Overclock.net banner

1 - 20 of 56 Posts

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #1 · (Edited)
The Goal: getting a 24/7 OC that gives you both your OC when you need it, and kicks in all the power saving features when it’s not.

The following guide shows you how to OC voltage tune an Intel 10th and 9th gen processor when using Adaptive Mode. This is not an full up OC guide; it assumes that a stable Manual Mode OC has been established. I hope it helps others dealing with getting a 24/7 Adaptive Mode OC dialed in for their cpus quickly and effectively.

Use the below at your own risk, I take no responsibility for what may happen. I welcome comments and criticism that will enhance, or possibly destroy, this guide's usefulness or accuracy.

Background: You have dialed in a stable OC using Manual Mode voltage, along with LLC and other settings that insure stability while keeping temps in check. Then you decide to use Adaptive Mode to enable lower voltages when the processor has light or no loads. But when you do, you find that the voltage value you’ve entered for using Adaptive Mode seems to be ignored; the load voltages are now higher than they were when using Manual Mode. The following discusses how to get around this increased load voltage behavior when using Adaptive Mode.

Technical background: The Adaptive Mode OC voltage tuning methodology is based on information Shamino posted on the ROG forum about using Adaptive Mode for a 9900ks. It is applicable to the 10th gen as well. For a more detailed description of what is at work here with the following tuning, this post by Shamino is highly recommended reading.

Shamino said:
per intel design:
1) adaptive voltage only takes effect at ratios > max single core boost ratio which is 50x in 9900ks case (so if you set 1.5v at 50x for eg, it gets ignored till you go up to 51x)

2) any volt set below the cpu's default vid gets ignored (eg setting 51x ratio with 1.1v when cpu's vid is 1.25v at 50x will get ignored)

On client you still have AC/DC loadline to manipulate to undervolt proc to a certain extent (you have no room to undervolt when AC/DC LL is already at minimum) to undervolt further you will have to rely on VRM Loadline.
This is saying that if you got a CPU which you target an all core OC that does not exceed its max single core turbo boost frequency, Adaptive Mode uses the VID voltage rather that your Adaptive Mode voltage setting. This explains the behavior of Adaptive Mode mentioned above. Shamino then gives a workaround: the IA AC loadline value can be used to lower (undervolt) this VID voltage, at least when IA AC LL is not at its minimum, which is 0.01. I’ve seen OC guides that recommend setting IA AC LL to this value to lower the load voltages but don’t explain why it works. A big clue is that the IA AC loadline value is indirectly set by the Asus bios setting “SVID Behavior”. Its “Best Case Scenario” setting sets IA AC LL to 0.01. On my ROG Maximus XII mobo, other “SVID Behavior” settings set IA AC LL to much higher values, such as 0.6 and 0.8. If I use “Best Case Scenario” in Adaptive mode (IA AC LL set at .01), my 5.0Ghz OC fails due to insufficient load voltages. All the other “SVID Behavior” setting yields load voltages in excess of my Manual mode setting’s load voltages. From this I’ve surmised that their associated AI AC LL values are too high, thus producing the excessive load voltages for my OC. So I tried directly setting the IA AC LL value. After a few trials setting IA AC LL, I found that a value 0.35 yielded the load voltages which my Manual voltage produced…success! Note that the SVID Behavior setting only takes effect if IA AC LL is set to Auto, and AFAIK IA DC LL should always be set to the IA AC LL value. A slightly unrelated gotcha: a processor may not drop to its low level/idle frequencies if both cache ratios values are set the same (other than auto). I only set its max value, min is set to auto and this works as expected.

Adaptive Mode OC voltage tuning methodology:

1. OC your computer using Manual Mode to find an optimal voltage and CPU ratio, LLC value, and all other setting (e.g. ram, cache ratio, SA, IO) that yields a stable OC (this is where an OC guide fits in). It is advised to use moderate values for LLC (Asus 4, 5).

