Trying to understand Ryzen 3000 series boost speed variations

[Waldorf]

@tabascosauz/lorry
had the same weird stuff until another bios update,
and the only time i saw 2 betas very shortly after another, before a full (non beta) came out

 

[lorry]

As for TSMC and Apple, Apple is either 1/2 or 1 process generation ahead. Same story for Qualcomm, 855/865 on N7/N7P. Like Ryzen 3000, the A12 in the XR was manufactured on N7FF, but the current A13 in the 11s is made on N7P, which is a product-improved N7, but not a more revolutionary upgrade that the EUV N7+ will be. That, and ARM has always been efficiency-minded due to its intended form factors (even performance-leading Apple SoCs), even discounting all the other big differences in architecture and performance between x86 and ARMv8.

Apple also isn't the one pushing the absolute performance limit out of N7, whereas AMD is. Around base clock to 4GHz, Vcore and thermals stay around 1.0V on Ryzen 3000, but Vcore blasts off as soon as we start approaching rated max boost clocks (save for that elusive 1-in-a-thousand unicorn chip). The jump to "sub-10nm" is markedly more efficient, but only if manufacturers do as Ice Lake does and keep clocks low. AMD doesn't want that, as they want to stick it to Intel, which is what got us here today, pushing N7FF a bit harder than it really wants to be pushed. For all the other ARM manufacturers, that level of power consumption and heat increase due to aggressively pushing the envelope would be unacceptable, as that would be big SD810 vibes all over again.

I get those fast and slow boots too, both before POST and during Windows loading, neither of which happen on my Intel PCs. Every time I change the timings on this 32GB kit, I get really fast pre-POST times for a few restarts, then it's back to the long POST. Almost every timing is manually filled out now, including tRFC2/4, doesn't make a difference for POST.

The unresponsive BIOS might be down to your board or the BIOS written for it. My F50 BIOS should be functionally identical as are most 400-series, just for a different board, doesn't do that. But I used to get similar unresponsive behaviour in Windows right after boot, like all the cores were limited to 200MHz or something. A quick restart would solve it, and it was a very rare occurrence that went away when I switched RAM kits, but still unexplained. That might also have been F42g or F50a, can't remember.

Hope your dog gets back to his usual self soon.

ARM though is aimed at a totally different market though, even if it is really two markets nowadays (mobile and extremely small form factor), and both have cooling restrictions due to their size. Neither can afford to push speeds as hard as AMD are able to do (just). I think that once you begin to push (overclock) yourself, you begin to realise/appreciate just How hard AMD are doing so out of the box.

My slow posts seem to be intermittent, with no reason that I can see for them to occur, as the bios will seem to 'hang' for a while on different Q-codes, but the ones that seem to occur the most are -
F8 Recovery PPI is invalid and more occasionally 6* CPU DXE initialization is started. It also never hangs once in Win itself, mine are All during the post

My dog seems to have been 'lucky' as in he is responsive etc this morning. he should hopefully be back on the right track once he eats and has insulin again at his regular time in roughly 3 hours
Thanks

Usually not taking advantage of some new features. Some instances they could fix some things with drivers.
Win10 v1909, November chipset drivers from AMD, new UEFI version, all came close to one another and after first (v1.0) of 1usmus power plan. Its not a coincidence...

I only asked as of course my win build was 1909 right from the word go and i spent very little time on the old chipset, as I updated to it almost immediately, so have little to base any difference on.

One interesting thing though is that I Had to drop down to 6 cores Just so that I could install windows! it would not complete the installation otherwise.
Once installed though and I had installed the Gigabyte App Centre - that then updated the Intel wi-fi driver and I was then able to go back into bios to enable all 12 cores.

This is why I feel that this particular motherboard has some peculiar issues with F50 and will hopefully be sorted out with the next bios update.

@tabascosauz/lorry
had the same weird stuff until another bios update,
and the only time i saw 2 betas very shortly after another, before a full (non beta) came out

I feel that on reflection, one bios cannot be completely compatible with Every single motherboard out there.

 

[Deleted member 185158]

went away when I switched RAM kits

That's where having a few memory kits comes in handy.

Might be wise to back the memory overclock down a bit and see if the Bios issues go away.

 

[lorry]

That's where having a few memory kits comes in handy.

Might be wise to back the memory overclock down a bit and see if the Bios issues go away.

"went away when I switched RAM kits"

Not sure who you are replying to there but it wasn't myself?

 

[Zach_01]

"went away when I switched RAM kits"

Not sure who you are replying to there but it wasn't myself?

The...
"went away when I switched RAM kits"
...was said by @tabascosauz at post #400 about the bios issues.

