9900K started needing more voltage for high OC, anything I can do?

[TheNaitsyrk]

Hai demons, it's ya boy again.

 

I recently added GPUs to my loop. Before I had the loop, CPU had 360mm and 240mm attached to just it.

 

I did that again because I wanted to test 5.3Ghz properly. But instead of wanting 1.375-1.4V for 5.3Ghz, it started wanting 1.425V and it's only stable in CPU-Z CPU Benchmark.

Cinebench R15 (according to core temp program) eats about 228W before I get a blue screen. The loop is still able to cool it, but barely at about 90C. Still crashes though.

I remember it being stable in TimeSpy also.

 

CPU only wants 1.28V for 5.0Ghz OC.

 

I'm using AORUS Master Z390 motherboard. Hottest point on the motherboard was 77C when running benchmarks.

 

Settings I'm using for 5.0Ghz:

LLC: Turbo

Voltage: 1.28V

Settings I'm using for 5.3Ghz:

LLC: Extreme

Voltage: 1.425V

 

It is cooled below it's TJ Max but still crashes when I give it 1.425V in Cinebench. 

 

Did I actually loose the lottery / I can't get it to 5.3Ghz stable any more?

[flibberdipper]

Welcome to the game of overclocking my guy. Some chips lose their skillz after you initially get them (though those high core temps sure can't be helping your case).

 

I know the pain of degradation too (G3258 went from doing 4.8/1.26 to walling at 4.6/1.3 in about two months).

[TheNaitsyrk]

7 minutes ago, flibberdipper said:

Welcome to the game of overclocking my guy. Some chips lose their skillz after you initially get them (though those high core temps sure can't be helping your case).

 

I know the pain of degradation too (G3258 went from doing 4.8/1.26 to walling at 4.6/1.3 in about two months).

Oh dang, really? I ACTUALLY did not know this and I'm building for many years now.

 

Really annoying as I wanted to try TimeSpy again with water cooled cards as they were air cooled before so I went to look into OC again and boom, it wants more voltage than before, but 5.0Ghz behaves the same weirdly.

 

I initially thought I had a really good chip as it only wanted 1.28V for 5.0Ghz, and now it just says "NOPE".

 

Oh yea, you're right: https://www.overclock.net/forum/5-intel-cpus/1673657-cpu-overvolt-death-degradation-stories.html

 

Not that I pushed it hard at all, just strange that it happened to me.

[Gix7Fifty]

5 minutes ago, TheNaitsyrk said:

Oh dang, really? I ACTUALLY did not know this and I'm building for many years now.

 

Really annoying as I wanted to try TimeSpy again with water cooled cards as they were air cooled before so I went to look into OC again and boom, it wants more voltage than before, but 5.0Ghz behaves the same weirdly.

 

I initially thought I had a really good chip as it only wanted 1.28V for 5.0Ghz, and now it just says "NOPE".

Im having the same problem, my OC was acting up last night 5Ghz@1.265v. Crashed twice when I was playing BF V last night. 

[TheNaitsyrk]

1 minute ago, Gix7Fifty said:

Im having the same problem, my OC was acting up last night 5Ghz@1.265v. Crashed twice when I was playing BF V last night. 

You have a similar board. Have you tried going past 5.0Ghz?

 

Also, what voltages do your CPU needs to get like 5.2Ghz?

 

Also, what LLC you run? Ultra / Extreme or Level 7?

[flibberdipper]

9 minutes ago, TheNaitsyrk said:

Oh dang, really? I ACTUALLY did not know this and I'm building for many years now.

 

Really annoying as I wanted to try TimeSpy again with water cooled cards as they were air cooled before so I went to look into OC again and boom, it wants more voltage than before, but 5.0Ghz behaves the same weirdly.

 

I initially thought I had a really good chip as it only wanted 1.28V for 5.0Ghz, and now it just says "NOPE".

 

Oh yea, you're right: https://www.overclock.net/forum/5-intel-cpus/1673657-cpu-overvolt-death-degradation-stories.html

 

Not that I pushed it hard at all, just strange that it happened to me.

Yep. In my fairly limited experience it seems like "modern" (aka 775 and up) Intel chips like to shit themselves a lot faster than FX and earlier AMD chips did. Thank god graphics cards seem to be a bit more resilient against this, though i suppose it helps that their voltage is pretty locked down.

