Overclock.net banner
7341 - 7360 of 8974 Posts

·
Noctua Fanboy
Joined
·
1,095 Posts
This was tested using Wprime and setting the thread setting, frequency monitored in Ryzen Master and all background apps where closed. This is running on W10 1909 and using the latest chipset driver from Amd(3-20-20). PBO Scalar testing a whopping 30-40mhz increase lol!

In a month or 2 i might be getting a 3900X to play with and sell this 3700X it was slightly abused but not by much i can't believe the used market for these why people are willing to pay 80%+ of the cost for a used part i have no idea but whatever haha personally i wouldn't pay over 250$ for a 3700X in today's world new with Zen 3 coming. This virus is having me go nuts with boredom but hey? Wish Navi 2 or geforce 3000 series would launch now instead of the end of the year.
 

Attachments

·
Overclock the World
Joined
·
2,995 Posts
Soo , 6.20A is an interesting bios
This is part 1 of the changes:

To my sources, JZ loves the performance with an offset
Auto values default to -0.625mV offset
Which makes the 3950X consume 200W boosting, and 180W AVX2 folding and pi-calculation

No Microcode, no EFI module updates, no new AMD CBS settings and no new AGESA
No new renoire modules in the bios
But a huge chunk of code changed
Investigating - need probably 1 more day to doublecheck everything
Could be PMU update, could be something else
I think it's PMU or SMU , by the location and chunk of code size
~ will update :)

focus on this location
Edit:
I would appreciate, if testers who stay on auto without PBO could doublecheck their applied SOC voltage under 3600 MT/s and higher
it lost 50mV defaulting to 1.028v instead of 1.087v (which are 50W savings on TDP)
- need information on how it scales now with MT/s
* consider to doublecheck vSOC and if you will have issues with FLCK >1800 without forcing SOC voltage by yourself
 

·
Noctua Fanboy
Joined
·
1,095 Posts
"To my sources, JZ loves the performance with an offset
Auto values default to -0.625mV offset" What does that mean exactly?

Veil please keep us posted man i was thinking about just skipping this bios but perhaps i should give her a try?

Been getting back into my other hobby which is just as expensive but more annoying due to the subjective matter on the subject. lol(headphones).
 

·
Overclock the World
Joined
·
2,995 Posts
"To my sources, JZ loves the performance with an offset
Auto values default to -0.625mV offset" What does that mean exactly?

Veii please keep us posted man i was thinking about just skipping this bios but perhaps i should give her a try?

Been getting back into my other hobby which is just as expensive but more annoying due to the subjective matter on the subject. lol(headphones).
On the screenshot you can see that Auto = -625mV
Optimal values are loaded as auto values
Failsafe are loaded on cmos rest or just bios reset

It changed to Offset mode, with a fixed offset of -625mV (Auto = offset mode)
i haven't seen any PBO flags active and there are no AGESA changes whatsoever
Soo keep it on auto and test without PBO :)
Also keep an eye on SOC voltage, it seems to be surprisingly low
Surprising, but nothing new ~ considering how 2nd gen vSOC range looks like :)

EDIT:
But there is more to this bios than just what we see here
Investigating, because JZ doesn't want to tell what's different :ninja:
 

·
Noctua Fanboy
Joined
·
1,095 Posts
Veil i tested it on auto everything but my timings and such and SOC still wants to go to a crazy 1.2V!

VDDP+VDDG voltage is the same 1.1V setting for each too on auto.


However it is actually hitting 4.4Ghz more often and this is no placebo i watch this stuff far to much to be considered sane lol


Yes for sure it is hitting 4.4ghz even during the boosttester software which is nuts its hitting it so much i'm wondering if perhaps if i turn on AutoOC if it would hit 4425mhz again?

This is NOT The same 6.2 Bios for sure its NOT something in here is tweaked.

I checked Vcore and its still at 1.472V as max but VID seems to be hitting 1.5V more now but it still kind of looks the same?

Not sure what magic Asrock or Jz did but something is for sure better everyone give it a try!!