But I believe that @ShrimpBrime ‘s suggestion to backdown the mem oc is for you too... and to see if anything occurs again. Does not matter if you have no errors or any obvious mishaps while in Windows.

 

[lorry]

The...
"went away when I switched RAM kits"
...was said by @tabascosauz at post #400 about the bios issues.

But I believe that @ShrimpBrime ‘s suggestion to backdown the mem oc is for you too... and to see if anything occurs again. Does not matter if you have no errors or any obvious mishaps while in Windows.

I have been having these occasional sluggish bios access issues from the start, near enough.
A save and exit (reboot) has always then seen the mouse/keyboard /bios return to it usual responsive

 

[tabascosauz]

@lorry I got you beat on that ST score, don't ask me how, I've no idea :



I guess 1usmus must be putting my Cinebench ST test on Core 5 and 7, because if it was on Core 0 then I'd be way down in Zen+ and Intel territory. But 7300pts is very respectable for your 3900X. 500pts is also pretty much at the top of the pack, so could be down to chip variance or normal benchmark variation.

I feel that on reflection, one bios cannot be completely compatible with Every single motherboard out there.

You are undoubtedly correct; however, Gigabyte evidently has a different opinion on that matter. Though in all fairness to them, all the vendors do it to some extent - no sense in writing something new from the ground up for every board of a given generation. Then again, it's Gigabyte, so they're copy-pasting hot garbage in place of a polished and verified BIOS.

Haven't heard anything regarding the next iteration of AGESA for Matisse, so you may be stuck with F50 for a while. Is it possible to downgrade Gigabyte BIOSes?


@NoJuan999 you are not on the "latest" Windows, 1909 has a new scheduler that v1.1 of 1usmus is designed to work with; you're supposed to use the Universal plan if you don't have 3000 on 1909, and Ryzen Power Plan if you have 3000 on 1909. Also, "2.2GHz" and "3.6GHz" minimum clockspeed is a HWInfo bug, your chip is doing a lot of idle behind the scenes that only Ryzen Master can show.

 

[NoJuan999]

I will put my 2 cents in here, the difference in performance between the various Power Plans is negligible.
I have Windows 10 (1903 Build 18362.535), the latest AMD Chipset drivers and the newest BIOS for my MB and I personally choose to use the AMD Ryzen balanced power Plan.
I have tried the 1usmus Ryzen Power Plan and it keeps my clocks set at 3.6GHz (no matter what my minimum processor state is set at) vs 2.2GHz minimum with the AMD Ryzen Balanced Power Plan set at 20% minimum/100% maximum.
No matter what Power Plan I use, I get the same multi and Single Core Performance.
So for me the AMD Ryzen Balanced Plan is the best option for performance vs power consumption/temps.
Therefore in my case the 1usmus and Power Saving plans are a waste of time.

And I have even tried the Ultimate Power plan (min and max set at 100%) and the difference in performance between them is negligible.
So I use the AMD Balanced Power Plan (set at 20%/100%) for everyday use on my 3700x.

I also run my 3600 G.Skill Ripjaw V (Hynix D die) RAM kit at 3733 with FCLK (IF) set at 1867 and it is 100% stable.

My best R20 run so far (using the AMD Ryzen Balanced Power Plan):

 

[lorry]

I will put my 2 cents in here.
I have the latest version of Windows 10 (1903 Build 18362.535), the latest AMD Chipset drivers and the newest BIOS for my MB and I personally choose to use the AMD Ryzen balanced power Plan.
I have tried the 1usmus Ryzen Power Plan and it keeps my clocks set at 3.6GHz (no matter what my minimum processor state is set at) vs 2.2GHz minimum with the AMD Ryzen Balanced Power Plan set at 20% minimum/100% maximum.
No matter what Power Plan I use, I get the same multi and Single Core Performance.
So for me the AMD Ryzen Balanced Plan is the best option for performance vs power consumption/temps.
Therefore in my case the 1usmus and Power Saving plans are a waste of time.

And I have even tried the Ultimate Power plan (min and max set at 100%) and the difference in performance between them is negligible.
So I use the AMD Balanced Power Plan (set at 20%/100%) for everyday use on my 3700x.

I also run my 3600 G.Skill Ripjaw V (Hynix D die) RAM kit at 3733 with FCLK (IF) set at 1867 and it is 100% stable.

My best R20 run so far:View attachment 140585

Latest build though is 1909?

 

[NoJuan999]

Latest build though is 1909?

Last Post edited.
Version 1903 Build 18362.535 is the December 10, 2019 update and the newest version available to me.