[TheNaitsyrk]

2 minutes ago, flibberdipper said:

Yep. In my fairly limited experience it seems like "modern" (aka 775 and up) Intel chips like to shit themselves a lot faster than FX and earlier AMD chips did. Thank god graphics cards seem to be a bit more resilient against this, though i suppose it helps that their voltage is pretty locked down.

Yeah, Nvidia is notorious for this.

 

I'm wanting to use LLC level 7. Is that healthy? What do you think? I believe it will allow me to drop voltages slightly.

[Gix7Fifty]

13 minutes ago, TheNaitsyrk said:

You have a similar board. Have you tried going past 5.0Ghz?

 

Also, what voltages do your CPU needs to get like 5.2Ghz?

 

Also, what LLC you run? Ultra / Extreme or Level 7?

Haven't messed with above 5Ghz, what do you think is the max safe voltage?

I set to:

Loadline Turbo

Package Power Limit 2 and 1(watts) 4090

Core current limin (Amps) 255

[flibberdipper]

17 minutes ago, TheNaitsyrk said:

Yeah, Nvidia is notorious for this.

 

I'm wanting to use LLC level 7. Is that healthy? What do you think? I believe it will allow me to drop voltages slightly.

Honestly can't tell you. I never really messed with LLC outside of setting it to whatever was recommended to me forever ago.

[TheNaitsyrk]

11 minutes ago, flibberdipper said:

Honestly can't tell you. I never really messed with LLC outside of setting it to whatever was recommended to me forever ago.

Ah I see, regardless thanks!

 

17 minutes ago, Gix7Fifty said:

Haven't messed with above 5Ghz, what do you think is the max safe voltage?

I set to:

Loadline Turbo

Package Power Limit 2 and 1(watts) 4090

Core current limin (Amps) 255

1.4V being max safe. I wouldn't go past that not even 1 mv.

[bobhumplick]

i know this is old but if anybody is planning on using these settings keep some things in mind.  

 

first there is a voltage drop when a cpu is under load.  just like if you run a power tool and the lights dim a bit or if you have  a high powered stereo in your car and the bass hits and the lights dim a bit.  this is natural.  some people claim its built in to protect the cpu but thats not exaclty it.  it just happens to work that way with circuits

 

but mobo makers added in ciruits that would raise the voltage under load.  so by the time you drop some voltage, then the mobo raises it, things even out.  this is called llc.

 

so the mobo opens the "valves" for more power when under load.  the "valves" it opens under load for a 1.35v setting might be the same "valves" it opens when you would have 1.43v set when not under load.  that basically means that if you use 1.35 with a flat llc (the voltage stays the same, on z390 gigabyte boards its turbo) then you are actually tell the board to raise voltage from 1.35 to 1.42v.  that is not good if you have something high set like 1.4 or higher!!!!.  and never use extreme llc.  there isnt much point in it.  

 

if you use a "flat"  llc (voltage doesnt drop but doesnt raise above what you set either) then i wouldnt go above about 1.32.  measuring at the back of the socket on an asrock z370 with an 8700k and later a 9900k, the flat (highest on asrock and turbo on gigabyte) llc setting actually raised the voltage to 1.39 when i set a voltage of 1.31 in the bios.  on that board if you set 1.4v and the flat llc then you would get about 1.48 at least.  THIS IS NOT GOOD!!

 

when you use llc you have to set your voltage much much lower.  you would probalby be better off just not using llc or using the one 1 or 2 levels below "flat"

 

THIS IS ALSO VERY IMPORTANT!!!  the llc level below flat often overclocks BETTER than flat or something insane like extreme llc.  i could run 1.315 at 5.1ghz on my old 8700k with llc set to the one just below flat (on asrock its level 2 with level 1 being highest i think) and it would be stable even though it was drooping down to 1.26 or so.  but if i used flat (level 1, like i thought you were supposed to) i would have to set the voltage to 1.32 and it would not droop much at all..  1.315 would droop to 1.31 or something and still not be stable, even though i just ran the chip stable at 1.26 or so.  i dont know why this is so but its true.  tried on 2 different boards with 2 different chips and was true both times.