I shut my PC off 3 times and i still get better boost speeds including resetting my bios this is for sure not a fluke

Not just clock stretching either!

I'm not getting off this bios and i'll be saving the file for the future for my 3900X just in case JZ takes it off their site or something.

Edit bingo 4425mhz again best bios since AGESA 1.0.0.3ABBA

Memory latency also dropped 0.8ns or so. 62.6ns now instead of 63.5
 

Attachments

·
Noctua Fanboy
Joined
·
1,095 Posts
Is it ok to be used on x470 Taichi?
Veil for sure would know that please don't try it until he says its ok

2T boost is slightly better as well but mainly the gains are in ST turbo and it seems like all 8 cores stick to 4325mhz on basically any lighter load, it's 5am here and stuck inside so i'll check tomorrow for games.
 

·
Registered
Joined
·
684 Posts
Veil i tested it on auto everything but my timings and such and SOC still wants to go to a crazy 1.2V!

VDDP+VDDG voltage is the same 1.1V setting for each too on auto.


However it is actually hitting 4.4Ghz more often and this is no placebo i watch this stuff far to much to be considered sane lol


Yes for sure it is hitting 4.4ghz even during the boosttester software which is nuts its hitting it so much i'm wondering if perhaps if i turn on AutoOC if it would hit 4425mhz again?

This is NOT The same 6.2 Bios for sure its NOT something in here is tweaked.

I checked Vcore and its still at 1.472V as max but VID seems to be hitting 1.5V more now but it still kind of looks the same?

Not sure what magic Asrock or Jz did but something is for sure better everyone give it a try!!

I shut my PC off 3 times and i still get better boost speeds including resetting my bios this is for sure not a fluke

Not just clock stretching either!

I'm not getting off this bios and i'll be saving the file for the future for my 3900X just in case JZ takes it off their site or something.

Edit bingo 4425mhz again best bios since AGESA 1.0.0.3ABBA

Memory latency also dropped 0.8ns or so. 62.6ns now instead of 63.5
Same here, VSOC on auto go to 1.2v!!

1.48v for CPU on auto
 

·
Riding on the edge
Joined
·
3,275 Posts
So if I'm still rocking a 1800X what bios should I install? Still on 5.50. Just 5.60? Is it worth upgrading beyond that?



If it's of any help I run a 2700x (Gen+) and with my testing on a C6H in rig sig I've settled very happily for AGESA 1.0.0.6.

I have played with all the newer versions after that and came back to good ole reliable.

After this is the where they (AMD) started adding features for the new gen2 CPUs. Buggy this, broken that (might not be this way for Asrock) but if I had to guess things aren't much different.
 

·
Overclock the World
Joined
·
2,995 Posts
First of all, thank you @iNeri & @polkfan for the reports
Speaking of, who is this Veil :p

Does anyone of you have a SPI flasher, or has an old flashtool backup for warranty recovery ?
We have some very significant changes on this new Bios
(1600 lines of code 100 modules i need to back&forth check ~ takes time, sorry)

But yes, we have very significant changes on 6.20A (full change-log in work)
To list a few:
- L1/2 cache mitigation patches
- AMD SMU firmware updates & new lockdowns
- s̶o̶m̶e̶ ̶t̶o̶u̶c̶h̶e̶s̶ ̶o̶n̶ ̶P̶M̶U̶ (EDIT: non changes, it's not under PMU)
- changed per CCX communication
- updated firmwares (CPM power management & FIT module, SOC range, smbus, AMD PSP recovery and access focusing on PCIe 4.0, NBIO GFX prediction and IOMMU update)
- chipset updates cover (Tiashan,Sandstone,Promontory,Huashan)
- updated firmware mosfets & vrm, also LLC behavior for: UP1674,IR3520
- updated SOC & Fabric communication for Mattise,Summit Ridge & Raven ridge
- updated PHY Memory Training algorithms (part of PMU but not everything)
apparently also AGESA touches without version update but covering nearly the whole package
this appears to me like pre-1005
sadly nothing Renoir related yet :eek: / although there is some new no-name module (exploring) :typer:

No wonder why JZ didn't want to spoil anything about this "gift" :D

I wish you wouldn't have updated tho, at least not everyone,
because AMD PSP updates are one way and integrated in the cpu only crosstalking with SMU
Meaning, once transmitted and sectors updated - our PSPTool might not work and SMU adress changes
* in the works of exporting anything, although as it's HMAC/SHA1 encrypted and likely security key changed ~ i'm not sure if we can transplant that over without updating PSP Firmware
** PSP Firmware access is crucial to even consider having any chance of getting access to the chips and removing PCIe 4.0 lockdowns (which need not only a firmware change, but also a bios change + leaked secret key)
Long story short, on one hand we have ROWHammer and L2 Bug fixes and new great fixes, but this might forever lockdown any potential chance of getting PCIe 4.0 to work

To everyone who hasn't updated:
Please doublecheck 4 things:
- visit https://github.com/irusanov/SMUDebugTool/ and help @infraredbg a bit by submitting SMU reports before this big update and after with corresponding new SMU Version
- make a full SiSoftware Sandra Multi-Core efficiency report (detailed) and export it, we need to check cross-CCX latency and communication speed <- this goes for before and after update
- Doublecheck your procODT range and list your selectable values (do we have 28Ω, do we have whole value 30Ω...)
- at last, if SiSandra reports behave how i think they would, try if you can pass 1930-1960FLCK in async mode on default auto mode
* while exporting SiSandra results, please consider to make no XMP 1200FLCK/1200MCLK/1200UCLK (all auto) tests and compare with your current running 1800/1900MLCK results (Detailed result, just untick online results)

@gupsterg you are experienced here,
what would be the most usable way of exporting the brand new SMU(PMU i know)
Issue is, they are encrypted - but communication should still work with 1004B PSP firmware without breaking CRC
Should i directly export PEI & DXE modules, or the old way of only HEX ?
Need to lookup your old tutorials and crosscompare, if sectors are similar
I can see we got AMI Update & Rom Layout update, stuff has shuffled around
 

·
Overclock the World
Joined
·
2,995 Posts
So if I'm still rocking a 1800X what bios should I install? Still on 5.50. Just 5.60? Is it worth upgrading beyond that?
We have a bit of issues with last bioses till 5.80, a lot of them where fully broken
The only functional bios that i found to be bugfree, was X470 3.60 modded
All bioses including 5.80 have throttle issues because tCTL & tDIE tracking is messed up
usually 0.0.7.2 with PMU update (memory controller) was the "best bios" for ryzen 1st gen
But 1003ABBA does work well and has this PMU Firware already, just <1003ABB has bugs - far to many of them including RGB controller issues

Without much work 5.80 is your best bet, even 5.10 is bugged (which is a bridge bios)
Wouldn't use 6.20 on first gen (6.20A likely yes, but wait a bit on that one)
Else if you are brave enough, best results you get on X370 to X470 Mod bios for first two ryzen gens - done with flashtool
Only bug that exists is RGB change from the Bios, rest works very well
Is it ok to be used on x470 Taichi?
Do you have a SPI flasher just in case?
To use X370 on X470 you optimally need to extend the rom to twice it's size
Adding another 16777216 b or 1000000h of FF
Take as example X470v3.60 The Stilt release as example:
https://1drv.ms/u/s!Ag6oE4SOsCmDhhEta6YIy70AhNAi from this thread
Usually you don't have to, bioses are still 16mb but X470 Taichi has a 32mb chip
To make flashing without issues and keep up official update path, better extend the size of the rom you are going to SPI flash with Flashtool (thread mentioned above, check #5)

Asking for a HW SPI flasher, because i would ask for someone to act as testing rabbit
ASRocks security check capsule should be broken now, i know it's location
Would love to test if it will accept modified Capsuled ROMs now, after it's gone
99% sure it will work, 1% unsure if it won't break signature afterwards ~ soo HW Flasher is recommended to check before SPI flashing

