Overclock.net banner
4181 - 4200 of 13158 Posts

·
Registered
Joined
·
2,328 Posts
Just wanted to give an little info on our PE level 3 and 4 as there seems to be some confusion on what it is doing. While yes it does increase the duration of how long the boost can hold...the biggest change takes place in the CPU MSR registers. Whatever The_Stilt was able to do does change how the CPU on die caches work. Somehow it speeds the L1, L2, and L3 cache up in terms of reducing latency.

The first screen shot is with PE Level 3 and the second screenshot is PE Level 4 Notice all latency improve some by good margin. However the boost level is the same with either PE level.
Aida64 latencies improve with CPU speed. I don't think it matters if it's PE 3 or PE4, but the actual CPU speed. Which in your case is increasing as you switch from PE 3 to PE 4. If you adjust your multiplier while your booted with PE 4 to match your PE 3 you will most likely find it's the same latencies. I've played with it quite a bit.
 

·
Registered
Joined
·
93 Posts
I can't get BCLK oc stable above 101, Bios version is 702, cpu is 2700x, ram is stable 2h GSAT, temp are ofcourse in check (Noctua d15, plenty of airflow) i am using 0.03 voltage offset with PE level 2.
Do i have the worst 2700x on the planet or what?
OC will get unstable randomly.
Do you have any SATA drive plugged into port 5/6 ? Then you won't get above 101.
It's a weird thing with this board and the way SATA port 5/6 is connected.
Keep port 5/6 free and you can go to ~103.4 most people can't go above 104.
 

·
Registered
Joined
·
1,262 Posts
No... i have a samsung 850 ssd in sata port 1 and a 3tb HGST 7200 ultrastar in port 2 (the first from bottom to top, bios says port 1 and 2) . I tried 804 bios but nothing changes... i see many getting 103 stable what is wrong with my system?
 

·
Meddling user
Joined
·
7,423 Posts
Aida64 latencies improve with CPU speed. I don't think it matters if it's PE 3 or PE4, but the actual CPU speed. Which in your case is increasing as you switch from PE 3 to PE 4. If you adjust your multiplier while your booted with PE 4 to match your PE 3 you will most likely find it's the same latencies. I've played with it quite a bit.

@mtrai

Above is true.

The core, L1 and L2 cache speed is permanently linked together as usual, however unlike with the previous designs the L3 cache now operates at core speed as well (i.e. full speed).
See heading The frequency relations of the CCX, here.
 

·
Registered
Joined
·
215 Posts
No... i have a samsung 850 ssd in sata port 1 and a 3tb HGST 7200 ultrastar in port 2 (the first from bottom to top, bios says port 1 and 2) . I tried 804 bios but nothing changes... i see many getting 103 stable what is wrong with my system?
yeap i have.

1 m.2
2ssd
2 normal disk 7200rpm

ch7 hero
2700x

and i can get 103.4 stable but i need to give it some more voltage of about eg: +0.250 or +0.0315
 

·
Computer Enthusiast
Joined
·
2,924 Posts

·
Stuck in the past
Joined
·
1,143 Posts

·
Computer Enthusiast
Joined
·
2,924 Posts
I've managed to get back to 3600MTs - but 'only' at 4.1GHz with CPU getting up to 74 degrees (P95). It used to be 4.2GHz and about 71 degrees max. Tight settings - 14 13 13 13 22 36 and tRFC at 280 currently, which is the setting i'm currently working on - I think I can get it tighter.

Image is taken after running RamTest, P95 35min run and IBT AVX Very High. Turns out I had to reduce my SoC voltage a bit to really get it reliable...

For those that don't know me, I do usually do much more thorough tests once I'm dialled in - RamTest to 4000%, P95 for over an hour etc. - when my system needs to be proper 100% stable. Often I'm just happy with my system and have no problems without feeling the need for extensive testing, and I don't really do all night tests!
 

Attachments

·
Meddling user
Joined
·
7,423 Posts
Great share gupsterg - very useful :)
NP :) , the information The Stilt pumps out amazes me TBH and not just on Ryzen.

Good stuff...it is hard to believe each day I still learn new or more in depth stuff with Ryzen CPUs.
NP :thumb:.

I've managed to get back to 3600MTs - but 'only' at 4.1GHz with CPU getting up to 74 degrees (P95). It used to be 4.2GHz and about 71 degrees max. Tight settings - 14 13 13 13 22 36 and tRFC at 280 currently, which is the setting i'm currently working on - I think I can get it tighter.

Image is taken after running RamTest, P95 35min run and IBT AVX Very High. Turns out I had to reduce my SoC voltage a bit to really get it reliable...

For those that don't know me, I do usually do much more thorough tests once I'm dialled in - RamTest to 4000%, P95 for over an hour etc. - when my system needs to be proper 100% stable. Often I'm just happy with my system and have no problems without feeling the need for extensive testing, and I don't really do all night tests!
That's damn nice result in my book. Perhaps temp difference due to room ambient variance between tests?
 