 

flat llc should probably not go above 1.32 for daily use.  that might be a bit high.  if you use extreme or anything ABOVE flat then i wouldnt go above 1.28 maybe less than that.  you would be better off going high llc (level 2) or even lower and just setting the voltage to 1.4v at max but better would be soemthing a bit less.

 

people say as long as you can cool these chips then it doesnt matter about the voltage because you will run into heat problems first.  but that is for large aircoolers or aio's under 240mm or so.  large custom loops run at max pump speeds will put you into dangerous territory with voltage without running into thermal throttling.  the guy in this post has either cpu degradation or the board's vrms have degradated.  dont do this.  

 

this guy should have just went for 5.1ghz and a nice low voltage.  maybe 5.2 with flat or below llc and maybe 1.35v max.  avx offsets are your friend and a 5.2ghz clock with an avx offset of 2 so it ran at 5ghz probalby could have been gotten on a 1.32v setting with llc just below flat.  dont do this!!!!!

 

measure your voltage at the back of the socket some time.  1.3v with llc extreme will give you 1.45v at least at the socket even if it reads much less.  like i said 1.315 would read near 1.4v just on flat llc (gigabyte turbo asrock level 1).  if you dont have any vdroop with your settings or if it actually raises voltage (negative droop) then about 1.3 is getting close to max voltage.

[Falkentyne]

On 12/23/2018 at 2:01 PM, TheNaitsyrk said:

Hai demons, it's ya boy again.

 

I recently added GPUs to my loop. Before I had the loop, CPU had 360mm and 240mm attached to just it.

 

I did that again because I wanted to test 5.3Ghz properly. But instead of wanting 1.375-1.4V for 5.3Ghz, it started wanting 1.425V and it's only stable in CPU-Z CPU Benchmark.

Cinebench R15 (according to core temp program) eats about 228W before I get a blue screen. The loop is still able to cool it, but barely at about 90C. Still crashes though.

I remember it being stable in TimeSpy also.

 

CPU only wants 1.28V for 5.0Ghz OC.

 

I'm using AORUS Master Z390 motherboard. Hottest point on the motherboard was 77C when running benchmarks.

 

Settings I'm using for 5.0Ghz:

LLC: Turbo

Voltage: 1.28V

Settings I'm using for 5.3Ghz:

LLC: Extreme

Voltage: 1.425V

 

It is cooled below it's TJ Max but still crashes when I give it 1.425V in Cinebench. 

 

Did I actually loose the lottery / I can't get it to 5.3Ghz stable any more?

 

The CPU degraded at 5.3 ghz because LLC was set to extreme at this voltage.

This is what caused problems.

That voltage would have only been safe at LLC "High".

LLC Turbo is only long term safe at most amp loads up to 1.35v set in the BIOS.

Yes I know I'm replying to a 6 month old thread, but I wasn't the one who bumped it

[TheNaitsyrk]

6 minutes ago, Falkentyne said:

 

What settings did you previously use?

Did you use 5 ghz long-term or did you actually use 5.3 ghz?

And how long was it at this speed and voltage?
And did you **IN THE PAST** actually use LLC EXTREME at 5.3 ghz?!?!?!?!  (And for how long?)

 

I need to know the answer, please.

 

Just for benchmarking purposes. I see 0 difference between 5.0Ghz and 5.3Ghz in games and I'm runnning 2080 Ti SLI. I don't even OC the GPUs now to be fair.

 

CPU is hard to kill even with a bit more voltage. I think the lifespan is gonna go down from maybe 7 years to 4? That's still plenty at that premium speed.

 

I could've gotten away with less voltage at 5.3Ghz IF it's just games.

 

Also, my CPU did not degrade I just used wrong settings that time.

 

I run my 9900K at stock currently, as I play The Witcher at 10320x1440 without anything OCed at high and I get around 150 FPS. (120FPS cap set on mostly to match refresh rate).

[Falkentyne]

13 minutes ago, bobhumplick said:

i know this is old but if anybody is planning on using these settings keep some things in mind.  

 

first there is a voltage drop when a cpu is under load.  just like if you run a power tool and the lights dim a bit or if you have  a high powered stereo in your car and the bass hits and the lights dim a bit.  this is natural.  some people claim its built in to protect the cpu but thats not exaclty it.  it just happens to work that way with circuits

 

but mobo makers added in ciruits that would raise the voltage under load.  so by the time you drop some voltage, then the mobo raises it, things even out.  this is called llc.