2. Record the load voltages for the stress tests you used, this info will be used later. Using more than one stress test is best. If you’ve found your OC surpasses the CPU's max single turbo frequency, congratulations on getting an exceptional chip though you most likely won't be able to use IA AC LL to tune your load voltages; you will find out if this is so in step 5, which also give alternatives to use.

3. Go into the bios and enable Adaptive Mode and leave all Adaptive Mode settings at default, i.e. auto. Alternatively, leaving the voltage setting on Auto also enables Adaptive Mode.

4. Set “SVID Behavior” to “Best Case Scenario”. For non-Asus users, you’ll have to figure out what your mobo’s equivalent is. If anything, you could directly set IA AC LL to 0.01 (always set IA DC LL to the same value).

5. Boot up and run your OC stress tests and observe the load voltages they produce. If they exceed the Manual mode load voltages, you can try any or all of these to lower the voltage: enabling TVB, using a negative voltage offset value for the OC frequency's applicable V/F point or lowering your LLC value; beware that from here you are on your own (see * below). If the load voltages are “as expected” for your recorded voltages, congrats you are done, go enjoy using your OC! If they are less (the usual case OCs within the turbo freqs), your computer will soon blue screen or lock up if it hasn’t already.

6. Go into the bios and set IA AC/DC LL to 0.1, boot, run your OC stress tests and observe the load voltages they produce. If they exceed the Manual Mode load voltages, go into the bios again and try subtracting 0.01 or 0.02 from IA AC/DC LL, boot, run your OC stress tests and observe the load voltages they produce. Repeat this subtraction process until you reach “as expected”. If the load voltages are at “as expected”, you are done tuning. If the load voltages are less than “as expected”, go into the bios again and add 0.1 to IA AC/DC LL, boot, run your OC stress tests and observe the load voltages they produce. Continue the processes of this step until you reach your “as expected” load voltages. Once you've completed the tuning, congratulations: your CPU is now running a 24/7 OC as efficiently as it can (just as is my 10700K)!

Obviously, rerunning your OC stability tests suite is recommended after completing this tuning. Enjoy!

* Looks that Intel gave my 10700k a V/F curve that targets it working on the weakest of mobos. I found there's V/F point tuning leeway available when using a good quality 10th gen mobo. Each V/F point has its own voltage offset that can be used to lower the point's VID by setting it to a negative value. Also, at least for a 10700k, enabling TVB significantly lowers the VID without having to change V/F point offsets (thanks to GeneO for pointing out this BIOS setting's effect). On 9th gen, or it these features are absent, the Adaptive Mode offset can be used in the same manner though it may cause stability side effects when idling.

Update: If your OC exceeds the max turbo frequency, Adaptive Mode uses the voltage of the highest frequency V/F point while applying all the rules given above. In this situation, if IA AC LL is at its min of .01, to get lower load voltages use either or both enabling TVB and lower the load voltage of the highest frequency V/F point by using a negative voltage for its offset. This is what I did to get a 5.2Ghz no HT Adaptive Mode OC for my i7 10700k.

Update for 9th gen CPUs: This guide was developed using my 10700k in a Z490 mobo. Though this guide will work for 9th gen CPUs, later in this topic and linked here, ThinbinJim posted these specifics for using IA AC/DC LL tuning on a Z390 mobo which you'll find helpful.
 

·
The last VRM burner
Joined
·
2,590 Posts
Uh, why not just leave C States and SpeedStep enabled and continue to work on override voltage mode?
 

·
Registered
Joined
·
2 Posts
just did quick test cause my 9900K need 1.48v at 5.1GHz to be stable, wish I knew about this before

before on manual stable 5.1
Bios Vcore 1.48v LLC5
HW Vcore min 1.376v max 1.492v
core VID min 1.272v max 1.329v
CPU power max 300w
temp idle 27c max 89c in prime95 112k avx.

