Newbie question about undervolting.

[jansimek]

Hello guys! so i've had my 9900k for over a month now and decided to undervolt. At stock settings with MCE off, the vcore under load is anywhere between 1.25-1.35v with occassional spikes up to 1.37. I did some digging and from what i understand, this is still completely fine and within specs, my temps are under control, too. However, i dont think this much voltage is necessary for 4,7ghz all core load, so i decided to undervolt. I have been playing with static vcore a couple weeks ago and was getting whea errors running aida64 up until i set the static vcore in bios to 1.190 so im thinking its safe to assume that my 9900k will need roughly 1.2V to do stock settings?

I dont want a static vcore though, so i did some more digging and found about a DVID option on my motherboard (Gigabyte z390 Aorus Master) and from what i understand, this setting should limit the maxium allowed vcore voltage supplied by the mobo while staying dynamic.

if i put a value of -0.150V in the DVID, that would mean the vcore limit would then be around 1.22v considering it creeps up to as high as 1.37 on auto vcore, right?

is this way of undervolting safe for long term use? i dont plan on overclocking right now, the processors performance is more than enough for me.

bios settings :

MCE : off
XMP 1 : ddr4 @2666 cl13 1.35v (creeps up to about 1.38 occassionaly which is fine i suppose)
Turbo Boost : enabled running at stock settings
Hyperthreading : enabled
Vcore : normal
DVID : -0.150v

 

[Besty]

Personally, I have used two options:

Option 1. LLC Turbo + Vcore manual set to 1.2v (or lower if good chip) most C-States enabled which will undervolt the chip as the turbo boost goes up and down.

Option 2. Leave all settings at stock and set a TDP Power limit again, with C-States enabled.

I would never use auto-vcore.

 

[jansimek]

I would never use auto-vcore.

I didnt like the auto vcore either because it overvolts. Would you still consider normal vcore mode with undervolt via the DVID offset auto? i am really green when it comes to overclocking stuff.

Been running latest version of p95 small ffts on all 16 threads and the cpu sat between 1.104 - 1.116v at the all core 4,7 boost. Also tried running the same test on only 2 threads to get the cpu to the stock 5ghz (2core max) turbo, in that case it sits at 1.130 - 1.150

Would you consider running at stock turbo behavior at those voltages safe when 100% stable? i will definitely do some more prolonged testing. Is this a case of "if it works, it works?"

(temps are not an issue at all, but its nice to see a 10c improvement while keeping the same clocks and performance)

 

[Jason Rage]

First things first when using adaptive what is the stock voltage? You use the offset to + or - from that. I.e, I know my stock voltage on my 8600k is 1.165. Overclocking using static voltage I found that 1.34v was the voltage I needed to run stable @5Ghz. So, if I want to use adaptive I take 1.34v minus 1.165v gives me my offset of +0.175 This allows me to draw voltage anywhere from 1.165v up to 1.34v.

In your case you would do the same but instead of finding your highest static stable you will be finding your lowest. Start with you normal voltage and remove 00.05v a time and run a better benchmark like IntelBurnTest or Prime95, aida64 is to weak. Keep lowering until you become unstable. Once you find that point add back 0.010v and that will be your static stable lowest voltage. Then you can work out your minus offset from that static number.

Leave all C states alone. I use High LLC, yes I might vdroop a bit more but Turbo seems to give me more Vboost. I like you do not like seeing say a sudden spike from 1.34v to 1.44v.

Lastly to control the vdroop/vboost a bit more set IA AC 1/IA DC 1. *The bellow link mentions leaving IA DC 0, this might apply to you more*

Here is a link with your board & 9900k. Might find it helpful. https://www.overclock.net/forum/6-i...78-gigabyte-z390-aorus-owners-thread-142.html

 

[jansimek]

At stock settings with MCE off, the vcore under load is anywhere between 1.25-1.35v with occassional spikes up to 1.37. I did some digging and from what i understand, this is still completely fine and within specs, my temps are under control, too. However, i dont think this much voltage is necessary for 4,7ghz all core load, so i decided to undervolt. I have been playing with static vcore a couple weeks ago and was getting whea errors running aida64 up until i set the static vcore in bios to 1.190 so im thinking its safe to assume that my 9900k will need roughly 1.2V to do stock settings?

if i put a value of -0.150V in the DVID, that would mean the vcore limit would then be around 1.22v considering it creeps up to as high as 1.37 on auto vcore, right?

