How to actually improve PBO performance?

kazablanka · Nov 10, 2019

VPII said:
Highest clocks I get is 4.175 basically 4.19ghz with my 100.4Mhz BCLK.

Can you try ptt 1000/ tdc 360 /edc 140 scalar x10 and llc auto and tell me the clocks with these settings please ,its seems that low llc doesn't efect all chips the same. Are you on 1.0.0.4B or abba?

VPII · Nov 11, 2019

kazablanka said:
Can you try ptt 1000/ tdc 360 /edc 140 scalar x10 and llc auto and tell me the clocks with these settings please ,its seems that low llc doesn't efect all chips the same. Are you on 1.0.0.4B or abba?

I'm on ABBA, the 1.0.0.4 gave me issues and lower clocks.

kazablanka · Nov 11, 2019

VPII said:
I'm on ABBA, the 1.0.0.4 gave me issues and lower clocks.

i have no issues with 1.0.0.4B but yes i have lower clocks too. Try something please ppt 1000/tdc 140A/edc 135A and scalar x10. When running cinebech look at hwinfo at clocks and at effective clocks(lates version required). See what score you have with this settings and with limits by motherboard and what clocks you have at clocks and effective clocks

agentesmith · Nov 11, 2019

I did some tests today to understand what's happening with PBO not reaching Max speed.
My config is: 3700x on Gigabyte x570 master with 1.0.0.4 bios liquid cooled PBO on auto +200 and power limits manually maxed on the values of my board.
To get freq values I used HWinfo tweaking polling rate to 100ms. The others are from ryzen master

So I started to play with Power Limits to get some quantitative results.
First of all I started trying different setting in PBO and XFR to identify wich one are applied or not and I found that XFR overrides PBO settings.
Then I played around with different settings from stock to manual maximum and in the end with maximum with EDC at 0.
In the beginning the vcore was normal -0.1 that took me to reach 4350mhz on the best core.
In every situation the maximum freq reached were the same unless I place EDC to 0. In this case I finally got 4375Mhz on best core.
Raising dyn Vcore to 0 made me to get 4400mhz so i raised it to +0.1 and I got 4425.
Then I started some qualitative test using cinebench in single and multi core.
And these are the results:

PPT PBO MAN 0 88 1200 1200
TDC PBO MAN 0 60 540 540
EDC PBO MAN 0 90 600 0
PPT XFR MAN 0 88 1200 1200
TDC XFR MAN 0 60 540 540
EDC XFR MAN 0 90 600 0
% PPT 100 10 12
% TDC 87 13 15
% EDC 83 16 100
CB20 MULTI 4390 4524 4700
CB20 SINGLE 493 493 499
FREQ MULTI 3850 4100 4100
FREQ SINGLE 4375 4375 4400

Lookin at the results it's clear that there is a problem with EDC value when changed from default. With defaults 90A i get capped in multithread bechmarks but overall goes faster than changing it to the maximum of my board.
What do you thing about that situation? I fell that my cpu can reach higher frequencies but something stop it. But I don't know where to look..
I hope that helps and I will keep doing tests tomorrow

VPII · Nov 11, 2019

kazablanka said:
i have no issues with 1.0.0.4B but yes i have lower clocks too. Try something please ppt 1000/tdc 140A/edc 135A and scalar x10. When running cinebech look at hwinfo at clocks and at effective clocks(lates version required). See what score you have with this settings and with limits by motherboard and what clocks you have at clocks and effective clocks

Hey @kazablanka I tried your settings and I found that where cpu speed what going 4.19ghz with 41.75 multi and 100.4 BCLK it now was mostly 4.216 with the 42 multi but would drop sometimes between 41.75 and here and there 41.5 multi. My CB score also went up but was lower at first until I changed EDC to 150 and TDC to 160. This is with a -0.1125 vcore offset. I tried lower but windows crash. I run Prime96 small FFT now to see what speed it sits at as before it was about 4.09 to 4.06ghz.

