pbo noob questions / overclocking

[Cramig88]

Little background, heres what I been up to:

 

Specs just in case anything is relevent:

-5950x

-Asus/Rog Dark Hero Viii - latest bios, i check weekly or more, just updated chipset drivers from amd/x570 area, all is well

-GPU's all have latest drivers

-Corsair vengeance pro argb, cmw32gx4m2d3600c18, yes latest firmware also.  4x dimms, 16g each, 18cl@3600, i think the timings are 18-15-15-22.  wild guess but i been in the bios enough lol.  the RAM was bought as a set, since there arent any 4 packs i had to get 2x 2 packs.  Yes, even checked serial numbers, ~90 is in 1A ~91 is in 2A, ~62 in 1B and ~63 in 2B (random numbers, regardless their paired by serial number)

-custom ekwb block made for the dark hero, it covers the cpu as well as the VRMs.   Cryonaut thermal paste

-strix 4090 OC, custom block also (top pci slot), used only with my samsung G9.  remaining GPUs are in the next slot down to the last one.

-xfx "speedster" AMD 6700xt, custom waterblock also. handles my vg32 something displays.  both are identical aside from brand and specs but the bezel is identical.  they're on the 6700 cause using super virtual resolution is easier to figure out.  Tried with nvidia control panel and gave up as setting a "custom resolution" proved to be a pain in the butt, even got to 60hz from 165(stock) and never worked at 2560x1440.  so just decided to let the AMD card handle that.  this card does not do any games, used to, but not since the 4090.

-nvidia quadro 4000 rtx which is set to do gpu and compute tasks, no monitors hooked up.  basically just use that when mining.  PC hardly ever gets shut off, so if shes gonna be up all night ill start mining.  No waterblock (but it needs one!!!!!!  I cant find one anywhere).  debating on removing the card, pulling everything down a slot and popping in my nvme RAID card with bitrification.  with 4 kingston fury's in there (fastest R\W and IOP drives i could find at the time), hey another 2tb on raid 1

-seasonic 80 bronze 1000w psu handling motherboard power, cpu and sata.

-"gamemax" 80 bronze 1050w psu handling the GPU's.  since im considering removing the quadro in place of an nvme card ill be able to remove this from the pile of crap.

-corsair hx1500i (take a guess how many watts, duh) 80 platinum psu currently not in use, but will replace both the seasonic and gamemax when i do pull out the quadro.  Got a hell of a deal on it! Roughly $250 for a ~$500+ psu!  brand new too!!

 

WATERCOOLING STUFF

-500ml corsair res/pump, DC5 naturally

-1L "fish tank" just for more water.  plus it looks cool with the UV led's under it as it sits on top of my case.

-2x 360mm radiators, one is slim other is 60mm.  big guy is mounted to the front pulling air in, narrow guy is on the backside of the case pulling air out.

-1x 420mm by 60mm radiator on the top also pulling out air. 

This guy has a seperate pump simply circulating the water from the "tank" into the rad and back again.  figured with all these rads having a cyclone in my largest radiator would be beneficial.  Since the rad had 2 extra 1/4 ports i figured hook it up to another pump.  kind of a seperate loop in that respect.

-1x 140mm (slim) on the rear exhaust, no fan as its at the end of the "cooling chain", passive cooling is all thats needed.

-the custom blocks as mentioned on mobo/4090/6700 all with cryonaut and thermal pads + extras.  i have tons in varying thicknesses, just followed the instructions and put more of of the same size across components pcb components that wouldnt have been covered otherwise.  figured if i had broken down a GPU to its bones, might as well.  Verified that the gpu/cpu on mobo IS making proper contact given the extra pads.  Anything I can do to distribute the heat across this huge block for better heat transfer.  Cant see any reason why it would hurt... do it right the first time

-a crapload of thermaltake something fans, all are the same except the 3 on the 420mm radiator.  best fans i could find at the time, think theyre 3000rpm.  idk but they arent rgb and they were bleeping expensive, plus had strong reviews across many sources.  woulda went noctua but that brown/tan color?  that would look terrible in my chassis

-3x of these are on the front side of the chassis (not inside case) pushing air in through the 360x60mm radiator.