First things first when using adaptive what is the stock voltage? You use the offset to + or - from that. I.e, I know my stock voltage on my 8600k is 1.165. Overclocking using static voltage I found that 1.34v was the voltage I needed to run stable @5Ghz. So, if I want to use adaptive I take 1.34v minus 1.165v gives me my offset of +0.175 This allows me to draw voltage anywhere from 1.165v up to 1.34v.

if i confirm that im stable at 1.2 manual vcore with other software such as what you recommended and newest prime 95, are my calculations correct? the board says stock is 1.2v in the vcore setting even when its on auto, but as i said it just creeps up as high as to 1.37

Thank you for the tips.

 

[Besty]

if i confirm that im stable at 1.2 manual vcore with other software such as what you recommended and newest prime 95, are my calculations correct? the board says stock is 1.2v in the vcore setting even when its on auto, but as i said it just creeps up as high as to 1.37

Thank you for the tips.

are you taking your voltage reading from ‘vout’ using hwinfo? if not download hwinfo and use that reading to gauge the voltage.

some bios settings if left to adaptive will avx boost the VID automatically in order to help with stability but that does not necessarily mean that the voltage the cpu is being fed has gone up. ‘vout’ on the other hand reads the true voltage feed

i would not use prime to test first, use realbench first, then cinebench20 multi, then use superpi32m then finally use prime.

prime is really a worst case scenario programme, it’s good to test your cpu and cooling with but you need to build up to it.

 

[Jason Rage]

if i confirm that im stable at 1.2 manual vcore with other software such as what you recommended and newest prime 95, are my calculations correct? the board says stock is 1.2v in the vcore setting even when its on auto, but as i said it just creeps up as high as to 1.37

Thank you for the tips.

You get that from your Bios I.e, when you first booted it up, you was at 1.2v that is the default voltage for your cpu/board combo, not software. From that number use static to keep shaving off 0.005 a time until you become unstable. If using prime 95 AVX instructions just let it pass 2 cycles a time, 12k-8k. When you think you have your lowest stable then let it run awhile. If all seems good then work out your offset from that and off you go!

Just read through some of the comments on the link I gave you and some people are stating that IA DC 1 is better. Read through it some and see for yourself.

 

[jansimek]

are you taking your voltage reading from ‘vout’ using hwinfo? if not download hwinfo and use that reading to gauge the voltage.

vr vout peaks at 1.318 at auto vcore, 1.178 at a manual static vcore of 1.2 and at 1.180 when using the -0.150 dvid offset.

You get that from your Bios I.e, when you first booted it up, you was at 1.2v that is the default voltage for your cpu/board combo, not software.

at completely stock settings, when i boot into bios, it sits at 1.284 even though the auto vcore setting says 1.2v next to it.

 

[Jason Rage]

Okay, lets break this down. Even with a static voltage you will get spikes, I.e, 1.2v then a spike of 1.3v etc.. This is called Vboost. LLC's effect this. That's why I said to use high and to set IA DC 1/IA AC 1. This will help control that. Even so you will always get a certain amount of both no matter what you do. It happens when a sudden load is placed on the system, it compensates to remain stable by adding extra volts.

Overclockers tend to set LLC to Turbo and run static to try and remain stable at higher volts. You do not want a sudden massive vdroop (You want 1.34v but instead get 1.2v making you unstable) when a sudden load hits the system.

You are undervolting. You do not need maximum power all the time trying to keep a tight overclock stable.

As to the stock voltage. Don't remember their being 2, unless your talking about the little slider box thingy. I just went of my Vcore in "Advanced Voltage Control" on an Aorus Gaming 5. So if that is showing 1.2v just go of that.

Keep dropping 0.005 and testing. It will take awhile but you will eventually be left with your lowest possible volt value for your cpu. Every cpu is different and this is the way if you want to reach your lowest stable volt on your cpu.

 

[jansimek]

As to the stock voltage. Don't remember their being 2, unless your talking about the little slider box thingy. I just went of my Vcore in "Advanced Voltage Control" on an Aorus Gaming 5. So if that is showing 1.2v just go of that.

Keep dropping 0.005 and testing. It will take awhile but you will eventually be left with your lowest possible volt value for your cpu. Every cpu is different and this is the way if you want to reach your lowest stable volt on your cpu.