·
Registered
Joined
·
1,262 Posts
yeap i have.

1 m.2
2ssd
2 normal disk 7200rpm

ch7 hero
2700x

and i can get 103.4 stable but i need to give it some more voltage of about eg: +0.250 or +0.0315
Check page 1-2 on the manual.. i've used sata ports 1-2 the ones on bottom... why i cant get more than 102?
If it can help im using ch7 wifi with Vega 64 gpu... i also have many devices plugged into usb ports too one of them is a DAC with 2 jbl305sr plugged in.
 

·
Computer Enthusiast
Joined
·
2,924 Posts
NP :) , the information The Stilt pumps out amazes me TBH and not just on Ryzen.

That's damn nice result in my book. Perhaps temp difference due to room ambient variance between tests?
Not sure - I do need to give the CPU more than I used to to achieve same OC. Don't seem to be able to do 4.2GHz AND have ram at 3600MTs on this bios, so am just going to get tFRC as low as possible.

Temps have been warmer here last few days - it's bound to have some effect I guess...
 

·
Premium Member
Joined
·
2,759 Posts
Ryzen Timing Checker 1.05

https://www.techpowerup.com/download/ryzen-timing-checker/

- Added an alternative code path required by PinnaclePI 1.0.0.6 AGESA.
- Added APOB NVC parsing support for Threadripper with Firmware TPM (fTPM) active.
- Disabled APOB NVC parsing for MCM4 Threadripper SKUs (2970WX & 2990WX) due to the piggyback structure of the CPU.

I'm pulling the plug after this version.
So no support for upcoming hardware, or even bug fixes.

Due to the fact that AMD is still unable (or unwilling) to provide a proper hardware access method to acquire the information RTC is displaying, maintaining the app has become a burden.
Since RTC has to rely on unofficial (and unorthodox) methods to acquire the displayed information, its functionality can be broken by even a single change AMD makes to the AGESA stack.
Trying to constantly reverse engineer the changes made by AMD, while maintaining the support for older software stack versions at the same time isn't very feasible at all.
 

·
Registered
Joined
·
2,328 Posts
Ryzen Timing Checker 1.05

https://www.techpowerup.com/download/ryzen-timing-checker/

- Added an alternative code path required by PinnaclePI 1.0.0.6 AGESA.
- Added APOB NVC parsing support for Threadripper with Firmware TPM (fTPM) active.
- Disabled APOB NVC parsing for MCM4 Threadripper SKUs (2970WX & 2990WX) due to the piggyback structure of the CPU.

I'm pulling the plug after this version.
So no support for upcoming hardware, or even bug fixes.

Due to the fact that AMD is still unable (or unwilling) to provide a proper hardware access method to acquire the information RTC is displaying, maintaining the app has become a burden.
Since RTC has to rely on unofficial (and unorthodox) methods to acquire the displayed information, its functionality can be broken by even a single change AMD makes to the AGESA stack.
Trying to constantly reverse engineer the changes made by AMD, while maintaining the support for older software stack versions at the same time isn't very feasible at all.

I can understand completely and I appreciate you creating this lastest release.
 

·
Registered
Joined
·
2,328 Posts
I've managed to get back to 3600MTs - but 'only' at 4.1GHz with CPU getting up to 74 degrees (P95). It used to be 4.2GHz and about 71 degrees max. Tight settings - 14 13 13 13 22 36 and tRFC at 280 currently, which is the setting i'm currently working on - I think I can get it tighter.

Image is taken after running RamTest, P95 35min run and IBT AVX Very High. Turns out I had to reduce my SoC voltage a bit to really get it reliable...

For those that don't know me, I do usually do much more thorough tests once I'm dialled in - RamTest to 4000%, P95 for over an hour etc. - when my system needs to be proper 100% stable. Often I'm just happy with my system and have no problems without feeling the need for extensive testing, and I don't really do all night tests!
Daaaaaamn. Alright let's see a RTC screen now too. I can't even get 14-14-14-14 stable at 3600.
 

·
Overclocker
Joined
·
7,038 Posts
Ryzen Timing Checker 1.05

https://www.techpowerup.com/download/ryzen-timing-checker/

- Added an alternative code path required by PinnaclePI 1.0.0.6 AGESA.
- Added APOB NVC parsing support for Threadripper with Firmware TPM (fTPM) active.
- Disabled APOB NVC parsing for MCM4 Threadripper SKUs (2970WX & 2990WX) due to the piggyback structure of the CPU.

I'm pulling the plug after this version.
So no support for upcoming hardware, or even bug fixes.

Due to the fact that AMD is still unable (or unwilling) to provide a proper hardware access method to acquire the information RTC is displaying, maintaining the app has become a burden.
Since RTC has to rely on unofficial (and unorthodox) methods to acquire the displayed information, its functionality can be broken by even a single change AMD makes to the AGESA stack.
Trying to constantly reverse engineer the changes made by AMD, while maintaining the support for older software stack versions at the same time isn't very feasible at all.
Thank you for your support and effort you put in this tool. Its a shame AMD goes this route, is it different for Intel or do they also provide no access to their hardware?