even though it was stable 24hr on prime95 and linpackX never felt comfortable at this voltage idle in windows or browsing 1.47v+ all the time so had to run it 5.0 with 1.41v

now with adaptive
Bios + additional turbo voltage 0.500v offset auto ACDC 0.8 LLC5 (still don't fully understand the relation between them I just throw random numbers until i got close to my manual setting)
HW Vcore min 1.341v max 1.394v
core VID min 1.286v max 1.376v
CPU power max 297w
temp idle 25c max 89c in prime95 112k avx.

did 30min prime95 112k no error so this looks promising gonna try 5.2GHz I may have some voltage room now

thanks man
 

Attachments

·
Optimal Pessimist
Joined
·
3,062 Posts
" adaptive voltage only takes effect at ratios > max single core boost ratio which is 50x in 9900ks case (so if you set 1.5v at 50x for eg, it gets ignored till you go up to 51x)"


With the 10xxxx series, you can overcome the "additional turbo voltage" being ignored below the max turbo multiplier by modifying the VID using the VF curve offsets.


.
 

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #5 · (Edited)
With the 10xxxx series, you can overcome the "additional turbo voltage" being ignored below the max turbo multiplier by modifying the VID using the VF curve offsets.
Thanks for bringing this up. I added this "back door" to the guide earlier today. Seems that that's the only way to tune load voltages if your OC's mult is the max single turbo mult. Looks that Intel gave "less than perfect" 10700k's a V/F curve that targets working on the weakest of mobos, hence there's V/F point tuning leeway for a good mobo. I took this route to get a stable 5.1Ghz OC (used a -0.065 V/F point offset), though anything AVX2 instantly thermal throttles, so on goes the AVX offset (ugh). Other than seeing if a 5.1 OC could be done, I'll probably not use it much. A have a 4.9 OC whose load voltages are really sweet...it will probably win my 24/7 derby (oh well).

thanks man
Glad that you figured it out. Hope that OC gets tied down.

Uh, why not just leave C States and SpeedStep enabled and continue to work on override voltage mode?
Uh, because it doesn't work that way: manual mode (override voltage) sets the CPU's VID to be the constant value of override voltage, which in turn effectively disables the c-states low voltages effect. The only way to get the EIST low voltages is to use either Adaptive or Offset modes.
 

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #6 · (Edited)
Here's a shameless bump (thread necromancy I suppose). I've cleaned this guide up a lot from its original form, and it seems to be getting a constant stream of web search hits. For those who have not upgraded CPUs for a long time, it's absolutely needed info on using Adaptive Mode for overclocking. I've seen a few recent guides that claim "set this voltage in Adaptive mode then use this offset with it too"...that's NOT the most efficient way to achieve an Adaptive mode OC...this guide's info is.
 

·
Registered
Joined
·
107 Posts
Been following this thread for awhile, great resource!

Could you explain the reasoning behind only using middle range LLC settings with AC/DC loadline? I'm rather curious... I found using a lower LLC setting (more vdroop) and moving from .01 AC/DC to .2-.3 allows me to get closer to my stable fixed voltage in general. I remember reading a Falkentyne post that touched on this but haven't been able to find the thread.

Here's a shameless bump (thread necromancy I suppose). I've cleaned this guide up a lot from its original form, and it seems to be getting a constant streams of web search hits. For those who have not upgraded CPUs for a long time, it's absolutely needed info on using Adaptive Mode for overclocking. I've seen a few recent guides that claim "set this voltage in Adaptive mode then use this offset with it too"...that's NOT the the most efficient way to achieve an Adaptive mode OC...this guide is.
 

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #8 ·
The reasoning for the advised LLC values was to both lower the idle voltage (for lower LLC) while maintaining the overshoot control benefits of LLC (for higher LLC). It's advised only for these reasons, but it's definitely not necessary.
 

·
Registered
Joined
·
107 Posts
Awesome, appreciate the explanation. I've based my LLC selection mainly on Buildzoid's recent Z490 motherboard findings. LLC 5 seems to be the best option for the Z490 Unify.
 