 

[lorry]

@lorry I got you beat on that ST score, don't ask me how, I've no idea :

View attachment 140583

I guess 1usmus must be putting my Cinebench ST test on Core 5 and 7, because if it was on Core 0 then I'd be way down in Zen+ and Intel territory. But 7300pts is very respectable for your 3900X. 500pts is also pretty much at the top of the pack, so could be down to chip variance or normal benchmark variation.

I feel that on reflection, one bios cannot be completely compatible with Every single motherboard out there.

You are undoubtedly correct; however, Gigabyte evidently has a different opinion on that matter. Though in all fairness to them, all the vendors do it to some extent - no sense in writing something new from the ground up for every board of a given generation. Then again, it's Gigabyte, so they're copy-pasting hot garbage in place of a polished and verified BIOS.

Haven't heard anything regarding the next iteration of AGESA for Matisse, so you may be stuck with F50 for a while. Is it possible to downgrade Gigabyte BIOSes?


@NoJuan999 you are not on the "latest" Windows, 1909 has a new scheduler that v1.1 of 1usmus is designed to work with; you're supposed to use the Universal plan if you don't have 3000 on 1909, and Ryzen Power Plan if you have 3000 on 1909. Also, "2.2GHz" and "3.6GHz" minimum clockspeed is a HWInfo bug, your chip is doing a lot of idle behind the scenes that only Ryzen Master can show.

I have a feeling that CB is quite fickle when it wants to be, lol. For ages I was stuck on 49* then hit 500,why I don't know, but we'll done, that's a great score there.

I thought that gigabyte generally had a decent enough reputation?

This board has a dual BIOS (another reason I decided on it), so I'm completely unsure if downgrading the bios is even possible, why would that be an idea though?

Last Post edited.
Version 1903 Build 18362.535 is the December 10, 2019 update and the newest version available to me.

Ah! Understood.

I have the distinct feeling that with Ryzen 3000 not only comparing apples with pears is impossible, but also comparing cox's apples with cranberry apples is as well.

In other words unless each two setups are physically the same, then different settings are extremely difficult to compare

 

[tabascosauz]

I have a feeling that CB is quite fickle when it wants to be, lol. For ages I was stuck on 49* then hit 500,why I don't know, but we'll done, that's a great score there.

I thought that gigabyte generally had a decent enough reputation?

This board has a dual BIOS (another reason I decided on it), so I'm completely unsure if downgrading the bios is even possible, why would that be an idea though?

They've got solid hardware, but their BIOS designs have always leaned towards dead last. I don't like Asus boards for their price and AM4 stock overvolting but their current BIOSes since Skylake have been really, really good. I miss DualBIOS, most GB Intel boards had that.

It's just an idea I've been toying with. My clocks have dropped slightly on 1.0.0.4 (and probably my benchmark scores, if I had tested R20 on ABBA) compared to 1.0.0.3ABBA which came in F42g and 1.0.0.3AB in F42c. Not sure I want to put up with the crazy idle again (which 1.0.0.4 fixed), but I have lost a bit of performance. Note that @NoJuan999's scores are on pre-1.0.0.4 AGESA (but his chip is probably binned better anyway).

Also no idea how 1usmus and Chipset Drivers would react to going backwards to 1.0.0.3ABBA, so I probably won't go be the guinea pig, lol.

 

[NoJuan999]

@NoJuan999 you are not on the "latest" Windows, 1909 has a new scheduler that v1.1 of 1usmus is designed to work with; you're supposed to use the Universal plan if you don't have 3000 on 1909, and Ryzen Power Plan if you have 3000 on 1909. Also, "2.2GHz" and "3.6GHz" minimum clockspeed is a HWInfo bug, your chip is doing a lot of idle behind the scenes that only Ryzen Master can show.

I do understand that HWinfo doesn't register the fact that most of my cores are sleeping and only shows the minimumas 2.2GHz, but when I use the 1usmus plan it shows my minimum clocks at 3.6 GHz.
My point was that with the AMD Balanced Plan my temps are lower at idle, But my CB R20 scores are the same.

PS
I am Installing Version 1909 now to see if the 1usmus power plan improves my performance.

 

[tabascosauz]

I do understand that HWinfo doesn't register the fact that most of my cores are sleeping and only shows the minimumas 2.2GHz, but when I use the 1usmus plan it shows my minimum clocks at 3.6 GHz.
My point was that with the AMD Balanced Plan my temps are lower at idle, But my CB R20 scores are the same.

PS
I am Installing Version 1909 now to see if the 1usmus power plan improves my performance.

