AMD Ryzen Master / Question about Best cores for PBO curve

Elric82

Limp Gawd
Joined
Jul 15, 2017
Messages
164
Hi

I have a question For the OC curve optimizer,


Hi,Which software should I trust for the prefered core and OC optimizer? Seems the informations are quite different. My bet is Ryzen master but just in case...

is it right :

In bios (0 to 11)
1 & 8 best
5 & 11 ""second best"
Is that correct?


https://www.casimages.com/i/210628105750159432.jpg.html

https://www.casimages.com/i/210628105750295253.jpg.html



My settings are now as this

Per ccx 47 /46 @1.26
Dynamic oc switcher 90amp
PBO limits PPT 165W TDC 120A EDC 150A
Core boost override : +100mz
Pbo curve : -15 on best core and second best / -30 on the rest
+140% cpu capability
LLC3
Power phase : phase response.


Thanks

My CB R23 score are
23400 points multi
1600 single
Cpuz : 10080 points / 680
77c max in CB

around 60c in games
 
Yes the star denotes the best core on each ccx and the white circle denotes the second best on each ccx. Also I’d put load line calibration back to stock and raise your manual oc voltage to compensate for vdroop without llc because llc also affects pbo (I don’t think pbo compensates for llc by lowering voltage but could be wrong). 1.5v top voltage bin for pbo plus llc 3 sounds like long term damage.

for me it was -12 on best and second best and -30 on the rest. I’m actually fiddling with general negative voltage offset to achieve better temps as prime 95 small ffts gets in the 80s with pbo. I know prime95 is unrealistic load but in my 2 decades of experience if it passes prime95 then it is a solid stable overclock. Plus more and more games and applications are starting to use avx so in the future some games might trigger a bsod if you don’t stress test your overclocks with an avx load.
 
Oh ok i didn't knew for llc.
Indeed I've seen it at around 1.5

I'm at 1.27 now with llc 3, how high should i go to compensate? Is llc auto same as no llc?

I've tried no llc but even at 1.3, CB R23 is crashing so...

That's what I have at idle witjout llc, is that fine,
 

Attachments

  • 11.jpg
    11.jpg
    49.7 KB · Views: 0
Last edited:
Oh ok i didn't knew for llc.
Indeed I've seen it at around 1.5

I'm at 1.27 now with llc 3, how high should i go to compensate? Is llc auto same as no llc?

I've tried no llc but even at 1.3, CB R23 is crashing so...

That's what I have at idle witjout llc, is that fine,
Unfortunately you will just have to trial and error. I’d try up to 1.375v MAX while monitoring temperature, if your cpu is in the 80’s for cb23 you have gone too far. Although I’d say you shouldn’t need more than 1.32-1.33v to make you cb23 stable. It’s okay to increase your voltage as you are trying to find the voltage that droops down to 1.27V under cinebench load. If you have an asus motherboard and have aisuite 3 installed then that is the best way to see the processors actual real voltage under cinebench load. Other monitoring programs only show you the CPU’s VID which is what the cpu asks for from your vrm but doesn’t show you the actual drooped voltage entering your cpu.

The good thing about llc3 is that it completely eliminates vdroop when manually overclocking so your VID is your real voltage but like I said earlier pbo is affected too so definitely don’t want that.

Also, I believe phase response increases cpu and vrm heat with little gain so perhaps optimized setting is better.

your idle voltage is about right for a 5900x. And yes llc auto means your processor controls llc which effectively means level 0 llc as the higher llc’s like llc 1,2,3 etc are all more aggresive than what the cpu will use under auto setting. May your tuning be fruitful!
 
Thanks! I have an asus board, dark hero but I've always hated ai suite, caused me lots of issues and was a pain to uninstall. That was a while ago though, has the soft improved?

What would be the max good temps in cb23 you think?
Thanks! I have an asus board, dark hero but I've always hated ai suite, caused me lots of issues and was a pain to uninstall. That was a while ago though, has the soft improved?


What would be the max good temps in cb23 you think?


Edit : i had to push to 1.31 to pass CB23, i got 23384 score multi and around 1620 single, 77c max,

CPU 120 / Dram 120 power phase optimized / llc auto / Vrm switch auto / +100 boost
 
Great! I would say 80C should be the limit as some avx loads are 10-15 degrees hotter than cinebench.
 
