• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Ryzen 3900X thread

Whilst I'm happy with the performance of my 3900x, I'm still confused with xfr/pb/pbo and their impact on performance so will welcome your thoughts.

Using Cinebench R20 as a benchmark

Stock = 6900-7000 / boost to 3950/4000mhz / temp at 72c (multiple runs)

Pbo off = 7100 /4050mhz/68c

Undervolt and pbo off(0.102) = 7350 /4175mh/65c

Undervolt and pbo on =7200 /4100mhz /63c.

I've tried under voting at different increment from 0.01 to 0.1 at 10mhz increments, anything less than 0.102 has a negative impact on scores and temps.

I'm confused as why it boost lower with pbo on? I've configured pbo to use motherboard settings, and I've also tried manual setting to max out the three settings.

In games and general usage, I will get 2-3 cores hitting 4.6 to 4.625 with the other cores hitting 4.575 on the first chiplet.

As I said, I'm happy with the performance, but confused with the route taken to get the performance.

I'll try :D

If we look at your results ;

Your .102 offset score looks too high, i would hazard a guess that ' clock stretching ' is occuring and skewing the readout figures.
7100/7200 is around ballpark for a stock cpu. The 'sweet spot' for undervolting seems to be around negative .06.

Your PBO off score of 7100 versus your PBO on score of 7200 is again what you would expect, a 100 point gain in CB20 - So that is fine.

Use HWinfo64 to monitor clock speeds, and look at the Effective Clock Speed, not the Core Clock.

Example, from my own testing;

Higher Boost clock speed ( 4.525 ) / Lower Effective Clock (ave 3737mhz) / 7000 CB20 Score
DX8USWc.png

Lower Boost clock speed ( 4.450 ) / Higher Effective Clock (ave 3845mhz) / 7100 CB20 Score
C3ZKWMj.png

Ryzen is a complicated beast, and i don't pretend to understand , but sometimes clock speeds can be deceiving.
 
Thanks for the reply, but PBO lowers my score by around 100 points consistently. Whether at stock voltage or with undervolt applied. The PBO off is at stock clocks vs PBO on with an undervolt., PBO with Stock volts is lower than both.

I will run some more test tonight, checking effective clocks also but PBO really doesn't make any sense to me. It lowers my peak clocks and cinebench scores, but also lowers temps a few degrees also as a result of lower clocks? It seems to be working in the complete opposite to how it should work.

I've tried with a 0.06 undervolt, and played games for a few hours and my peak clocks are always lower (4.50 vs 4.625) when compared to a 0.102 undervolt.

I'm using a Gigabyte X570 elite and a 360mm Thermaltake AIO.
 
Mine is faster with a - 0.1 offset as well, any more than this loses performance. I don't have any explanation, more just posting to let you know that someone else's chip is the same as yours!
Heehee, as I said this isn't a grumble, I'm genuinely impressed with this cpu. Just wish I understood why settings that supposedly add performance make things worse, and things such as a large undervolt make it fly.
 
Last edited:
Team Group 8Pack Edition 32GB (2x16GB) DDR4 PC4-28800C16 3600MHz Dual Channel Kit & Asus 570-f motherboard.

I'm trying to set the xmp to the profile it has stored, but the machine doesn't post, it hangs on the DRAM light on board.

If I try just setting the memory speed to 3600 it also won't post and ends up booting to windows with 1200!!

Any ideas how I can get around this please?

Thanks,
Adam
 
1 ) Make sure XMP is also upping your DRAM voltage to 1.35v.
2 ) if it is, and no post, try 1.38v
3 ) if that fails try 3200Mhz see if that posts


Thanks. Ok checked and it is 1.35, upped to 1.38 no post again.

Just tried manual mode and setting to 3200 and no post.

Confused.


When I first got the equipment yesterday it was posting at 3600mhz. Set manual mode and 3600mhz with no problems
Started playing around with some mild overclocks on ryzen master (3900 per core) and it stopped posting.
Uninstalled ryzen completely now.
 
Thanks. Ok checked and it is 1.35, upped to 1.38 no post again.

Just tried manual mode and setting to 3200 and no post.

Confused.


When I first got the equipment yesterday it was posting at 3600mhz. Set manual mode and 3600mhz with no problems
Started playing around with some mild overclocks on ryzen master (3900 per core) and it stopped posting.
Uninstalled ryzen completely now.