 

so the mobo opens the "valves" for more power when under load.  the "valves" it opens under load for a 1.35v setting might be the same "valves" it opens when you would have 1.43v set when not under load.  that basically means that if you use 1.35 with a flat llc (the voltage stays the same, on z390 gigabyte boards its turbo) then you are actually tell the board to raise voltage from 1.35 to 1.42v.  that is not good if you have something high set like 1.4 or higher!!!!.  and never use extreme llc.  there isnt much point in it.  

 

if you use a "flat"  llc (voltage doesnt drop but doesnt raise above what you set either) then i wouldnt go above about 1.32.  measuring at the back of the socket on an asrock z370 with an 8700k and later a 9900k, the flat (highest on asrock and turbo on gigabyte) llc setting actually raised the voltage to 1.39 when i set a voltage of 1.31 in the bios.  on that board if you set 1.4v and the flat llc then you would get about 1.48 at least.  THIS IS NOT GOOD!!

 

when you use llc you have to set your voltage much much lower.  you would probalby be better off just not using llc or using the one 1 or 2 levels below "flat"

 

THIS IS ALSO VERY IMPORTANT!!!  the llc level below flat often overclocks BETTER than flat or something insane like extreme llc.  i could run 1.315 at 5.1ghz on my old 8700k with llc set to the one just below flat (on asrock its level 2 with level 1 being highest i think) and it would be stable even though it was drooping down to 1.26 or so.  but if i used flat (level 1, like i thought you were supposed to) i would have to set the voltage to 1.32 and it would not droop much at all..  1.315 would droop to 1.31 or something and still not be stable, even though i just ran the chip stable at 1.26 or so.  i dont know why this is so but its true.  tried on 2 different boards with 2 different chips and was true both times.

 

flat llc should probably not go above 1.32 for daily use.  that might be a bit high.  if you use extreme or anything ABOVE flat then i wouldnt go above 1.28 maybe less than that.  you would be better off going high llc (level 2) or even lower and just setting the voltage to 1.4v at max but better would be soemthing a bit less.

 

people say as long as you can cool these chips then it doesnt matter about the voltage because you will run into heat problems first.  but that is for large aircoolers or aio's under 240mm or so.  large custom loops run at max pump speeds will put you into dangerous territory with voltage without running into thermal throttling.  the guy in this post has either cpu degradation or the board's vrms have degradated.  dont do this.  

 

this guy should have just went for 5.1ghz and a nice low voltage.  maybe 5.2 with flat or below llc and maybe 1.35v max.  avx offsets are your friend and a 5.2ghz clock with an avx offset of 2 so it ran at 5ghz probalby could have been gotten on a 1.32v setting with llc just below flat.  dont do this!!!!!

 

measure your voltage at the back of the socket some time.  1.3v with llc extreme will give you 1.45v at least at the socket even if it reads much less.  like i said 1.315 would read near 1.4v just on flat llc (gigabyte turbo asrock level 1).  if you dont have any vdroop with your settings or if it actually raises voltage (negative droop) then about 1.3 is getting close to max voltage.

This is actually wrong.

The "Lowest" (Flat loadline) LLC ON your Asrock does *NOT* raise the voltage at all at load !

Elmor and another Asrock user actually tested this with Mode 1 and Mode 2, with mode1 being the completely flat LLC.

They were able to measure more accurately, but there was a LOT of noise as it wasn't properly grounded and I am still not sure if they were measuring VCC_Sense or MLCC caps!

 

The Socket MLCC is what you measured yourself, and this is actually equal to the Gigabyte "ITE 8792E" measurement and the Gigabyte onboard read points.  This reads "flat" at LLC=Turbo on the Gigabyte (which we all know is wrong since LLC Turbo is 0.4 mOhms of vdroop on a 8 core CPU and 0.52 mOhms of vdroop on a 6 core CPU.  

Here is what you were actually looking at with your measurements:

 

https://www.overclock.net/forum/27686004-post2664.html

 

Just for reference, if you are using a true "Flat" LLC (measured via VCC_SENSE)--VR VOUT is an accurate VCC_Sense measurement but VR VOUT does NOT show transient spikes or drops, and transients will increase with tigher loadline!, you should *NOT* exceed 1.28v set in BIOS unless you are under sub-ambient cooling and if you are attempting to run AVX prime95, you should *NOT* exceed 1.24v set in BIOS!  That's because at 193 amps of current (8 core), max VCC_Sense voltage (with respect to ground), with 193 amps being absolute max allowed by Intel, is between 1.212v-1.230v (depending on where it's measured).