·
Registered
Joined
·
6 Posts
Thanks a lot for the guide! I had my 9900K on 48 (I don't need a high overclock and prefer lower thermals & noise) with Manual 1.25V, LLC 5. Vcore was 1.248 in idle and 1.184 while stresstesting with max. 73C.
After setting Adaptive and lowering IA AC/DC LL from 0.1 down to 0.06 I ended up at around 1.2V idle, 1.165V load and max. of 69C.
I'm wondering though, if it's normal for the voltage to be quite constant around 1.2V when doing only light computing, it seems not to fluctuate almost at all. I think that on stock settings it was sometimes even falling under 1V.
 

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #11 · (Edited)
I'm wondering though, if it's normal for the voltage to be quite constant around 1.2V when doing only light computing, it seems not to fluctuate almost at all. I think that on stock settings it was sometimes even falling under 1V.
What you are seeing is normal. With light loads, vdroop is not applied so the effective voltages remain at the VID which the CPU requests. The VID is set by Intel so it's definitely safe. The application of vdroop is in relation to the CPU's power draw, its purpose is to minimize voltage spikes when there are sudden changes in power draw. With light loads the power draw changes are minimal, hence why vdroop isn't being applied.

To fully use the Intel power saving features, c-states and EIST must be enabled in the bios, and the Windows power plan in use must allow low power states, e.g. Balanced. What all this does is allow the CPU to drop its clock multiplier really low when not in use, which also results in really low voltages. By your name, seems that you might be a Linux user; I imagine it has something similar to the Windows power plan.

I'm surprised you got your OC stable with a lower load voltage than you had with Manual Mode. Nice! I didn't see that behavior with my i7 10700k.
 

·
Registered
Joined
·
6 Posts
What you are seeing is normal. With light loads, vdroop is not applied so the effective voltages remain at the VID which the CPU requests. The VID is set by Intel so it's definitely safe. The application of vdroop is in relation to the CPU's power draw, its purpose is to minimize voltage spikes when there are sudden changes in power draw. With light loads the power draw changes are minimal, hence why vdroop isn't being applied.
That makes a lot of sense, thanks! I was wondering, because by default (stock BIOS settings), the voltage was jumping around between I think 1V or even lower and up to 1.4V (the main reason I even started bothering with OC, undervolt and whatnot).

To fully use the Intel power saving features, c-states and EIST must be enabled in the bios, and the Windows power plan in use must allow low power states, e.g. Balanced. What all this does is allow the CPU to drop its clock multiplier really low when not in use, which also results in really low voltages. By your name, seems that you might be a Linux user; I imagine it has something similar to the Windows power plan.
Okay, I will have a look for those options, maybe they're on by default but turned off once you start OC'ing? Because I never touched those. Yeah I mostly use macOS and Linux, but didn't change anything in the operating systems, so if they had "power saving mode" before and they don't anymore, it's weird. I'll have a look in Windows though, if the voltage also stays stable on 1.2V or drops below sometimes.

I'm surprised you got your OC stable with a lower load voltage than you had with Manual Mode. Nice! I didn't see that behavior with my i7 10700k.
Yeah, me too, I am honestly not sure why, but if it works I guess I'm happy with that :D
 

·
Flight Sim Enthusiast
Joined
·
330 Posts
Discussion Starter · #13 ·
When your computer is idling the CPU multiplier should drop down to 8 with voltages less than 1v. If you saw that voltage drop behavior with full stock bios settings, the power saving features are working with your OS. This points to something with your OC settings is preventing the idle power states from being entered. I'm not having any such problems with my OC, and at best can only wildly guess what's going on with yours (so I won't). If you can't isolate it, start a new topic on your problem; people will see it and respond.
 
  • Rep+
Reactions: unixb0y

·
Registered
Joined
·
18 Posts
This method is extremely underrated as nobody bothers trying to understand how adaptive works when trying to OC. Not a bad guide at all but I do have some comments as someone who's been using the AC/DC LL method for overclocking/undervolting for 2 years now with the 9900k, and 9900ks subsequently.

Adaptive Mode OC voltage tuning methodology:

1. OC your computer using Manual Mode to find an optimal voltage and CPU ratio, LLC value, and all other setting (e.g. ram, cache ratio, SA, IO) that yields a stable OC (this is where an OC guide fits in). It is advised to use moderate values for LLC (Asus 4, 5).