-3x more are on the slim rad next to it facing the back side of the chassis where the wires are.  wires are nice and tidy so airflow is as good as its gonna get, pulling air from the inside or other radiator.

-3x as exhaust fans, again not in chassis, mounted "outside" on the top of the thick 420mm rad.

-chassis is a thermaltake view T71.  yeah its made for e-atx, well aware i have atx.  reasoning was to fit all this crap inside there (oh that was a challenge).  given its all glass, the noctua fans would have looked like sh**

 

^^^^^^yes i know it all seems overkill but these parts were acquired over time.  since i repair and build pc's, when it came to doing my own i just went thru my closet of stuff and popped it into the config.

-----------------------

 

BACK TO IT:

 

"stock overclock" and results:

So lately i have been dabbling with the PBO feature in BIOS.  Yes its the latest version, same with my chipset drivers and all.  Fans set to turbo via AI Suite app as its a Asus\ROG dark hero viii.

First I reset my bios.  Applied resizable BAR (not IV-SOV or whatever, read it hurts performance, not much but whatever), setup my ram via docp settings then did a TPU II auto-tune overclock.

I threw in cinebench r23, got 11 passes at 26,488 then single core 1,340 with only one pass.  Custom five minute test.  Feel like I have to say this, during these stress tests i let windows finish doing stuff, i can tell cause the clocks or heat go down.  this applies to ALL stress tests I have done.  Then obviously I have hwInfo running, taskman, cinebench and notepad journaling changes, temps and so forth.  Otherwise if im not typing notes im not using the mouse or doing anything.  i start the stress tests at 0-1% utilization.

Looking at my notes, 4.1ghz/75ish celcius on MC and SC stresses.  The cpu didnt fluctuate in terms of speed, it just ran through and did its job, cool.  Though since it did setup the multiplier, PBO isnt going to work.  Least I havent gotten it to yet. 

Multi core score: 26,488 @ 11 passes, temps in 75 area, 4.1ghz

Single Core: 1,340, same temps and speed, MP ratio 19.77x

 

Dabbling with PBO:

Last night I decide to play with PBO feature.  Saved bios settings to usb then back to defaults.  After 8+ hours of restarting into bios and stress testing i have came up with many LOWER scores and quite confused as to how things turned out.  This is my best setup as of yet

At this current setup im hitting roughly 5.1ghz - but not on all cores.  Yes I realize some are better than others as show in HWinfo.  Ive done my best in terms of the pbo scalars, those negative numbers are very confusing!  The scores on this setup (best manual OC scores ive gotten, and stable) are 24,288, unknown how many passes.  Reached around 90c and the cpu started to throttle, no surprise.  I think it got down to 4.2ghz and was still at ~90c.  whatever i guess, its the best score all day!  the single core was 1535, with an mp ratio of 15.82x.  whats that? why does it change with every configuration no matter how slight?  Should also mention the temps on Single Core were ~74c area and holding 5.0-5.2 steady.  Yes I stress tested it 12+ hours on corecycler using aida64?  prime 95?  whatever, u get the point.

Im not sure how cinebench works for single core stressing but looking at taskman usually id see 0-3 doing work while rest are idle.  if anybody can shed light on how R23 does its single core stuff that might be helpful.

Multi core score: 24,288 cpu hit 90c then throttled to 4.2ghz while maintaining 90c or a little less.

Single Core: 1535 MP ratio of 15.82x.  74c +/- 2 degrees holding 5.0-5.2ghz consistently.

 

Current Config:

then just recently out of curiosity i saved the pbo settings to a usb (never know!) and set to defaults.  All i did was set docp up, put 3600 as target ram speed (the docp standard says 3594 or something, prob irrelevent), then put the lclock?  memory clock?  whatever, at 1800 since its half of my ram.  the bclk is the 100.00 value, so not that.  I put in my stock RAM voltage (its on the box) to 1.35v then went to the digi+ phase tuning.  the phase tuning is setup from some youtube video, after watching many specifically with my cpu and mobo I came up with the most stable config.  My only knowledge regarding Digi+ tuning and what "this" does is from AI Suite.  That being said after seeing enough videos I just tossed caution into the wind and applied some high settings.