Are your idle temps actually consistently lower on Ryzen Balanced over a number of days, or did you just happen to have more background tasks going on the day that you decided to switch? The idle temps on these chips fluctuate way too much to properly measure. I can't think of any reason why the 1usmus plan would run hotter; it pretty much just shifts the load around and parks some cores. Your R20 scores should be the same, it doesn't really give you any real increase in fully threaded performance.

 

[oxrufiioxo]

single core for a 3900X should be in this range. I've seen as high as 535.

 

[NoJuan999]

Are your idle temps actually consistently lower on Ryzen Balanced over a number of days, or did you just happen to have more background tasks going on the day that you decided to switch? The idle temps on these chips fluctuate way too much to properly measure. I can't think of any reason why the 1usmus plan would run hotter; it pretty much just shifts the load around and parks some cores. Your R20 scores should be the same, it doesn't really give you any real increase in fully threaded performance.

The idle temps in RM were consistently around 5c higher when using the 1usmus plan on 1903.
I just manually updated to 1909 so I will see if that is still the case now and post back.

 

[lorry]

They've got solid hardware, but their BIOS designs have always leaned towards dead last. I don't like Asus boards for their price and AM4 stock overvolting but their current BIOSes since Skylake have been really, really good. I miss DualBIOS, most GB Intel boards had that.

It's just an idea I've been toying with. My clocks have dropped slightly on 1.0.0.4 (and probably my benchmark scores, if I had tested R20 on ABBA) compared to 1.0.0.3ABBA which came in F42g and 1.0.0.3AB in F42c. Not sure I want to put up with the crazy idle again (which 1.0.0.4 fixed), but I have lost a bit of performance. Note that @NoJuan999's scores are on pre-1.0.0.4 AGESA (but his chip is probably binned better anyway).

Also no idea how 1usmus and Chipset Drivers would react to going backwards to 1.0.0.3ABBA, so I probably won't go be the guinea pig, lol.

i don't fancy going backwards in bios, especially if i have no obvious reason to, which I don't. i have only really used 1.0.0.4

They've got solid hardware, but their BIOS designs have always leaned towards dead last. I don't like Asus boards for their price and AM4 stock overvolting but their current BIOSes since Skylake have been really, really good. I miss DualBIOS, most GB Intel boards had that.

the BIOS being clunky is an aesthetic though, doesn't affect how it works at all, I can show you any number of bikes and cars that are as ugly as sin and a bitch to ride/drive but they go like a bat out of hell.
Their design reminds me somewhat of the old Nokia mobiles where you had to drill down through the menu to get to the bit that you wanted.

Also just noticed a massive temperature reading difference between RM and HW

 

[tabascosauz]

i don't fancy going backwards in bios, especially if i have no obvious reason to, which I don't. i have only really used 1.0.0.4

the BIOS being clunky is an aesthetic though, doesn't affect how it works at all, I can show you any number of bikes and cars that are as ugly as sin and a bitch to ride/drive but they go like a bat out of hell.
Their design reminds me somewhat of the old Nokia mobiles where you had to drill down through the menu to get to the bit that you wanted.

Also just noticed a massive temperature reading difference between RM and HW

For me, having duplicate RGB settings and redundant / system-breaking (somehow) voltage settings, as well as the hilarious "PWM/°C" setting in older GB BIOSes is pretty indicative of where GB is at in BIOS design, haha. It works for the most part (as every BIOS should, at the minimum), but I haven't yet been able to change SoC voltage without either causing major errors or nuking my Windows installation.

The RM reading is usually closer to the new CPU Die (average) reading added a few revisions back, than the old Tctl/Tdie, but unfortunately, all BIOSes as far as I can tell still only read Tctl/Tdie.

Also, have you fooled around with the PBO power limits and temperature limits? I eked out a bit extra in clocks and a CPU-Z bench, but had no improvement in CB R20 while drawing about 90W. Maybe I'm reaching the limits of the U9S and my chip, but yours might fare better. Maybe you'll even get a little surprise like this:



Louder fans, hotter chip, more power draw, and stock voltage all for some HWInfo and CPU-Z e-peen. I went right back to PBO disabled afterwards, lol.

 

[Zach_01]

The 1usmus power plans (both) are primarily designed for optimal core loading. What it does, is to keep Windows’s scheduler to load the best/high quality cores and prevent the threads from jumping around to all cores CCXs and CCDs in single and light threaded situations. Those are the perf# numbers in HWiNFO. I remind (to all) that this requires UEFI settings to work properly, the right plan version for the right win version. Latest win version, latest chipset drivers and latest UEFI version are also implement the same logic.
These plans are specifically designed for 2 CCD CPUs like the 3900X/3950X but I saw difference in loading at mine with 1 CCD.