Flash your bios to the latest version that supports your CPU, do not use stored BIOS settings if those aren't wiped.
Set XMP and try again.
 
Intel be like.

tenor.gif
 
Thanks man!Sorted updated bios. Seems to have done the trick.

I've saves the working profile now so I can start playing around.

Welcome.
If you want to tinker use DRAM Calculator and head for 3800C16 and IF 1900
When you are all stable (run the DRAM Calc ram tests), try to go for 3800C16-15-14-28 (usually run at same power) then tinker with the following trying to get them as close to values tFaw 16 and tRFC 285. (between changes run the DRAM Calc tests).

When you are there, try to do a dash for 3800C14-14-14-28, just in case, you might be lucky :p

As for CPU overclocking, I would leave everything to Auto. Activate CPPC & CPPC Prefered Cores. Cool'n'Quiet (if your CPU has it) and try XFR.
Don't try for all core overclock etc if you play games. Is better to have 2 cores boosting to 4575-4625 and the rest operating between 4350-4500 than been stuck at 4300.
 
Welcome.
If you want to tinker use DRAM Calculator and head for 3800C16 and IF 1900
When you are all stable (run the DRAM Calc ram tests), try to go for 3800C16-15-14-28 (usually run at same power) then tinker with the following trying to get them as close to values tFaw 16 and tRFC 285. (between changes run the DRAM Calc tests).

When you are there, try to do a dash for 3800C14-14-14-28, just in case, you might be lucky :p

As for CPU overclocking, I would leave everything to Auto. Activate CPPC & CPPC Prefered Cores. Cool'n'Quiet (if your CPU has it) and try XFR.
Don't try for all core overclock etc if you play games. Is better to have 2 cores boosting to 4575-4625 and the rest operating between 4350-4500 than been stuck at 4300.
Thank you for this panos.
Much appreciated already feel a little out of my depth here coming from an intel architecture.

Also do you use ryzen master or bios?

Also I'm assuming this is the boost mode kicking in:
looking at my ai suite and the cores seem to be hovering around 3800 to 4400mhz and vcore going as high as 1.45!

Temps are great idle around 33c, under load no higher than 62c.
But I'm a little alarmed at that vcore swinging so high?
 
Last edited:
Thank you for this panos.
Much appreciated already feel a little out of my depth here coming from an intel architecture.

Also do you use ryzen master or bios?

Also I'm assuming this is the boost mode kicking in:
looking at my ai suite and the cores seem to be hovering around 3800 to 4400mhz and vcore going as high as 1.45!

Temps are great idle around 33c, under load no higher than 62c.
But I'm a little alarmed at that vcore swinging so high?

Keep away from Ryzen Master if you do not want to do some tedious per core overclocking. Other than that, is better not to have it installed on your system at all.

If you try to run a benchmark you will see your CPU won't go over 1.25-1.26v (temp and load depending) at 100% core loads because electricity current is high.
The 1.45-1.5v is just the reported voltage of the milisecond in time the CPU is ordered to boost from idle to max speed and occurs at very low current so is very safe and within spec, is also per core. It will never try to push 1.45-1.5v on all cores trying to boost all of them, only one at the time. Otherwise it would settle for the safe spec of 1.27 and lower clocks. You shouldn't worry about it.

3900X is a CPU over the last 10 months has amazed me how much thought and AI trickery AMD has incorporated to make it work so clever.
It saddens me when I see people (even youtubers like Jay2Cents) preaching "power too high, AMD crazy, manually set to 1.35v" without have done a single research or observed the CPU default behaviour. AMD is not crazy to burn their CPUs let alone want it's customers to have degrading silicon issues. (that occurs when you go over 1.28v with high electricity current that would never happen if not manually do it).
Thats why XFR & PBO are also so great to work with. You raise the power limits and the speed of the CPU can change the clocks and the AI will decide whats safe at any time based on gazillion of variables including temps (per core), electricity current (per core), hundreds upon hundreds of variables.

FYI Install the latest chipset drivers from AMD again if you just flashed your bios.
 
@Serial45 https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/9.html

Pick up the memory test there. TM_5 & config, there is another config (v3), which lasts several hours and is the ultimate memory stability test.