Now im "stock" in windows... well, nearly.  My CPU is displaying 5.0 using CoreTemp but thats only the highest clocked cpu out of 16.  Im at 65 idling in windows typing this.  the biggest thing im noticing is my max cpu is going from 5.0 to 3.6 to 4.8, just all over the bleeping place and thermals are under control.  mind you "performance boost" is on auto and multiplier is at auto.  I have noticed when using the multiplier it fixes my clock to whatever and it wont utilize PBO.  Why cant I use both?  I dont have a dynamic OC option as shown in many other videos, guessing cause they pulled it out the BIOS after various updates.  Still, if i could run at 4.0 or 3.8 while doing less intensive tasks then boost to 4.9ghz+ when duty calls, that would be the ideal situation.  especially since the water and components would have time to dissipate the heat.

Multi Core: 8 passes, 21,050 score.  Temps ~68 / 100% utilization / 3.3-3.4ghz.  did my pbo settings stay even after hitting the bios reset button?  if so PPT would explain the low temps

Single Core: 1 pass, 1,564 score MP ratio 13.46x.  Temps ~78-80 / 5% utilization / 4.7-4.9ghz. 

 

 

 

Ryzen Master hasnt worked since I started this:

its all stock aside from the above and stupid stuff like resize bar - figured thats a given.  OH also I did read somewhere that enabling hyper-v/SMT/virtualization lowers your scores.  I have tried this with the auto-tune and its true, but only by like 100 points - nothing serious.  Yes core isolation is off.  Much as I enjoy running VM's, ill deal with that part later.  Still Ryzen Master wont work!  Even did the registry thing, disableUnknownDrivers (whatever the d-word is) and set it to zero.  saved, restart.  uninstalled ryzen master, restart.  downloaded it again, installed, restart.  boot up windows and.... no dice.  What "drivers" are installed?  where can I find these in computer management\device manager tab??  as mentioned I updated the x570 chipset drivers to latest as of october (?) 2024.  but.... what drivers are potentially being installed by Ryzen Master?  removing them (as there is no clean uninstall) is supposedly a way to fix the program but, where are they? 

Should be mentioned as a windows "power user" I have disabled certain junk from booting, some windows services, a bunch of bloatware apps i dont need running constantly or booting on startup (armory crate for example).  Zero issues with games, .net is all updated, SDK framework and so forth, thats all happy. 

I would like to get Ryzen Master working so I have a starting point of PBO values to adjust instead of taking a wild guess then do the stress, reboot, adjust, rinse/repeat.  Throughout this guessing game I would save as V1,2,3,4,5,etc, then change stuff and repeat.  Trouble is my understanding of PBO and being somewhat dyslexic has proven challanging.  And yes I have written down the order of the best performing cores down to the worst, info obtained from hwinfo while running stress tests or after.

Any suggestions on how to get Ryzen Master working would be great!  Im thinking it has nothing to do with drivers, instead some service i disabled from starting up.

 

So here are a few questions, yes noob questions im sure but will probably help clarify things for me and others-

-whats that multiplier in cinebench even mean?

-how does the single core bench work in R23 since it appears to only hit 4 cores out of 16?

-why is the auto-overclock without PBO get about 1000pts more on MT, yet 100 points less on single thread?  better temps and lower speeds??  Why is the overclock at a steady speed outperforming my curve?

-related to the above.  on my various attempts at shoving numbers into PBO and potentially turning this or that off in bios (usually not aware of what it does other than google), why is it that even if i manage to get 5ghz+ across all cores is my score LESS than auto-tune at 4.1 steady?  yes temps are high, 90c.  Adjusted the max temp to be 95 just cause the pc crashes due to hitting 90.8c or something for a split second. Why am I pulling lower scores at higher clocks? 

-how does the PBO work in regards to positive and negative values?  Doing a per-core config, all negative, is a LOWER number better for poorer cores (ie -30, -25) or is a higher number better? (ie, -5, -3)  I have looked into this extensively but as mentioned its confusing given dyslexia.  a lower value increases the voltage?  Does that mean -30 is a waterfall and -3 is spilling a little water?

-what does PBO actually DO, restrict voltages and somehow in turn increases clock speeds yet somehow increases heat?  Basically break down the PBO part for me.  Its a simple concept yes but the "lower" setting does... what?