————————

You can go back in BIOS version only if the board has bios flashback button. Otherwise I think the integrated utility does not accept earlier version.

————————

The RM temp report is matching the average report from HWiNFO.

————————

I remember that different AGESA versions required different PBO settings in order to keep single core CPU boosting the same. And it’s not universal settings. Every board/bios has its own response to AGESA. One must try different settings to find the best.

 

[lorry]

For me, having duplicate RGB settings and redundant / system-breaking (somehow) voltage settings, as well as the hilarious "PWM/°C" setting in older GB BIOSes is pretty indicative of where GB is at in BIOS design, haha. It works for the most part (as every BIOS should, at the minimum), but I haven't yet been able to change SoC voltage without either causing major errors or nuking my Windows installation.

The RM reading is usually closer to the new CPU Die (average) reading added a few revisions back, than the old Tctl/Tdie, but unfortunately, all BIOSes as far as I can tell still only read Tctl/Tdie.

Also, have you fooled around with the PBO power limits and temperature limits? I eked out a bit extra in clocks and a CPU-Z bench, but had no improvement in CB R20 while drawing about 90W. Maybe I'm reaching the limits of the U9S and my chip, but yours might fare better. Maybe you'll even get a little surprise like this:

View attachment 140597

Louder fans, hotter chip, more power draw, and stock voltage all for some HWInfo and CPU-Z e-peen. I went right back to PBO disabled afterwards, lol.

The 1usmus power plans (both) are primarily designed for optimal core loading. What it does, is to keep Windows’s scheduler to load the best/high quality cores and prevent the threads from jumping around to all cores CCXs and CCDs in single and light threaded situations. Those are the perf# numbers in HWiNFO. I remind (to all) that this requires UEFI settings to work properly, the right plan version for the right win version. Latest win version, latest chipset drivers and latest UEFI version are also implement the same logic.
These plans are specifically designed for 2 CCD CPUs like the 3900X/3950X but I saw difference in loading at mine with 1 CCD.

————————

You can go back in BIOS version only if the board has bios flashback button. Otherwise I think the integrated utility does not accept earlier version.

————————

The RM temp report is matching the average report from HWiNFO.

————————

I remember that different AGESA versions required different PBO settings in order to keep single core CPU boosting the same. And it’s not universal settings. Every board/bios has its own response to AGESA. One must try different settings to find the best.

________________________________

I can state, having watched thread usage in HWINFO during various benchmarks, that the thread load is mainly spaced out over the 3 best threads and that even shows in Core Temp as well

________________________________

This MB only supports Q-FLASH, Not flashback, sadly

________________________________

the RM temp is Not matching the HWINFO average with mine, if you look you will see that HWINFO shows the average as 46C yet RM reports 40.2C

________________________________

I haven't tried other settings really, as you know.

 

[Zach_01]

View attachment 140600



________________________________

I can state, having watched thread usage in HWINFO during various benchmarks, that the thread load is mainly spaced out over the 3 best threads and that even shows in Core Temp as well

________________________________

This MB only supports Q-FLASH, Not flashback, sadly

________________________________

the RM temp is Not matching the HWINFO average with mine, if you look you will see that HWINFO shows the average as 46C yet RM reports 40.2C

________________________________

I haven't tried other settings really, as you know.

View attachment 140605

Yes I know that you know... I wrote all the above for all the conversation participants.

about the temp...
RM temp is matching the current value of “CPU DIE (average)” not the average value of Tctl/Tdie.
That’s the first value of second green line for your HW that matches the RM.

 

[lorry]

Yes I know that you know... I wrote all the above for all the conversation participants.

about the temp...
RM temp is matching the current value of “CPU DIE (average)” not the average value of Tctl/Tdie.
That’s the first value of second green line for your HW that matches the RM.

Ah okay, except that RM seems to be more responsive than HWINFO. The temp spikes last a lot longer in HWINFO than on RM, just in case no one had noriced that

 

[Zach_01]

From what I’ve seen on mine the RM temp seem more stable on low-middle values than any temp of HWiNFO. As I said before, a few days ago, AMD uses a proprietary report method in RM not known (it’s undisclosed) publicly, and how much of accurate it is.
Personally I don’t pay much attention to RM for temp.

 

[lorry]

From what I’ve seen on mine the RM temp seem more stable on low-middle values than any temp of HWiNFO. As I said before, a few days ago, AMD uses a proprietary report method in RM not known (it’s undisclosed) publicly, and how much of accurate it is.
Personally I don’t pay much attention to RM for temp.

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

 

[Waldorf]

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...