Seems i have weird issue when setting Power phase / cpu current capability above 100, I have some qcode 00 which is cpu non detected at boot or few min after.

Any idea what could be the issue? Voltage too lower maybe?

With digi+ on stock, no crash
 
I commend you guys on overclocking with RM and the PBO. I tried and failed miserably, yet I can overclock the snot out of it via old fashion bios methods. I gave up on the RM/PBO crap.
 
I don't use RM, I do it all via bios. RM is just to get info i don't tweak anything in it
 
Seems i have weird issue when setting Power phase / cpu current capability above 100, I have some qcode 00 which is cpu non detected at boot or few min after.

Any idea what could be the issue? Voltage too lower maybe?

With digi+ on stock, no crash
Hmm that is interesting, so I have a dark hero so when I did my oc I just set my manual oc voltage in the per core ratio menu, the same menu that has the option that allows you to dynamically switch between manual oc and pbo. I leave the main voltage on the extreme tweaked page as auto. Maybe that’s your problem? It would help being able to see screenshots of your bios showing what all you have changed.

also stock phase control is more than adequate as AMD has smarter phase control than intel processors.
 
3-2.jpg

2-2.jpg

4-1.jpg

In this one I've put extreme and not power phase response. If i set it back to auto, no issue but extrême or 140 causes the crash

5-1.jpg

6-1.jpg

I've changed it since this screen for +100. Pbo limits 160/120/170
 

Attachments

  • 6-1.jpg
    6-1.jpg
    77.3 KB · Views: 0
From what I can tell the only thing wrong is your core VID in the per core ratio menu. You might be right, not enough core voltage. I basically use the same settings as you but core VID at 1.32V, platform thermal throttle limit to 75 degrees, and power phase to optimized. Sorry I can’t be more help, unfortunately I’ve never run into this problem so have no experience diagnosing it
 
Why do you say the core vid is wrong? You mean not high enough?
 
Last edited:
Yeah 1.27V is too low from my experience with my 5900x. Like I said it takes 1.32V core VID to stabilize my 4.7 ghz for cinebench without llc
 
Oh I've put 1.31 now and i passed cinebench

My only issue were those crashes when putting the phase on extreme and cpu current capability at 140.

But i guess 100% is enough and optimized works all fine?
 
Haven't tried, i will do more tests.

For the rest of my settings you do the same with dynamic oc? What are your pbo limits?
 
Also, Are those voltage safe it was when gaming on biomutant, temps are ok and clocs are good but this 1.481 and average 1.38 and sometimes more worries me.

biomutant.jpg
 
Also, Are those voltage safe it was when gaming on biomutant, temps are ok and clocs are good but this 1.481 and average 1.38 and sometimes more worries me.

View attachment 371599
My pbo limits are “motherboard” limits. And yes if load line calibration is set to auto and your primary cpu voltage is also set to auto then when your cpu is in PBO mode those voltages are perfectly fine. Your cpu knows what it is doing.
 
I've let auto limits for pbo, is that fine?

I was afraid motherboard limits will make it run too hot, is it the case?

Also for the boost you have +100?
 
Last edited:
Yeah that’s fine, idk what the settings auto uses for limits but it should be less than motherboard. Idk if using motherboard limits is too hot, it all depends on your cooling setup. For me it’s fine. If you want to try it out it’s okay as your cpu will downclock if it actually gets hot enough. I just do +50 MHz because I like seeing my cpu at 5.0 ghz but it’s fine to increase it more.
 
I'll try motherboard and run benches


For the scores right now i have

23500 CB 23 multi / 1620 single
CPU z 10130 / 684
Geekbench : 15300 /1773
 
I saw the same issue when setting CPU Power Phase Control to extreme, I would get random crashes with Q-Code 00. Switch back to auto and everything is rock solid, no other changes were necessary. Maybe there is something buggy going on with the extreme setting.
 
Good info here, nothing to do with docs, just how to fine tune PBO.






Part 5 contains the benchmarks.



Looks like curve optimizer and PBO offered the best results.
 
I saw the same issue when setting CPU Power Phase Control to extreme, I would get random crashes with Q-Code 00. Switch back to auto and everything is rock solid, no other changes were necessary. Maybe there is something buggy going on with the extreme setting.