Else you can do the old methods, just increase size soo you retain the update path
If you want to keep warranty up, you need to first flashtool export your original bios
And transfer over the ethernet mac adress and serial number with a 3rd party tool before you use flashtool to clean SPI flash
Just again, wait a bit before you do that as this Bios is very experimental
And you will need to play with ASRock polychrome
You can go that path , but just wait at least one more day
~ can provide you later this extended rom package, and try if you have a HWFlasher to break ASRocks security capsule permanently
Just give me a bit of time, too many things at once to finish
* it might make more sense to transplant this changes into 3.92 from JZ page instead of update PSP Firmware (is in the chip not in bios) with this bios
 

·
Noctua Fanboy
Joined
·
1,095 Posts
Someone will have to take this bios out of my cold dead hands before i give it up lol.

This is crazy good work and i hope everyone gets AGESA 1.0.0.5 i knew something major changed lower latency on memory and higher ST turbo that constantly hits its rated turbo or exceeds it and i have a average chip here i need 1.26V or so for 4.1Ghz to be stable. Also most of my other cores are up 25mhz or so too and i checked like 5 times and its effective clock as well

I don't even need the custom powerplan anymore i just use balanced and it works perfectly on 1909.

Edit haha been calling you Veil i have no idea why sorry Veii!
 

·
Overclock the World
Joined
·
2,995 Posts
Someone will have to take this bios out of my cold dead hands before i give it up lol.

This is crazy good work and i hope everyone gets AGESA 1.0.0.5 i knew something major changed lower latency on memory and higher ST turbo that constantly hits its rated turbo or exceeds it and i have a average chip here i need 1.26V or so for 4.1Ghz to be stable. Also most of my other cores are up 25mhz or so too and i checked like 5 times and its effective clock as well

I don't even need the custom powerplan anymore i just use balanced and it works perfectly on 1909.

Edit haha been calling you Veil i have no idea why sorry Veii!
I don't mind, thought you where mocking me for fun ^^#
Although i wouldn't give up CPPC powerplans - windows theirs is still worse than 1usmus his or ones from computerbase/hardwareluxx

After you've updated, it would be fine if you could focus on the reports (when you have the time)
And check if our FCLK is finally unlocked or not
I have a suspicion that our old recommended voltage&resistance values are invalid now with this bios
We would need to know how Inner-CCX Latency looks like
Memory delay won't change by frequency, only L3 delay and bandwidth will be affected by frequency

If i remember correctly, we should have your old SiSandra report somewhere flying around
Just unsure if it was a full report or only shows complete inter-core latency without the ones in between cores
But yes, assist a bit with SMU readout submission, and maybe we can see if there is a version update or not
Haven't decompiled that part yet to check all changes ~ still wip

It is for sure interesting if you have procODT 28ohm as option or 30ohm
1.2v SOC is too high, but it scales up because VDDG is high on default
Could you try to find out what procODT runs right now ?
Issue is , installing ryzen master still will disable CPPC - that hasn't changed :rolleyes:
Soo afterwards, you'll need to disable and enable it from the bios to effectively "enable" it back again, after RM is wiped
 

·
Noctua Fanboy
Joined
·
1,095 Posts
Sure i'll run all those tests or most that i can. and edit this with the results! Edit running the 1usmus ryzen plan why not i guess haha don't have too.


Anyways i noticed just doing normal tasks in the PC is smoother as it is WAY more aggressive with the ST turbo now to check games and see if its better before it was setting at 4325mhz and dropping to 3600mhz in even the sims 3 a dual threaded game and it had rare spikes to 4350-4375.

Then i'll play some fallout 4 and see or GTA5 which seems to eat 6 threads or so. Edit Sims 3 i get 4325-4375mhz jumps more often and with GTA5 i get 4225-4325mhz with no drops basically the same as AGESA 1.0.0.3ABBA!
 

