Asus M/B Offset and Adaptive Voltage mode: Voltage Spikes

[dmt_taran]

I've read tons of threads all over the internet with such a name (or close to it), but I've not found at least suitable answer or solution to this issue (exept for chit chats, "i dunno dude, I was just passing by" sort)

I have 8700K with maximus x hero system. CPU is capable of 4.8 GHz at 1.24V, 4.9 at 1.3, 5.0 at 1.355, 5.1-5.2 over 1.4V, uncore ~4.5GHz at every step, no AVX offset. Every mark, every clock step I've tested gradually last 4 months, and I'm 200% sure it Is stable at those clocks with voltage mentioned. In addition, I've used manual voltage and LLC6 for additional stability, IA lines are set to 0.01.

Target is 5.0GHz
The problem is - I can't get stable OC and voltage while using offset and adaptive. Unlike manual voltage, adaptive and offset makes max load voltage fluctuate at 0.05V range. So either I have unstable OC with 1.312-1.328V constant supply with spikes up to 1.36V, or I have stable OC, but with spikes up to 1.408-1.424V.
The problem becomes more complicated under heavy workloads (e.x. BF1). These spikes get more rapid - I could even say that 50% of time CPU gets more excessive voltage.

However, it's not enough to utilize stable OC with lowered voltage - for e.x., suggesting spikes go 50% of workload time and even more, I could lower voltage below stability point, achieving sweet spot between spikes and undervoltage - but it doesn't work, OC is not stable! :mad:

So what is it about? Why voltage is so unstable while using adaptive and offset? How can I solve this problem? Even if it cannot be fixed while under load, maybe there is some way to decrease these spikes, at least to 0.01V mark?

[bass_junkie_xl]

Unluggy71ab wrote:
thanks, can confirm i made adjustments to bios as you recommended. ran realbench 3x, seems no issues, though not sure why my score goes down per each run. hwmonitor shows vcore max was 1.344V (was this max even before loading realbench), which i'm not sure how as i set it to 1.330v and left the + adaptive to auto . min vcore 0.664. ill run heavenbench soon too and maybe bit of prime95 to confirm stability but so far is good, can see lower idle temps/idle vcore voltes which is more or less what i wanted to acheive. whilst maintaining stability. guess ill find out for sure when playing some games tonight 🙂

questions:
* how about speedshift and speed step. should they be enabled?
* with the adaptive mode, i selected +. wasn't sure exactly what you meant by offset=auto(none), does that mean leave it at auto or put it to 0? for now it's left at auto.
* i have avx on 3 as noticed temps go up bit more and things get louder in those situations, how does that tie in with that situatiion you mentioned about spikes? to tighten it up would i benfit from going back to avx-1 rather than avx-3?

bios changelog (the ones not shown in changelog, i already had it there from yoru sugegstions in past posts)
cpu svid support -> enabled
bclk aware adaptive voltage -> disabled
cpu core/cahce voltage -> adaptive mode
additional tubro mode cpu core voltage auto -> 1.330v
cpu load line calibration lvl 5 -> lvl 6
long duration package power limit auto -> 4095 (this is max on this mb)
short duration package power limit auto -> 4095
ia ac load line auto -> 0.01
ia dc load line auto -> 0.01
cpu c-states disabled -> auto
speedstep enabled

questions:
* how about speedshift and speed step. should they be enabled? - yes

with the adaptive mode, i selected +. wasn't sure exactly what you meant by offset=auto(none), does that mean leave it at auto or put it to 0? for now it's left at auto. - offset means you can make a + ( positive ) *add some volts * to your cpu volts or you change the + ( positive ) to a - ( negative sign ) to * take a way * some volts - example to under stand , so say your cpu needs 1.3v for 5.0 ghz in * manual vcore mode * now u want switch to adaptive mode so you type in total adaptive volts = 1.3 v so you load up windows and you notice that your getting 1.328 v - 1.332 v under load ... so you go back to the bios under adaptive mode offset and change the + * positive symbol to a - negtive symbol ( we need to take some volts away ) and you would do some simple math
1.300v = stability but adaptive is giving me 1.328 v
1.300v - 1.328 v = 0.028 v difference thats what you type in offset volts - 0.028v so now in windows you wont see 1.328v you should see 1.3 v since you made that offset under stand ? same thing goes for + positive offset say you got a crappy 8700k that needs 1.4v for 5 ghz but in adaptive mode its only giving it 1.37v under load you would add a + positive offset of + 0.030 ( 1.37v + 0.030 v = 1.4v ) so your stable .

