub With the launch of the Ryzen 3000 series processors, we’ve noticed a distinct confusion among readers and viewers when it comes to the phrases “Precision Boost 2,” “XFR,” “Precision Boost Overdrive,” which is different from Precision Boost, and “AutoOC.” There is also a lot of confusion...

www.gamersnexus.net

Thanks and yes I completely agree with you about it being confusing, and I feel conflicting, information.
Take 1usmus power plan for instance, from what I can see, others weren't able (or willing?) to reproduce his results and claims, yet many individuals say that they have improved their own setups. So who is right in just that one particular case?
I also read about ZEN3 last night and that they will possibly see a 20% overall increase over ZEN2 with higher clock speeds matching Intel !

Zen 3 is rumored to be flaunting monumental IPC gains in early testing

As AMD confirmed at the HPC-AI Advisory Council UK conference, their next-generation server architecture, Milan, is being tested by their biggest customers. Milan and Ryzen 4000 will...

www.techspot.com

 

[Zach_01]

Thanks and yes I completely agree with you about it being confusing, and I feel conflicting, information.
Take 1usmus power plan for instance, from what I can see, others weren't able (or willing?) to reproduce his results and claims, yet many individuals say that they have improved their own setups. So who is right in just that one particular case?

I believe that users who claim not seeing anything with 1usmus plan are in 2 categories
1. They expected different things from what this power plan was meant to be
2. They did not set/install all that are required for it to work properly

I also read about ZEN3 last night and that they will possibly see a 20% overall increase over ZEN2 with higher clock speeds matching Intel !

Zen 3 is rumored to be flaunting monumental IPC gains in early testing

As AMD confirmed at the HPC-AI Advisory Council UK conference, their next-generation server architecture, Milan, is being tested by their biggest customers. Milan and Ryzen 4000 will...

www.techspot.com

I will read this... but I do not believe that we will see any time soon Intel like boost clocks. Those figures (of 5GHz) will become long past even for Intel after introducing a whole new 7nm architecture.

 

[lorry]

I believe that users who claim not seeing anything with 1usmus plan are in 2 categories
1. They expected different things from what this power plan was meant to be
2. They did not set/install all that are required for it to work properly


I will read this... but I do not believe that we will see any time soon Intel like boost clocks. Those figures (of 5GHz) will become long past even for Intel after introducing a whole new 7nm architecture.

I'm not so sure, ZEN3 looks to be radically different from ZEN2 from the little that I have read and I wouldn't rule anything out at the moment. Sure, when more leaks/info leaks out we will know a lot more, but for now anyway, AMD look to be the better bet when it comes to 7nm.

 

[Zach_01]

For now yes it is AMD better in 7nm because Intel doesn’t even have 7nm. And 10nm are still with a lot of issues. For now Intel is left behind, but will catch up on AMD in the next 2-3 years top.

As for ZEN3 I believe (and the techspot article said the same) that single core clocks will increase about 100-200MHz max. We may see a larger clock increases for all core clocks but not something ground breaking. The war of performance from now on will focus on the IPC gains and the multicore optimization.
It’s physically impossible to keep increasing clocks as more and more shrinking occurs and a higher density is achieved. The clock wall is near. And the shrinking wall is getting closer (~3nm) also, with the current UV/eUV method and the materials that manufacturers are using.

 

[lorry]

I found that those numbers are 2 tier.
65W TDP CPUs. (like mine)
105W TDP CPUs (like yours)

This review of GamerNexus explains a lot for, PrecisionBoost, PBO and other factors of ZEN2 CPUs.
It’s very confusing...

Explaining AMD Ryzen Precision Boost Overdrive (PBO), AutoOC, & Benchmarks | GamersNexus

stub With the launch of the Ryzen 3000 series processors, we’ve noticed a distinct confusion among readers and viewers when it comes to the phrases “Precision Boost 2,” “XFR,” “Precision Boost Overdrive,” which is different from Precision Boost, and “AutoOC.” There is also a lot of confusion...

www.gamersnexus.net

I had read that a while back, but rereading it now make it interesting reading. In R20 my single core results do not match there, with PBO on or off - they hit 520 whilst my very best so far is 502, but my multi core is 100 points higher, also for myself PBO does seem to increase multi core scores.
I feel that PBO is such a finicky thing even between the same setup but at different times, that it looks to be that there are no 'rules' or guidelines that can be given. Yes you Might increase your performance by a percent or two, but nothing significant beyond besting your best benchmark by a few points, certainly nothing usable in real use. By far the biggest factor that many didn't appreciate and still don't is temperature. i did realise that and the reason why I went for the best air cooling that I could at the time (even going to the expense of adding another Noctua Chromax fan to the NH DI-15S). I know that by far the best method is to get an open cooled loop and I am saving up for that, but at £400 roughly and limited funds, that won't happen next week either.

For now yes it is AMD better in 7nm because Intel doesn’t even have 7nm. And 10nm are still with a lot of issues. For now Intel is left behind, but will catch up on AMD in the next 2-3 years top.

As for ZEN3 I believe (and the techspot article said the same) that single core clocks will increase about 100-200MHz max. We may see a larger clock increases for all core clocks but not something ground breaking. The war of performance from now on will focus on the IPC gains and the multicore optimization.
It’s physically impossible to keep increasing clocks as more and more shrinking occurs and a higher density is achieved. The clock wall is near. And the shrinking wall is getting closer (~3nm) also, with the current UV/eUV method and the materials that manufacturers are using.

Yes Intel will, but also remember that AMD would have also been developing and fine-tuning their own 7nm products that would be already out. So, as to whether they match each other or one remains ahead only time will tell.

I have a distinct feeling that given the designs of both Intel and AMD, that Intel will likely always be ahead in the fastest speeds and that AMD will be ahead in multicore usage. Unless of course either or both of them radically change the designs of their CPUs.

I feel that the clock wall being near is relative though. It is near, but I also do not expect it to be reached within the next couple of years. I am sure that the actual production process will be far more of a hindrance as the chiplets become thinner and that will then slow the progress to the thinnest wafer that can be made.

It may even be that we see a radical change in design. For instance, instead of going thinner and thinner, reaching an obvious end point, what if say AMD were able to develop and expand upon the Infinity Fabric to increase the distance between cores without loss of communication speed? That way you even see the size of the CPU increase significantly to allow distance between everything and dissipate the generated heat better, thus increasing performance that way?