2. If you’ve found your CPU is an exceptional OCer (comfortably surpasses max single turbo frequency), congratulations though you won't be able to use IA AC LL to tune your load voltages (see the notes below these steps). Whatever the OC, record the load voltages for the stress tests you used. Having more than one stress test is best.

3. Go into the bios and enable Adaptive Mode and leave all Adaptive Mode settings at default, i.e. auto.

4. Set “SVID Behavior” to “Best Case Scenario”. For non-Asus users, you’ll have to figure out what your mobo’s equivalent is. If anything, you could directly set IA AC LL to 0.01 (always set IA DC LL to the same value).

5. Boot up and run your OC stress tests and observe the load voltages they produce. If they exceed the Manual mode load voltages, you can try any or all of these to lower the voltage: enabling TVB, using a negative voltage offset value for the OC frequency's applicable V/F point or lowering your LLC value; beware that from here you are on your own (see * below). If the load voltages are “as expected” for your recorded voltages, congrats you are done, go enjoy using your OC! If they are less (the usual case OCs within the turbo freqs), your computer will soon blue screen or lock up if it hasn’t already.

6. Go into the bios and set IA AC/DC LL to 0.1, boot, run your OC stress tests and observe the load voltages they produce. If they exceed the Manual Mode load voltages, try subtracting 0.01 or 0.02 from IA AC/DC LL and repeat this subtraction until you reach “as expected”. If the load voltages are at “as expected”, you are done tuning. If the load voltages are less than “as expected”, add 0.1 to IA AC/DC LL and continue the process of this step until you reach your “as expected” load voltages. Once you've completed the tuning, congratulations: your CPU is now running a 24/7 OC as efficiently as it can (just as is my 10700K)!

Obviously, rerunning your OC stability tests suite is recommended after completing this tuning. Enjoy!
1. You can benefit from the superior transient response of lower LLCs by leaving it at level 2-3, if your cpu is stable with droopier load lines. Middle LLCs like 4-5 are generally OK, but "standard" llc (level 2) perform consistently good across different types of work loads (see buildzoid's latest 2021 probinator vids).

2. Just a nitpick - unless Intel messes up and sets VID too high on a golden sample, I doubt any cpu would be truly stable with 0.01mOhm AC LL on stock LLC.

3. That is redundant. Simply leaving vcore at auto nets the same results. Auto (and "auto" adaptive) derives its voltage from the base VID, modified by TVB, AC LL and VRM loadlines. TVB is typically enabled (basically constant), while AC LL and VRM loadlines can be manually set by the user, pretty much making auto voltage "constant" (as in it will perform consistently across reboots).

4. You can opt to directly input AC/DC LL instead of touching SVID behavior. SVID behavior only works on auto AC/DC LL anyway, making it kind of redundant.

5/6. You can directly test AC/DC LL by manually setting the VRM load lines and AC LL (e.g. Level 2 & 0.80mOhm). Afterwards, disable speedstep/speedshift and c-states and it will perform like a static vcore, except with the effects of TVB being present. The main benefit from this is that you don't have an additional step to find the same load vcore.
 

·
Registered
Joined
·
6 Posts
When your computer is idling the CPU multiplier should drop down to 8 with voltages less than 1v. If you saw that voltage drop behavior with full stock bios settings, the power saving features are working with your OS. This points to something with your OC settings is preventing the idle power states from being entered. I'm not having any such problems with my OC, and at best can only wildly guess what's going on with yours (so I won't). If you can't isolate it, start a new topic on your problem; people will see it and respond.
Yeah, it now goes down to like 1.4GHz only, but I'm also quite happy with how it's performing at the moment, in the end I think with a voltage between 1,17V and ~1,3V I'm not gonna harm my CPU. Maybe I'll play with LLC, like ThinbinJim suggests!
 