i was saying that if you ( * do not use * ) a - avx offset you will or * could see * up to 0.040 v spikes under load on top of your already set vcore
example 1.3v no avx offset = up to 1.34 v under loads in avx witch i was saying that if you use a avx offset of - 1 you will only see up to a 0.020 v spike vs the 0.040v with out one , if you use a avx offset of - 2 you will not have any or little volt spikes ( maybee 0.016 v tops ) .

its a trade off to not get those volt spikes in adaptive mode . trade off is , instead of a constant 5 ghz @ 1.3v + 0.040 v spikes you get 4.9 ghz on avx loads and only 1.32 v spikes .

some cpus dont spike as high as others as they have a better svid voltege table vs other cpus that require more volts for a givin speed .

i read threw a 90 page fprum on over clock .net and drank a pot of coffee can learn alot from those guys that have been there done that for years .

i only been around since the core 2 duo first came out e6600 @ 3.4 ghz .... lol

[Unluggy71ab]

bass junkie xl wrote:
questions:
* how about speedshift and speed step. should they be enabled? - yes

with the adaptive mode, i selected +. wasn't sure exactly what you meant by offset=auto(none), does that mean leave it at auto or put it to 0? for now it's left at auto. - offset means you can make a + ( positive ) *add some volts * to your cpu volts or you change the + ( positive ) to a - ( negative sign ) to * take a way * some volts - example to under stand , so say your cpu needs 1.3v for 5.0 ghz in * manual vcore mode * now u want switch to adaptive mode so you type in total adaptive volts = 1.3 v so you load up windows and you notice that your getting 1.328 v - 1.332 v under load ... so you go back to the bios under adaptive mode offset and change the + * positive symbol to a - negtive symbol ( we need to take some volts away ) and you would do some simple math
1.300v = stability but adaptive is giving me 1.328 v
1.300v - 1.328 v = 0.028 v difference thats what you type in offset volts - 0.028v so now in windows you wont see 1.328v you should see 1.3 v since you made that offset under stand ? same thing goes for + positive offset say you got a crappy 8700k that needs 1.4v for 5 ghz but in adaptive mode its only giving it 1.37v under load you would add a + positive offset of + 0.030 ( 1.37v + 0.030 v = 1.4v ) so your stable .

i was saying that if you ( * do not use * ) a - avx offset you will or * could see * up to 0.040 v spikes under load on top of your already set vcore
example 1.3v no avx offset = up to 1.34 v under loads in avx witch i was saying that if you use a avx offset of - 1 you will only see up to a 0.020 v spike vs the 0.040v with out one , if you use a avx offset of - 2 you will not have any or little volt spikes ( maybee 0.016 v tops ) .

its a trade off to not get those volt spikes in adaptive mode . trade off is , instead of a constant 5 ghz @ 1.3v + 0.040 v spikes you get 4.9 ghz on avx loads and only 1.32 v spikes .

some cpus dont spike as high as others as they have a better svid voltege table vs other cpus that require more volts for a givin speed .

i read threw a 90 page fprum on over clock .net and drank a pot of coffee can learn alot from those guys that have been there done that for years .

i only been around since the core 2 duo first came out e6600 @ 3.4 ghz .... lol

thanks a lot mate. just posting back to say i've taken your advice on the settings and couple days in now with several hours in overwatch + 20ish runs of realbench and 5ish hours prime95 has proven stable. the small spike up to 1.344 comes in when the avx instructions come on from prime95 it seems, all other occasions logging with hwinfo the vcore is nice and low in idle and stable at 1.328 under load conditions. this marks the success of adaptive mode for me, which i had previous spent so many hours on and failed, so thanks for your help! 🙂

[bass_junkie_xl]

Unluggy71ab wrote:
thanks a lot mate. just posting back to say i've taken your advice on the settings and couple days in now with several hours in overwatch + 20ish runs of realbench and 5ish hours prime95 has proven stable. the small spike up to 1.344 comes in when the avx instructions come on from prime95 it seems, all other occasions logging with hwinfo the vcore is nice and low in idle and stable at 1.328 under load conditions. this marks the success of adaptive mode for me, which i had previous spent so many hours on and failed, so thanks for your help! 🙂

good to hear , thanks to the few fellows @ over clock.net and there convos i read through .