Okay just tested and in Prime95 it is barely 4Ghz while stress testing. Still not bad with vcore sitting at 1.15v or there about.

kazablanka · Nov 12, 2019

VPII said:
Hey @kazablanka I tried your settings and I found that where cpu speed what going 4.19ghz with 41.75 multi and 100.4 BCLK it now was mostly 4.216 with the 42 multi but would drop sometimes between 41.75 and here and there 41.5 multi. My CB score also went up but was lower at first until I changed EDC to 150 and TDC to 160. This is with a -0.1125 vcore offset. I tried lower but windows crash. I run Prime96 small FFT now to see what speed it sits at as before it was about 4.09 to 4.06ghz.

Okay just tested and in Prime95 it is barely 4Ghz while stress testing. Still not bad with vcore sitting at 1.15v or there about.

Thanks for the info!

kazablanka · Nov 12, 2019

Ihave

agentesmith said:
I did some tests today to understand what's happening with PBO not reaching Max speed.
My config is: 3700x on Gigabyte x570 master with 1.0.0.4 bios liquid cooled PBO on auto +200 and power limits manually maxed on the values of my board.
To get freq values I used HWinfo tweaking polling rate to 100ms. The others are from ryzen master

So I started to play with Power Limits to get some quantitative results.
First of all I started trying different setting in PBO and XFR to identify wich one are applied or not and I found that XFR overrides PBO settings.
Then I played around with different settings from stock to manual maximum and in the end with maximum with EDC at 0.
In the beginning the vcore was normal -0.1 that took me to reach 4350mhz on the best core.
In every situation the maximum freq reached were the same unless I place EDC to 0. In this case I finally got 4375Mhz on best core.
Raising dyn Vcore to 0 made me to get 4400mhz so i raised it to +0.1 and I got 4425.
Then I started some qualitative test using cinebench in single and multi core.
And these are the results:

PPT PBO MAN 0 88 1200 1200
TDC PBO MAN 0 60 540 540
EDC PBO MAN 0 90 600 0
PPT XFR MAN 0 88 1200 1200
TDC XFR MAN 0 60 540 540
EDC XFR MAN 0 90 600 0
% PPT 100 10 12
% TDC 87 13 15
% EDC 83 16 100
CB20 MULTI 4390 4524 4700
CB20 SINGLE 493 493 499
FREQ MULTI 3850 4100 4100
FREQ SINGLE 4375 4375 4400

Lookin at the results it's clear that there is a problem with EDC value when changed from default. With defaults 90A i get capped in multithread bechmarks but overall goes faster than changing it to the maximum of my board.
What do you thing about that situation? I fell that my cpu can reach higher frequencies but something stop it. But I don't know where to look..
I hope that helps and I will keep doing tests tomorrow

i have done the same testing with various cpus , i have 3700x too , set the pbo scalar x10 ,i think that if you set scalar x10 you will get better clocks

My settings are
ppt 1000
tdc 105
edc 90
scalar x10
+100 mhz
llc8 (lower)
and 1usmus power plan (it helped me to get 4450mhz at single core)

agentesmith · Nov 12, 2019

kazablanka said:
Ihave

i have done the same testing with various cpus , i have 3700x too , set the pbo scalar x10 ,i think that if you set scalar x10 you will get better clocks

My settings are
ppt 1000
tdc 105
edc 90
scalar x10
+100 mhz
llc8 (lower)
and 1usmus power plan (it helped me to get 4450mhz at single core)

Hi, i've already set scalar 10x and i've also tested single core bech setting manual affiinity in windows to only my best core and the results are the same. Not a single mhz more. The edc problem looks like a bug they have to solve because right now is useless.
Today I will do some more tests

VPII · Nov 17, 2019

I've always been a manual overclocker when it comes to my processor so I really did not bother with PBO on these Ryzen chips. Now seeing this thread a few weeks back I decided to try PBO with my Ryzen 9 3900X to see what it would actually do. Now at first I set everything to max to see what clocks I get and upon seeing the clocks for all core sitting around 4.16 to 4.19ghz using 100.4 BCLK I decided to do a negative off-set voltage having seen what the processor can do with a manual overclock. I decided at first to go with -0.075vcore and all worked well, so I decided to drop it a little further up to -0.1125 and all worked well. I noticed a nice drop in cpu temps when running CB20. But seen that I am unable to get a manual overclock to work running Prime95 small FFT as I usually drop 1, 2 or 3 cores while it is running, I decided to see what it would do.