I have 3 different config files and duplicate the program and use shortcuts.
Here it is (Replace text in MT.cfg in the program/bin directory):
Memory Test config file v0.02
Copyrights to the program belong to me.
Serj
testmem.tz.ru
[email protected]

[Main Section]
Config Name=Default
Config Author=1usmus_v3
Cores=0
Tests=16
Time (%)=100
Cycles=20
Language=0
Test Sequence=6,12,2,10,5,1,4,3,0,13,9,14,7,8,1,11,15

[Global Memory Setup]
Channels=2
Interleave Type=1
Single DIMM width, bits=64
Operation Block, byts=64
Testing Window Size (Mb)=880
Lock Memory Granularity (Mb)=16
Reserved Memory for Windows (Mb)=128
Capable=0x0
Debug Level=7

[Window Position]
WindowPosX=1379
WindowPosY=529

[Test0]
Enable=1
Time (%)=100
Function=RefreshStable
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=0
Test Block Size (Mb)=0

[Test1]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=1
Pattern Param0=0x1E5F
Pattern Param1=0x45357354
Parameter=0
Test Block Size (Mb)=16

[Test2]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0x14AAB7
Pattern Param1=0x6E72A941
Parameter=254
Test Block Size (Mb)=32

[Test3]
Enable=1
Time (%)=100
Function=MirrorMove
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=1
Test Block Size (Mb)=0

[Test4]
Enable=1
Time (%)=100
Function=MirrorMove128
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=510
Test Block Size (Mb)=0

[Test5]
Enable=1
Time (%)=100
Function=MirrorMove
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=4
Test Block Size (Mb)=0

[Test6]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0x5D0
Pattern Param1=0x143FBC767
Parameter=125
Test Block Size (Mb)=1

[Test7]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=0
Test Block Size (Mb)=2

[Test8]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0x153AA
Pattern Param1=0xDC7728C0
Parameter=358
Test Block Size (Mb)=0

[Test9]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=0
Test Block Size (Mb)=4

[Test10]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0x2305B
Pattern Param1=0x97893FB2
Parameter=477
Test Block Size (Mb)=8

[Test11]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0x98FB
Pattern Param1=0x552FE552F
Parameter=8568
Test Block Size (Mb)=16

[Test12]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0xC51C
Pattern Param1=0xC50552FE6
Parameter=787
Test Block Size (Mb)=32

[Test13]
Enable=1
Time (%)=100
Function=SimpleTest
DLL Name=bin\MT0.dll
Pattern Mode=2
Pattern Param0=0xB79D9
Pattern Param1=0x253B69D94
Parameter=8968
Test Block Size (Mb)=64

[Test14]
Enable=1
Time (%)=100
Function=MirrorMove
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=16384
Test Block Size (Mb)=0

[Test15]
Enable=1
Time (%)=100
Function=MirrorMove128
DLL Name=bin\MT0.dll
Pattern Mode=0
Pattern Param0=0x0
Pattern Param1=0x0
Parameter=2
Test Block Size (Mb)=0

As far as my own situation:

I've swapped out micron-e for b-die patriot viper steel 4400. I had the M2Z3600c18 @ 3800 (I posted before in this thread). I got the 4 dimms of b-die in and found a setting of tWR = 13 which my system liked and posted with. The temperature has gone up 8'c here in last 48 hours and my memory failed to post and gave errors @ 3733. I was using 1.44vDDR (1.46v reported) so I set 3666, 1.4vDDR and kept my tWR13 setting and now I run 299 tRFC because 299 is a multiple of 13.

I did have the 20 pass test done but it was tricky to get it to post sometimes. I feel like I need a fan on my RAM, and I'm out of PWM to molex fan adapters (to set pwm fan to 12v), I do have a spare silent fan. My priority was to have at least something stable and I even tried cl15 GDM disabled but got some really wobbly responces from the motherboard. I've not tried lowering tFAW which is 16. My SoC is overkill even for 3800MT/s as I have gameplay of it running 1069 & 1088mV (that's AMD's reading, gigabyte's is 1.116v).

As expected I've dropped some score in the membench: 196.96 > 203.69 on default.

https://imgur.com/a/TC1KA01 Last screenshot is the most recent/relevant.

This is my earlier post with my system specs:
https://imgur.com/a/2uxhgym
 
Last edited:
Keep away from Ryzen Master if you do not want to do some tedious per core overclocking. Other than that, is better not to have it installed on your system at all.

How come Panos? genuine question, just moved to a ryzen system, no plans to overclock, why not install ryzen master to keep an eye on things?
 
Back
Top Bottom