Does it also differ from board to board or is it strictly AMD?
 

·
Registered
Joined
·
1,262 Posts
Ryzen Timing Checker 1.05

https://www.techpowerup.com/download/ryzen-timing-checker/

- Added an alternative code path required by PinnaclePI 1.0.0.6 AGESA.
- Added APOB NVC parsing support for Threadripper with Firmware TPM (fTPM) active.
- Disabled APOB NVC parsing for MCM4 Threadripper SKUs (2970WX & 2990WX) due to the piggyback structure of the CPU.

I'm pulling the plug after this version.
So no support for upcoming hardware, or even bug fixes.

Due to the fact that AMD is still unable (or unwilling) to provide a proper hardware access method to acquire the information RTC is displaying, maintaining the app has become a burden.
Since RTC has to rely on unofficial (and unorthodox) methods to acquire the displayed information, its functionality can be broken by even a single change AMD makes to the AGESA stack.
Trying to constantly reverse engineer the changes made by AMD, while maintaining the support for older software stack versions at the same time isn't very feasible at all.
Thanks for being the most useful guy to this community.
Might i dare to ask you if you have any clue why i can't OC bclk above 101 on my CH7 while others clearly can? This is driving me mad
 

·
Premium Member
Joined
·
2,759 Posts
Thank you for your support and effort you put in this tool. Its a shame AMD goes this route, is it different for Intel or do they also provide no access to their hardware?

Does it also differ from board to board or is it strictly AMD?
The issue exists mostly due to the security features.
It seems that the security was an afterthough, which was added only after everything else was completed.
The registers which could and normally would be used to read the parameters do exists, but they belong to a region which is "secured" by the platform security processor (PSP).
Securing a certain range means that it cannot be read or written. AMD could have easily allowed read only access to this region, but for some reason they decided not to.

Only the most basic timing and frequency information can be read on this platform. Everything else (such as ProcODT, etc) can be only read from a dynamic PSP NV copy, intended for debugging.
None of this is documented and because of that even the bioses cannot read them.

The worst memory related issue on this platform is that there is no way to tune the memory controller parameters.
They are hard coded into the memory controller firmware (PMU) and cannot be changed by anyone but AMD. This prevents the manufacturers from optimizing the parameters specifically for their designs.

For example on Z390 motherboards, on which I've worked on for the last couple weeks there are more than a hundred adjustable parameters. Intel provides the base line (MRC code) and the manufacturers can
then tune them further to better suit their designs. As an example, a specific dual rank B-die memory kit was giving me trouble on a Z390 motherboard. It was able to post at 3200MHz (which is very low for the platform) and higher
frequencies resulted in failed memory training and failures to post. About three hours later the very same kit was running at 4000MHz, with tighter timings than originally. Obviously the memory controllers on Intel and AMD CPUs have a very little in common, but
it is hard to say if the issues on Ryzen memory controllers are due to the memory controller IP itself being weak / bad or simply due to the fact that there are no adjustments available.
 

·
Registered
Joined
·
58 Posts
Ryzen Timing Checker 1.05

https://www.techpowerup.com/download/ryzen-timing-checker/

- Added an alternative code path required by PinnaclePI 1.0.0.6 AGESA.
- Added APOB NVC parsing support for Threadripper with Firmware TPM (fTPM) active.
- Disabled APOB NVC parsing for MCM4 Threadripper SKUs (2970WX & 2990WX) due to the piggyback structure of the CPU.

I'm pulling the plug after this version.
So no support for upcoming hardware, or even bug fixes.

Due to the fact that AMD is still unable (or unwilling) to provide a proper hardware access method to acquire the information RTC is displaying, maintaining the app has become a burden.
Since RTC has to rely on unofficial (and unorthodox) methods to acquire the displayed information, its functionality can be broken by even a single change AMD makes to the AGESA stack.
Trying to constantly reverse engineer the changes made by AMD, while maintaining the support for older software stack versions at the same time isn't very feasible at all.
I know that you just said that this is the last version, but do you have any idea why I keep getting "InitializeOIs failed." error? It used to work but not anymore, not even the latest version.
 

·
Premium Member
Joined
·
2,759 Posts
I know that you just said that this is the last version, but do you have any idea why I keep getting "InitializeOIs failed." error? It used to work but not anymore, not even the latest version.
It seems to be Windows related.
No know cures besides re-installing Windows, at least to me.
 

·
Registered
Joined
·
2,328 Posts
I know that you just said that this is the last version, but do you have any idea why I keep getting "InitializeOIs failed." error? It used to work but not anymore, not even the latest version.
This eventually cured itself for me. I can't recall if it was related to the subsystem id bug in the 1101 BIOS, but I believe it fixed itself after reverting back to BIOS 1001 and a few attempts to run RTC.
 
4181 - 4200 of 13158 Posts
Top