This is what i mean by the stock voltage its showing in the advanced voltage control : https://imgur.com/a/88iXWfM it says 1.2v next to the auto box where manual vcore is applied so i guess thats the same thing you just said? and the "slider box thingy" is where the current vcore is being reported (i think) thats where i saw the 1.284 in bios.


Guys, thank you very much for your time, i appreciate it! will report back when i figure this out

 

[Jason Rage]

Yes! So 1.2v it is then! Lastly remember that if you change your cpu clock ratio then you will have to tinker again. I saw you mentioned 4.7Ghz (47 cpu clock ratio) so you will be finding your lowest for that. If you decide you want to set your cpu clock ratio to say 48 for 4.8Ghz then you would do the same adding 0.005 volt until stable. Once you have that number work out your offset again.

I had a headache myself doing adaptive lol

Good luck!

 

[jansimek]

Lastly remember that if you change your cpu clock ratio then you will have to tinker again. I saw you mentioned 4.7Ghz (47 cpu clock ratio) so you will be finding your lowest for that. If you decide you want to set your cpu clock ratio to say 48 for 4.8Ghz then you would do the same adding 0.005 volt until stable.

Yeah im undervolting for a 47x all core clock ratio which is stock max all core turbo, but the cpu will still be boosting up to 50x when only up to 2 cores are active, 49x up to 3 active cores, 48x up to 5 cores active and 47x from there

my last question : when i find the perfect lowest voltage for the all core 47x ratio, will it cause instability in instances where less cores are working, but are at a higher frequency? Im assuming that should be fine

 

[Jason Rage]

Last question : when i find the perfect lowest voltage for the all core 47x ratio, will it cause instability in instances where less cores are working, but are at a higher frequency? Im assuming that should be fine

Okay here is the problem. If just playing with adaptive+offset you can pass benchmarks all day but crash at idle because your voltage is to low. So using static we rule out that problem for now. Booting into Windows, starting a stress test will reveal the instability and allow us to +or- accordingly. What we have to be careful about is having to low a voltage at idle. That's why we do it nice and slow with 0.005 increments static. Once we have our lowest stable static we should not have a problem with 5Ghz as we will be offsetting from 1.2v. So at idle volts will lower but when put under stress we can use up to 1.2v as that is the starting point for offset.

As Besty said, far easier just to go his route. All you need to do then is set the volt you want and run two benchmarks. One to run at 47x and one to run at 50x 2 cores.

 

[jansimek]

Thank you once again. Have a nice day!

 

[Jason Rage]

Just a heads up, try what I said by all means but looking more into the 9900k can see that it works slightly differently than the previous generations.

As with all these things there seems to be different opinions but this looks like solid information. https://rog.asus.com/forum/showthre...voltage-mode-on-Z390-Maximus-XI-Formula-9900k He's using a different board but its what Shamino replied that is of particular note.


I'm not experienced with the 9900k and thought it would be the same as say an 8700k. Reading through the above forum I'm not the first to make this mistake.

As said, if it does not pan out have a read through the above forum post and try following Besty advice who might have more experience with the 9900k than I do. Who knows though, I have read elsewhere doing what I suggested and step lowering.

Problem seems to be with offsets and adaptive. The general census is "static" is better, but then again that has been said forever. Adaptive does work but might need a different setting or 2 on the 9900k that I do not have first hand knowledge of.

Sorry if it turns out my way is trash Wish you all the best and good luck!

 

[jansimek]

That is interesting, i either dont understand this at all (very likely) or my board doesnt behave like what Shamino described in that thread.

Anyway, been playing with the voltage a bit more and found out that it starts throwing whea errors and becomes unstable once i set the DVID to -0.165. It still seemed to run good at -0.160, but i set it to -0.155 just to give it a tiny bit more juice.

So right now, with basically everything in bios at stock except for a xmp profile, mce off, vcore set to "normal" and a DVID of -0.155 it passed this so far :


10 min p95(ver. 29.8) small fft all 16 threads

15min p95(ver. 29.8) blend test

10min aida64 system stability test (cpu, fpu, cache, memory)

3 cinebench r20 runs



Minimum Vcore recorded in HWINFO is 0.720 at idle and the max is 1.200v

VR VOUT shows values of 0.754 minimum and a peak of 1.184

The cpu package peaked at 80c, which is 15-20c less than i would have at "auto" vcore during a small fft run on a warm day.