·
Registered
Joined
·
6 Posts
5/6. You can directly test AC/DC LL by manually setting the VRM load lines and AC LL (e.g. Level 2 & 0.80mOhm). Afterwards, disable speedstep/speedshift and c-states and it will perform like a static vcore, except with the effects of TVB being present. The main benefit from this is that you don't have an additional step to find the same load vcore.
Thanks for your insight on this topic! So disabling speedstep and c-states means I will automatically see the "maximum voltage" at that LLC and AC/DC LL without performing various benchmarks, do I understand correctly? This would mean you'd recommend to disable those 2 settings to find the correct AC/DC LL setting (lower by 0.01, reboot, check HWInfo, repeat), then when it's found, enable them again?
 

·
Registered
Joined
·
18 Posts
Thanks for your insight on this topic! So disabling speedstep and c-states means I will automatically see the "maximum voltage" at that LLC and AC/DC LL without performing various benchmarks, do I understand correctly? This would mean you'd recommend to disable those 2 settings to find the correct AC/DC LL setting (lower by 0.01, reboot, check HWInfo, repeat), then when it's found, enable them again?
No problem! I'm glad that the ACDC method is gaining traction at all. Yes, disabling those settings will allow you to see "true" voltages after the effects of power-saving features are removed. This ensures that you're able to test a given auto/adaptive overclock on equal footing with manual voltage overclocks.

0.01mOhm has an absolutely tiny effect on VID. It's far more practical to lower AC LL (fyi DC LL is only used for package power calculations) by 0.10mOhm per step until you reach instability then incrementing it by 0.05mOhm. But yeah, you got the gist of it.

When using very droopy load lines, watch out for VID hitting the ceiling of 1.52v. The VID system is limited in its voltage request range and caps out at 1.52v, I'm not sure about the implications of exceeding this range. My XI Hero is able to extend VID well into the 1.570v range, but that may just be an Asus z390 feature. If you notice VID clipping at ~1.520v on your board, you'll have to increase the LLC and correspondingly reduce AC LL (fyi DC LL only affects package power calculations).
 

·
Registered
Joined
·
6 Posts
No problem! I'm glad that the ACDC method is gaining traction at all. Yes, disabling those settings will allow you to see "true" voltages after the effects of power-saving features are removed. This ensures that you're able to test a given auto/adaptive overclock on equal footing with manual voltage overclocks.
Yeah, I think there's not a single video talking about it. I don't think it's a good idea to have systems running more than 12h/day on a fixed voltage, yet most people who overclock do that because they are guided to do so :D Also I still don't understand, why "adaptive mode" can't be implemented in an intuitive way, e.g.: input the "true" voltage but set the settings that let it go down when not needed so much voltage. Instead, voltage often goes UP when using adaptive, so nobody uses it ^^

Very nice, the speedstep+c-state trick is super good to know!

0.01mOhm has an absolutely tiny effect on VID. It's far more practical to lower AC LL (fyi DC LL is only used for package power calculations) by 0.10mOhm per step until you reach instability then incrementing it by 0.05mOhm. But yeah, you got the gist of it.
I did it with 0.01 since it was in the original guide but I noticed I had to go down lots of times to get it instable :D Would you also recommend to start at 0.1? Or what is a starting point for AC/DC LL that is safe to use? I have no idea at which setting you could start "breaking things"..

When using very droopy load lines, watch out for VID hitting the ceiling of 1.52v. The VID system is limited in its voltage request range and caps out at 1.52v, I'm not sure about the implications of exceeding this range. My XI Hero is able to extend VID well into the 1.570v range, but that may just be an Asus z390 feature. If you notice VID clipping at ~1.520v on your board, you'll have to increase the LLC and correspondingly reduce AC LL (fyi DC LL only affects package power calculations).
Okay! So a too low LLC can cause VID to go extremely high, interesting. I'm on LLC 4 or 5, will check later and maybe go down a step at a time. I'm also very interested to see what voltages I get without speedstep+c-state, if that'll be lower than my manual OC or higher ^^
 

·
Registered
Joined
·
18 Posts
Yeah, I think there's not a single video talking about it. I don't think it's a good idea to have systems running more than 12h/day on a fixed voltage, yet most people who overclock do that because they are guided to do so :D Also I still don't understand, why "adaptive mode" can't be implemented in an intuitive way, e.g.: input the "true" voltage but set the settings that let it go down when not needed so much voltage. Instead, voltage often goes UP when using adaptive, so nobody uses it ^^

Very nice, the speedstep+c-state trick is super good to know!