-can or should I use positive adjustments on a per-core basis?  if so what would that do? 

-summarizing all the PBO stuff, how to effectively implement it?  should my hot cores get a lower number?  higher number?  all negative values?  lower again being -30 as opposed to -1.  should my cooler cores get a larger or smaller value?

-what are the three values and how do they relate to the functionality of PBO let alone the neg or positive offsets?  i think ones PPT, forget the other two.  I do know that im at 250/200/150 in that order.  Apparently a forum read was right, the first number, PPT i think, should be lowered if the CPU (Tctl/Tdie) gets too hot.  That of which i did, it was at 300PPT so i lowered it to 250, temps are about 10c lower from 90.  forget if it affected the clocks or not but whatever, irrelevent.  what do the 3 numbers do in relation to pos\neg, voltage(watts?) and however PBO works?  possibly provide a visual aid such as a graph would greatly help me to understand, again as i easily mix up higher and lower since they do the opposite with a negative offset\curve.

-what are your PPT/something/something numbers and why?  What are your best cores set at for PBO?  plz explain why u set things as such, would help with my understanding of all this.

-wtf is MP Ratio after a "single core" bench?  why does it fluctuate between setups?  is higher ideal or lower?  or... does it not matter at all?

-other than all that, are there any options i should enable or disable in BIOS to improve performance and/or temps?  Should I Uncore/SoC, splitting infinity fabric, bclk and all that to run independently but with fixed values?  idk just an example of stuff ive tried.

-in terms of figuring out the best core, should i go off the temps at idle?  ignore the avg/min/max and just watch them?  higher temp = underperforming cpu versus lower temp = potential for PBO jazz?  Again, u can see how this is confusing!  Maybe its the opposite, higher temps = better cpus?  seems unlikely but thats why i ask.

-How to figure out the best cores to medium to terrible.  should I go off HWinfo's "core clocks" section that says (perf #X/X)? According to the instructions means the perf#(number) is your preffered core and the number to the right is the core hardware-labeled.  Say the "best" core is #1 but its actually idk, core 6.  just an example.  OR is this Window's evaluation of which are PREFERRED to use rather than what the best ones are relative to overclocking headroom?  For example, I love to drive my fast ferrari but the slow chevy uses less gas, costs less to insure and gets me to the same destination.  therefore the chevy is my car of choice, uses less resources and still gets me there.

-why do i get higher scores with lower clocks versus higher clocks and lower scores?  seems backwards to me - even if the temps are ok but not great?

-lastly, ANY help with Ryzen Master would be AMAZING!!!!!!!!!  Is there a way to figure out *why* it wont start?  I havent checked the error logs yet, just been playing with all this manually in bios.  Is it possible i disabled a service it depends on?  if so, can i find out which one(s) they are?  I'd rather not reinstall windows/drivers/apps just to get it going...

 

i swear, im sure all it would take is a single puzzle piece and ill be able to see the entire picture.  Help, anybody, ahhhh!!

[Cramig88]

also for simplicity i have manually written down the prefCore / Hardware Core #.  Listed below.  Strange cause it seems like theyre nearly the same, 1 pref is core 1/2, almost like the hardware ID is +1 from the pref core

so yeah, should i be adjustiing my pbo settings based on temp or the pref/hardware cores?  and how?  whats it do?  eh u prob read the questions

 

Preffered / HW core number

1 / 1

1 / 2

2 / 3

3 / 4

4 / 5

5 / 6

6 / 7

7 / 8

8 / 9

9 / 10

10 / 11

11 / 12

12 / 13

13 / 14

15 / 15

15 / 16

and YES im aware amd (and everything pc related) starts at ZERO, not 1.  so yes im entering in values correctly, i double\tripe\quad check

[freeagent]

I run my 5900X at 260/170/200 -25 all core, +200.

 

She boosts pretty hard.