I am happy with this and will monitor how it behaves in the next few days, hope i dont crash at idle or while web browsing.



Would there be a reason NOT TO keep it set like this if im 100% satisfied and it ends up working under all circumstances? (question for anyone experienced in the OC mastery who passes by and is willing to spend a minute )

 

[Jason Rage]

That looks perfect! Run that 24/7

Have a read of this https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review/21. Notice the first thing talked about was changing boards (Notice: When we initially posted this page, we ran numbers with an ASRock Z370 board. We have since discovered that the voltage applied by the board was super high, beyond normal expectations. We have since re-run the numbers using the MSI MPG Z390 Gaming Edge AC motherboard, which does not have this issue.

In one way or another all board manufactures do this. They want the end user to be happy with fast results. As all cpu's are different they have a pretty wide margin of "what is needed". All they care about is great benchmarking results, no one wants to read about how power conservative a board is.

All you have done is tighten that margin for what your cpu needs & not what is needed to run every cpu as fastest as possible regardless of the silicon lottery.

Lastly VR VOUT is your correct reading, its the new more accurate one replacing the old vcore. I unfortunately have a Z370 board that does not have this sensor that was introduced to Gigabyte boards in the Z390.

If it turns out that my advice was not total turd after all then enjoy your less voltage happy and cooler cpu!

 

[jansimek]

Okay, so after some more testing i'd love some opinions on whether or not i should consider this stable and stop tweaking.


Ended up manually changing the cpu clock ratio to 47.
MCE OFF
Turbo Boost OFF
XMP 1
and a DVID offset.

-0.160 error 5m into p95(version 29.8) small fft (worker error & stop +a whea error)

-0.155 error 12m into p95(version 29.8) small fft (worker error & stop +a whea error)

-0.150 error 30m min into p95(version 29.8) small fft (worker error & stop +a whea error)

-0.140 error 1h 20m into p95(version 29.8) small fft (this time only the worker stop & error happened with no whea log in windows. Sign of getting closer to 100% stable?)

-0.130 passed 3hours 48minutes p95(version 29.8) small fft with no errors. Cpu package max was 86c after almost 4 hours of what is the absolutely worst case scenario, if i understand the prime version differences right? im assuming that running the 26.6 non AVX version after this is pointless as that could probably go on forever.

Vcore under the newest p95 small fft test load im running is between 1.116 - 1.128 with the latest -0.130 DVID offset.



is this 100% stable?


Edit : i had no issues, errors, crashes or anything of that sort running games such as battlefield V and other stress tests such as Aida64, Realbench, Cinebench even at an offset of -0.150. It's just the P95 that kept producing errors up until i bumped it up to -0.130.

 

[Jason Rage]

Yes you are right. You would only need to run the non AVX prime if you was running an AVX offset or testing the 2 core boost.

Prime is a bugger to run and as you rightly say the worst case scenario. However, if your passing with no errors your pretty much guaranteed a 100% stable system. So yes, your passed prime your stable!

Some people do not even use Prime as its so demanding and unrealistic. Do not let this overly worry you as prime is becoming a bit antiquated and just dam right over the top. Most people eventually settle on an AVX offset but I do not like this. I found that if I'm overclocked to 5ghz with an AVX offset of 2 literally all programs/games use, or at least your Bios believes its using, some form of AVX and just downclocks my cores to 4.8ghz all of the time anyway. So why bother with the extra volts & heat for something I get only on desktop?

So there only really 3 options, keep you -0.130 for a guaranteed 100% rock solid underclock, use the middle ground and use -0.150 what in all probability will be fine, only time will tell. Lastly set an AVX offset of 1 and test again.

Me personally? I would try your 0.150 and give it some time. If you crash on gaming drop 5 a time, -0.150-0.145 etc.. until you are rock solid again. Save your -0.130 in Bios just in case your ever do some very intensive cpu work that only takes a few seconds to switch to for security. I have 3 different settings that I switch between depending on what I am doing.

As to your voltage its very good! I need 1.33v to run Prime AVX 100% stable @4.8ghz on my 8600k.

 

[jansimek]

Thank you for all the information and helping me understand Jason! i appreciate it.