·
Noctua Fanboy
Joined
·
1,095 Posts
This is default Sisoftware results and RM also this is default meaning bios set to default and not a single thing changed but the boot order

https://docs.google.com/document/d/1ypeUiXfOZdt96zdDQlVZjME70SRCyxwNwIBtbok3NSQ/edit?usp=sharing

FCLK easily unlocked from memory all i did was change it to 1900mhz and boom memory still default, note how the VDDG or VDDP voltage stays the same ?


Here is the Sisoftware results for memory at default with FCLK at 1900mhz

https://docs.google.com/document/d/1XIldbSN5uVFnLPTgmt5eCWWOeBS9jhmda8-p-5nRKps/edit?usp=sharing


Check out that reduction in latency from moving to default FCLK to 1900mhz 100ns to 75ns! Cores within the CCX aren't effected by it at all! For sure we all can't wait for Zen 3!!!

For fun why not include memory at 3800mhz and the fabric at default notice how VDDP+VDDG+SOC jumps WAY up after i set it this way even with the FCLK at default!
https://docs.google.com/document/d/1cRwUAeLvAMLzmhxnlfM6uNbpQ2Xee8y3FuiSJBeqvPc/edit?usp=sharing

This is interesting to me there is more gains here from moving from 1900FCLK then moving from 2133-3800mhz memory in terms of intercore latency. Also i tested if 1933mhz worked on FCLK and nope a no go i tested this since when i first got my CPU anything over 1800mhz didn't work and didn't even boot.

Also to let others know your OC profiles work perfectly so you don't have to manually do the timings or anything i even manually checked the CBS settings and they stayed the same too with the update!
 

Attachments

  • Rep+
Reactions: Veii

·
Overclock the World
Joined
·
2,995 Posts
This is default Sisoftware results and RM also this is default meaning bios set to default and not a single thing changed but the boot order
https://docs.google.com/document/d/1ypeUiXfOZdt96zdDQlVZjME70SRCyxwNwIBtbok3NSQ/edit?usp=sharing

FCLK easily unlocked from memory all i did was change it to 1900mhz and boom memory still default, note how the VDDG or VDDP voltage stays the same ?

Here is the Sisoftware results for memory at default with FCLK at 1900mhz

https://docs.google.com/document/d/1...it?usp=sharing

Check out that reduction in latency from moving to default FCLK to 1900mhz 100ns to 75ns! Cores within the CCX aren't effected by it at all! For sure we all can't wait for Zen 3!!!

For fun why not include memory at 3800mhz and the fabric at default notice how VDDP+VDDG+SOC jumps WAY up after i set it this way even with the FCLK at default!
https://docs.google.com/document/d/1...it?usp=sharing
Wonderful, thank you for this one :clock:
1.025v vSOC seems to be used for anything under <2800MT/s
ProcODT prediction is still a mess, no difference there - good to know :thumb:
Let's see what the bios allows you to select

Also yes, the differences are huge
consistent 72ns across each CCX and consistent 27ns access time between each core inside CCX
I wait for one more SiSandra report from you with your 1900:1900 MCLK-FLCK setup on auto voltages
^ EDIT:

Nevermind :D
"Artificial Limits" still exist where default MCLK behave better,
let's call it a bug - because 1900 FCLK behaves significantly better, 30% better and consistent across core to between the furthest CCX (this should have a very strong effect on any game, pushing 0.1% and 1% FPS)
At least that part is "fixed" finally
Hmm, i wonder if something internally throttles down because it predicts far to high voltage :thinking:
~ and like we know, high voltage is bad on high frequency, high voltage alone makes no issues

Also remember this post :D
Starting Preset:
  • ProcODT 28-36.9Ω SR / 36.9-53.3Ω DR
  • CAD_BUS 24-20-20-24 / CadBusClkDrvStrengh (first value) can be pushed to 30-60Ω - DR 60-120Ω
  • CLD0_VDDP 900mV
  • CLD0_VDDG 950mV
  • VDDG CCD 950mV
  • VDDG IOD 950mV
Seems to align perfectly, prediction is still a bit messed up
You can run 28ohm with this low VDDG voltage on 1867FLCK without issues, like Yuri showcased on sTR4
I wonder what causes this spikes :thinking:
> Can you try to ignore voltages, let it be predicted and just use your RTT, CAD_BUS values ?
Starting with 34.4ohm going down to 28ohm procODT
Two results should happen,
- either no post because VDDG is too high predicted
- Bios is finally inteligent enough and drops VDDG to 950mV, while VDDP has to stay at 900mV
* vSOC will adopt to VDDG and VDDP predefined ranges / too high procODT limits maxOC on this old IMC