I did it with 0.01 since it was in the original guide but I noticed I had to go down lots of times to get it instable :D Would you also recommend to start at 0.1? Or what is a starting point for AC/DC LL that is safe to use? I have no idea at which setting you could start "breaking things"..



Okay! So a too low LLC can cause VID to go extremely high, interesting. I'm on LLC 4 or 5, will check later and maybe go down a step at a time. I'm also very interested to see what voltages I get without speedstep+c-state, if that'll be lower than my manual OC or higher ^^
Hi, a good starting point for ACLL on default LLC (standard/LLC2) is the default ACLL that your board uses on optimized default settings. You can check IA AC LL using HWiNFO64 in summary view -> processor information:
2483254

On my Asus z390 board, it's 1.20mOhm @ LLC 2 while gigabyte z390 boards have it at 1.00mOhm @ LLC standard and the evga z390 dark is 0.80mOhm. As you can see, it varies by manufacturer and possibly by chipset too (not sure). Once you know the starting AC LL, you can work your way down and reduce the ACLL for every round of stability testing.

When increasing LLC from standard/LLC2, you can try lowering IA AC LL by a fair amount like say, 0.10-0.20mOhm per LLC level and comparing load voltages during a heavy load like p95 avx small or AIDA64 fpu only (note that cpu must not throttle!), ensuring that the load voltages are equal between both sets of AC LL/LLC settings.

Also, I worded it wrongly in my previous post. It's the high amounts of IA AC LL in combination with a high base VID (aka target voltage) from the 50x core multiplier that inflates the VID. It'll only get worse the moment you start using cache ratios >43 and cpu ratios >50. When you're using cache ratios higher than default (43x for 9900k), or core ratios higher than 1c boost (50x for 9900k) with adaptive voltage (see shamino's comments), the target voltage can be checked using HWiNFO64:
2483253


Also, I wouldn't expect voltages to be all that different under load. The main advantage of adaptive is the idle voltage reduction from c-states/Speedstep/speedshift and temperature-based voltage reduction from TVB (can also be thought of as a voltage boost under high heat).

Lastly, do keep in mind that the same IA AC LL and VRM Loadline settings will result in different voltages to different cpu samples! This is due to varying base VID between samples, as programmed by intel. So do watch the vcore when playing around with these parameters. As they are variables that work in conjunction with another variable (base VID), there is no universal 'safe' AC LL and LLC settings.
 

·
Registered
Joined
·
6 Posts
Hi, a good starting point for ACLL on default LLC (standard/LLC2) is the default ACLL that your board uses on optimized default settings. You can check IA AC LL using hwinfo64 in summary view -> processor information. On my z390 board, it's 1.20mOhm @ LLC 2 while gigabyte z390 boards have it at 1.00mOhm @ LLC standard and the evga z390 dark is 0.80mOhm. As you can see, it varies by manufacturer and possibly by chipset too (not sure). Once you know the starting AC LL, you can work your way down and reduce the ACLL for every round of stability testing.
Okay, alright, thanks a lot! So starting at 0.1 how *AcidBath* suggested is way too low then?
Lastly, do keep in mind that the same IA AC LL and VRM Loadline settings will result in different voltages to different cpu samples! This is due to varying base VID between samples, as programmed by intel. So do watch the vcore when playing around with these parameters. As they are variables that work in conjunction with another variable (base VID), there is no universal 'safe' AC LL and LLC settings.
Yeah of course! I was just interested to understand what range of values are "normal" ;) Ballpark numbers..
 
1 - 20 of 56 Posts
Top