Every ASUS board I've owned in the last 6 years overclocks better on Optimized than Extreme. I think Extreme is better for LN2, but worse for air/water.
 
I saw the same issue when setting CPU Power Phase Control to extreme, I would get random crashes with Q-Code 00. Switch back to auto and everything is rock solid, no other changes were necessary. Maybe there is something buggy going on with the extreme setting.
What were your other settings to see what could be the wrong combination? Core voltage, vid, per ccx, curve, pbo...?


Is it better to use optimized or auto?
 
Don't increase the base voltage, that takes away the whole point of curve optimizer. Leave it stock, set PBO to motherboard, then drop your 2 (or 4) best curves by X and the remaining cores by Y, where Y>X, and play around with it, benchmarking in between each change to ensure you aren't murdering your performance. That's the "AMD way" to do it, where you maintain the best possible lightly-threaded performance while not compromising on multicore perf.
 
Don't increase the base voltage, that takes away the whole point of curve optimizer. Leave it stock, set PBO to motherboard, then drop your 2 (or 4) best curves by X and the remaining cores by Y, where Y>X, and play around with it, benchmarking in between each change to ensure you aren't murdering your performance. That's the "AMD way" to do it, where you maintain the best possible lightly-threaded performance while not compromising on multicore perf.
I haven't touched the base voltage, only core vid per ccx because i have a dark hero board. I've posted all my settings above :)
 
What were your other settings to see what could be the wrong combination?
PBO on motherboard limits, CO at -25 for most cores, two cores at -15, Fmax enhancer off, and DRAM/FCLK at 1900. Everything else relevant on auto/default. I hadn't messed with RAM timings yet when the 00 codes started cropping up.
Is it better to use optimized or auto?
I honestly don't know, I haven't tried optimized yet (or really any tweaking at all) due to lack of time since then. Just recently got my weekends back though, so will be really diving into Ryzen tweaking soon.

On a side note, I didn't do the whole best cores = x, 2nd best cores = y etc in CO. What I did was arbitrarily started at -25 all cores then tested with Prime and used HWinfo to identify which cores would crash (you'll see a specific core clock drop to idle when a worker crashes out), then bumped those cores up 5 "CO units" at a time until they were stable. The two that needed a bump were 1 "best core" and 1 other random core. One of these days I intend to find the true max CO limits of each core individually, but yeah, that going to be a bit time consuming.
 
PBO on motherboard limits, CO at -25 for most cores, two cores at -15, Fmax enhancer off, and DRAM/FCLK at 1900. Everything else relevant on auto/default. I hadn't messed with RAM timings yet when the 00 codes started cropping up.

I honestly don't know, I haven't tried optimized yet (or really any tweaking at all) due to lack of time since then. Just recently got my weekends back though, so will be really diving into Ryzen tweaking soon.

On a side note, I didn't do the whole best cores = x, 2nd best cores = y etc in CO. What I did was arbitrarily started at -25 all cores then tested with Prime and used HWinfo to identify which cores would crash (you'll see a specific core clock drop to idle when a worker crashes out), then bumped those cores up 5 "CO units" at a time until they were stable. The two that needed a bump were 1 "best core" and 1 other random core. One of these days I intend to find the true max CO limits of each core individually, but yeah, that going to be a bit time consuming.
Oh ok so it seems it's just the extreme settings that messes thing up.


For me i've set up -15 on best 4 cores (as seen in Ryzen Master) and the rest at -30
I've checked in event manager and created a custom rule to see if cores are crashing which looked like this for instance.

I had no crashes for a while so I suppose it's stable enough but haven't tried more. I think that -15 on 4 cores and -30 on the rest is good enough !

1625463858147.png
 
ahh ok ill try some of these i used the settings in that vid from Der8auer when he talked about he dynamic oc switcher didnt touch on the settings found here.
 
I saw it but i wasn't sure about his settings, he must have awesome cooling, the Core VID he put was very high. Also the amps for the oc swtich i'm still not sure what is the good value
 
hmm well i dont mind fine tuning with the help of this community i havent seen temps above 68c under load CB max was 70 but i have a nice 420 aio setup from arctic.
 
always have this handy as its what i used lol check in settings @5:16 mark
 
Last edited:
Back
Top