At the end, might you attach a visual (detailed) picture please with the 3 results selected - the graph
deselect online results just yours between this 3 tests ~ and writing down what color is which result (one pic)
i think our artificial limiter is gone, because it consistently scales up, need to see a full report on 1:1 mode only to doublecheck if 27ns change away or that's fixed for now by design (BLCK should push that lower)

Very interesting to note is
Bios is intelligent enough to disable Spread Spectrum on 1900 FCLK
and as we can see VDDG is not needed for higher FCLK, nor is VDDP pushing needed till 1900FCLK
only messes up signal integrity pushing too much voltage here
~ very helpful results for 3rd gen users, thank you ! :thumb:
 

·
Noctua Fanboy
Joined
·
1,095 Posts

·
Overclock the World
Joined
·
2,995 Posts
1900FCLK+3800mhz memory

https://docs.google.com/document/d/1P2pg9vZQLs5GwR4ibgbLrPb_09Hf3bTEMdvNu4qbw3U/edit?usp=sharing

For sure they did some kind of magic here as my intercore latency is down 2-3ns and communication within the same CCX is always 27ns now before it would jump to 28-29 at times.

Also again i think this is why latency on memory is slightly down too.
46.9ns Inter-Core latency, 66ns Inter-CCX latency, 27.5ns Inner-CCX Latency O _ O
This result might be the best that exists so far :applaud:
Let's later push memory a bit so you actually get global #1 ~ if you aren't already :p

Under what procODT was this ?
https://docs.google.com/document/d/1P2pg9vZQLs5GwR4ibgbLrPb_09Hf3bTEMdvNu4qbw3U/edit
compared to:
https://docs.google.com/document/d/1cRwUAeLvAMLzmhxnlfM6uNbpQ2Xee8y3FuiSJBeqvPc/edit
It rly looks like one of the limits is gone :thinking:
Was this 1900MCLK without fabric 1:1 ?
Or did it actually throttle down because of high voltage, resulting in Bad Inter-CCX Latency
 

·
Noctua Fanboy
Joined
·
1,095 Posts
OMG HAHA these are the default timings and settings for 3800mhz for the LOVE OF JESUS people please never ever keep it at these settings i gurentee these are worse then stock 2133/15 cas timings lets do the math here

3800/28=135.7
2133/15=142.2

Oh and it turns on geardownmode haha *** so its way worse then 2133mhz/15

Also as one can tell auto settings for all the memory and timings and training everything default. Now i'll bump that down to 3600 and 3200 and see

Just for laughs i'll re-run Sisoftware and also test memory latency at my optimized settings the test in ryzen calc completes in 107 seconds.

Link for the Sisoftware results for the horrid default 3800/1900FCLK

https://docs.google.com/document/d/12fnvc1X_mID_J_QIT1-WyeVXR9ei9Fe2G7jeMEp28gE/edit?usp=sharing

Ryzen calc results surprising not the worst latency results i seen on Zen 2 with some terrible memory setups

Also the intercore latency doesn't change at all with these horrible timings i was expecting something like 20-22-22-46 or something as the default. These are Samsung B-die kits too

As for the ProcODT Settings 28.2 is the lowest setting i can pick at 3800/1900 nothing lower is available as one can see it boots fine and still runs at 1.1V VDDP+VDDG and 1.2V SOC

3600/1800
3200/1600

auto settings down below it keeps the same voltage as the 3800/1900 settings for VDDP+VDDG but 3200 drops the SOC down to 1.1V
 

Attachments

7341 - 7360 of 8974 Posts
Top