But first a member here @kazablanka gave me some advice as in dropping TCD and EDC to around 140 to 150. Well so I did but ended up dropping it to 160Amps each and all core speed went up to 4.2ghz. So I tested with Prime95 Small FFT and it runs without a problem but CPU speed around 4.06ghz with the 100.4BCLK and it runs without an issue. Cpu temps do rise to about 81.5C after 10 to 15 minutes but it is keeping it there about without an issue with the vcore fed to the cpu being 1.2v.

wingman99 · Nov 17, 2019

VPII said:
I've always been a manual overclocker when it comes to my processor so I really did not bother with PBO on these Ryzen chips. Now seeing this thread a few weeks back I decided to try PBO with my Ryzen 9 3900X to see what it would actually do. Now at first I set everything to max to see what clocks I get and upon seeing the clocks for all core sitting around 4.16 to 4.19ghz using 100.4 BCLK I decided to do a negative off-set voltage having seen what the processor can do with a manual overclock. I decided at first to go with -0.075vcore and all worked well, so I decided to drop it a little further up to -0.1125 and all worked well. I noticed a nice drop in cpu temps when running CB20. But seen that I am unable to get a manual overclock to work running Prime95 small FFT as I usually drop 1, 2 or 3 cores while it is running, I decided to see what it would do.

But first a member here @kazablanka gave me some advice as in dropping TCD and EDC to around 140 to 150. Well so I did but ended up dropping it to 160Amps each and all core speed went up to 4.2ghz. So I tested with Prime95 Small FFT and it runs without a problem but CPU speed around 4.06ghz with the 100.4BCLK and it runs without an issue. Cpu temps do rise to about 81.5C after 10 to 15 minutes but it is keeping it there about without an issue with the vcore fed to the cpu being 1.2v.

After lowering the core voltage did the score reduce for CB20?

VPII · Nov 18, 2019

wingman99 said:
After lowering the core voltage did the score reduce for CB20?

Nope it actually improved, reason being lower temps under load. From what I've seen with these Zen 2 chips, when running stock or PBO there is minute changes in core speed under load and these changes are impacted by core temp.

rastaviper · Nov 18, 2019

wingman99 said:
After lowering the core voltage did the score reduce for CB20?

Why lowering the voltage would reduce the performance??
Less voltage->less heat->better clocks->better performance

Sent from my ONEPLUS A6003 using Tapatalk

wingman99 · Nov 18, 2019

rastaviper said:
Why lowering the voltage would reduce the performance??
Less voltage->less heat->better clocks->better performance

Sent from my ONEPLUS A6003 using Tapatalk

The reason in the video.

Satanello · Nov 21, 2019

kazablanka said:
Ihave

i have done the same testing with various cpus , i have 3700x too , set the pbo scalar x10 ,i think that if you set scalar x10 you will get better clocks

My settings are
ppt 1000
tdc 105
edc 90
scalar x10
+100 mhz
llc8 (lower)
and 1usmus power plan (it helped me to get 4450mhz at single core)

Hi, i tried with scalar x10 but the Vcore for ST and MT apps/bench is very high I think It's too much for everiday use.
If I set scalar x3 with LLC8 durign MT Cpuz bench I reach 1,35/1,36v :-\

How to actually improve PBO performance?

kazablanka

VPII

kazablanka

agentesmith

Attachments

VPII

kazablanka

kazablanka

Attachments

agentesmith

VPII

wingman99

VPII

rastaviper

wingman99

Satanello

Top Contributors this Month

Recommended Communities