[bass_junkie_xl]

Unluggy71ab wrote:
thats awesome dude! sorry for amateur question but would you mind putting screenshots/pics of the relevant areas to be able to acheive the relatively stable adaptive mode setup on this mobo?

i have z370-h (BIOS 1704), 8600k 5ghz, 1.330v stable in manual mode. any options i've tried with adaptive mode seem to have been a failure 😞 i guess tehre's no inherent problem with what i have but always felt it could be a bit more efficient atleast .

In bios do these. If your 5 ghz is stable @ 1.33v

Multy = 50
Avx = - 1 ( only if u want up to a max of 0.016 v spike )
No avx u will see up to 0.040v spike 1.33 + 0.040 = 1.37v

So try avx 1 if u see it spike alot past 1.33 -1.35 v
Svid =on
Svid = best case cinario
Cpu power options to max min/ max deration = 4555.5
( what ever max is prevents throttling)

Ia / dc load line both to = 0.01

Load line = 6 ( 7 is max )
Blck adaptive aware volts = disabled.


Cpu volts adaptive mode
Total adaptive volts = 1.33v
Offset = auto ( none ) unless u spike .
Bios c states to all auto.

Xmp on.
Set windows power plan to balanced.

Should be good.

Just keep hardware monitor 64 open keeping an eye on min/avg/ max vcore .

I would rather use a 5.1 ghz with a - 1 avx ( 5 ghz on avx loads ) @ 1.36v with a spike up to 1.376 v then not use a avx offset and see those damn 0.040v spikes.

Some good quality cpus ( higher binned ones )
See less volt spikes then others that are lower quality since we are using intels svid tables on the cpu aka
Adaptive mode.

[kalle-m]

bass junkie xl wrote:
ok i have narrowed down the "adaptive mode volt spikes " issue

if you " do not use " a avx offset you will see " up to " 0.040 v spikes in adaptive mode .

example 5 ghz , load line=6 , ia/dc =0.01 , total adaptive volts = 1.28 v to be stable avx = 0 you will see spikes " up to " 1.312 - 1.32 v .

if you use a avx offset of = - 1 the " volt spikes " will only be " up to " 0.016 v - 0.020 v
example : 5 ghz 1.28v avx = -1 , = " up to " 1.296 v spike vs 1.312v - 1.32v spikes with out a avx offset .

i was reading in the over clock .net forums and guy narrowed it down and tested 8 8700k , a few 8086k in adaptive mode and had it down to a science ,

no avx offset = up to 0.040 v spikes
avx offset = -1 = up to 0.016 - 0.020 v spikes
avx offset = - 2 = no spikes or minamal up to 0.016v tops !

so what i did was use my 8700k thats stable @ 5.1 ghz @ 1.36v manual switched to adaptive mode set a avx offset of = - 1
total adaptive volts = 1.36v ( cpus svid tables work in 0.016 v incriments on load ) load line = 6 and now i only get max of
1.376 v on avx loads ( +0.016 v spike ) on avx loads .
playing games with out avx bf4 it sits @ 1.36v no spike ,

i also tested 5.2 ghz with avx offset of =-2 @ 1.4v and didnt get any spikes period .

in the end live with a - 1 avx and only get up to a 0.016v spike or dont set a avx offset and recieve up to 0.040 v spike . - 2 avx = no spike .

this is awesome I have to test this also, I never use avx offset because I only dun 4.7ghz. but if this is true for my 8600k also then I am gonna push it to whatever frequency 1.25v will give me!

was running occt on medium and it was doing 1.28v, when in every other benchmark and test it goes only to 1.18 max...…

[bass_junkie_xl]

kalle-m wrote:
this is awesome I have to test this also, I never use avx offset because I only dun 4.7ghz. but if this is true for my 8600k also then I am gonna push it to whatever frequency 1.25v will give me!

was running occt on medium and it was doing 1.28v, when in every other benchmark and test it goes only to 1.18 max...…

yeah it works tested it for a few days .

load line level 6 and say 1.3 total adaptive volts to 1.3 no adaptive offset left at auto will give you 1.34v on avx loads .

-set a - 1 avx offset and your 1.3v will only spike to 1.316v - 1.32v max max

-set a - 2 avx and you will have 0 spike if at all .

- dont use avx offset and live with 0.040v spikes