 

 

[Cramig88]

260/170/200, what do those values mean?

my understanding is PPT/TDC and EDC in that order.  a lower ppt (say 200) will improve temps on the Tdie (lower them)

TDC and EDC, higher the values, worse at single thread.  Lower, better at single thread loads.

why do -25 on all cores?  just curious, u could get more mhz tuning them individually.  from what ive noticed so far the closer the neg value is to zero the "worse" that core is.  the further from 0 (-25) the better that individual core is.  so with 16 to deal with, lotta work testing single threaded loads and assigning cineR23 (thru taskman) to use core 0 or 5 or idk 15
that being said my understanding is u take the ones overheating and give them a value closer to zero which in turn increases the mhz and reduce heat.  take the ones not overheating then farther from zero u go it will increase the temps while increasing speeds.  then in the end your left with a balanced set of cores with thermals in check.  Accurate?  not sure just reading and reading.

 

whats your boost frequency at and temp if u dont mind me asking?  i had a 5900x but upgraded not long ago to this 5950 (might be why ryzen master wont work)

[freeagent]

The first CCD on my CPU has a few cores that hit 5150MHz, a few that hit 5GHz, and the other CCD they vary from 4950 down to 4850, that is effective clocks.

 

For all core, it depends on what's running, Linpack Xtreme 10GB load runs at 4550MHz, and something light like TM5 will run at ~4750MHz all core.

 

My CPU is old, like first run old.

 

She rips though.

 

I can go further down on the negative scale, but nothing likes it and it will crash.

 

Edit:

 

Ryzen Master is shit, I just use the bios.

[Cramig88]

hm well that is informative, not all cpu's can/will hit the peak.  suppose for my own setup ill reduce the boost to the peak of the slowest and pair them using that speed.  later on bump it up.  did not know that!  given what ive learned so far about the 3 numbers thats sort of a guessing game.  ive seen forums with ppl that have the same equipment, aside from ram, and suggest 260/170/200, 225/325/245, heck one post said edc185-190, tdc165-170 and ppt270-300 (out of order from the rest).  i tried that last one and the higher ppt numbers like 250-300 and i typically burn up at a lower clock.

 

heres one i found online, defines the three numbers - for anybody else reading, helpful!

PPT only matters in the sense that it determines your max TDC.  you could set PPT to 300W, but if your tdc is limited to 50A you wont ever hit that 300w target limit.

     -basically to me, it sounds like a limiter.  tdc cant go past "this number". the tdc will start slow and get up to ppt related to the workload.

TDC is what will determine your maximum clock speed.  Simply put, higher tdc usually means higher clocks.  This will plateau at a certain point.

     -so it determines how fast the core will go..  ultimately our limit is ppt though, why the need to input a value?  is that the starting amps/watts/whatever, then under load it hits the ppt limit?

EDC (increasing it) allows for you to set a more aggressive CO.  Setting mine to 200 allowed me to set an all-core -30, whereas at 160 i couldnt even get at -20 on my best cores, -15 and -10 on the rest.

     -more aggressive curve.  so i guess more velocity takes more distance to stop.  this guy takes the TDP in relation to its PPT limit and pushes the curve harder.  no different than slamming on the gas versus casually getting up to speed?

 

also as mentioned another post i found mentioned that the tdc/edc on lower numbers, work better with single thread (they prob meant core) situations.  so inversely a high tdc/edc would not work as good on a single core process.

 

far as ryzen master goes I figured it would be a good way to get numbers to start with.  assuming it can do per-core PBO optimization?

using easy apps is how i used to OC my 5900x except i wasnt using PBO at all, id be stuck on all cores.  that performance multiplier or something along with noob one-click tuning apps.  managed to get her going at 5.1 around 75c - constantly.  though now that im actually diving into PBO its reasonable to think there was a much better way of doing things.  it got really old hearing all these fans going berserk!!  toned it down to 4.2 just bc a sony ps2 (3?) emulator would crash the system.  i just wanted to play skate!

 

now that ive arranged my good/bad/ugly "recommended" settings based on max temps, lets see how they work when i punch em in!

hell right now im at 5.1 and 47c - seems windows cooled down.  prob happened so slowly due to the EDC at 190

[Cramig88]

just got ryzen master working.  apparently i just had to remove amd gpu stuff, restart then install ryzen master.  weird...

whatever, back to bios!

[freeagent]

Don't mind the Thermal limit.. using shitty TIM right now, got my regular stuff on the way thankfully.