 

 

[Falkentyne]

21 minutes ago, TheNaitsyrk said:

Just for benchmarking purposes. I see 0 difference between 5.0Ghz and 5.3Ghz in games and I'm runnning 2080 Ti SLI. I don't even OC the GPUs now to be fair.

 

CPU is hard to kill even with a bit more voltage. I think the lifespan is gonna go down from maybe 7 years to 4? That's still plenty at that premium speed.

 

I could've gotten away with less voltage at 5.3Ghz IF it's just games.

 

Also, my CPU did not degrade I just used wrong settings that time.

 

I run my 9900K at stock currently, as I play The Witcher at 10320x1440 without anything OCed at high and I get around 150 FPS. (120FPS cap set on mostly to match refresh rate).

I actually deleted and edited that post as you were replying as I realized that it was a 6 month old post.

But anyway, here you go.

 

1.40v with LLC Extreme definitely caused degradation on the 9900K.  And trying to increase voltage to 1.425v with LLC Extreme will just cause more degradaton faster.

LLC Extreme is precisely a 0.2 mOhm loadline slope.  Knowing this, we can calculate the maximum safe VR VOUT (CPU on-die sense voltage or VCC_Sense) at different amp loads.

To do this, we need to use max VID (1.52v) and max amps (193 amps) for 8 cores, and then use the default loadline (1.6 mOhms--this is milliOhms), using Ohm's law to calculate vdroop, converting any volts to millivolts, since the other unit is in milliOhms.  Loadline calibration is DEFAULT (standard) here (1.6 mOhms!).

 

193 amps: 1520 - (1.6 * 193) (milliOhms (oops, wrote millivolts there, before, sorry, edited now) * amps is equal to vdroop) = 1.211v max VR VOUT 

 

150 amps (typical cinebench R20 or Realbench 2.56 power draw at 5 ghz-5.1 ghz):

1520 - (1.6 * 150) = 1.280v max VR VOUT.

 

100 amps (typical light to medium gaming loads):

1520 - (1.6 * 100) = 1.360v max VR VOUT.

 

Etc etc.

 

Now, since you are using a custom loadline calibration, this throws off the entire specification. (0.2 mOhms)

Assuming your stress test was drawing 150 amps (1280mv max safe voltage here):

 

1400 - (150 * 0.2) = 1.370v.

So you were pulling 1.370v on VR VOUT when 1.280v should have been the limit, at that power draw.

Increasing to 1.425v would have just made things even worse here.  This is still ignoring the issue of transient spikes which would have been pretty huge here also (and not detectable on HWinfo sensors; an oscilloscope is needed).

 

For 150 amps, Loadline Calibration=High would have been safe at 1.40v set in the BIOS, since LLC High is 0.8 mOhms:

1400 - (0.8 * 150) = 1.280v !

[TheNaitsyrk]

2 hours ago, Falkentyne said:

I actually deleted and edited that post as you were replying as I realized that it was a 6 month old post.

But anyway, here you go.

 

1.40v with LLC Extreme definitely caused degradation on the 9900K.  And trying to increase voltage to 1.425v with LLC Extreme will just cause more degradaton faster.

LLC Extreme is precisely a 0.2 mOhm loadline slope.  Knowing this, we can calculate the maximum safe VR VOUT (CPU on-die sense voltage or VCC_Sense) at different amp loads.

To do this, we need to use max VID (1.52v) and max amps (193 amps) for 8 cores, and then use the default loadline (1.6 mOhms--this is milliOhms), using Ohm's law to calculate vdroop, converting any volts to millivolts, since the other unit is in milliOhms.  Loadline calibration is DEFAULT (standard) here (1.6 mOhms!).

 

193 amps: 1520 - (1.6 * 193) (milliOhms (oops, wrote millivolts there, before, sorry, edited now) * amps is equal to vdroop) = 1.211v max VR VOUT 

 

150 amps (typical cinebench R20 or Realbench 2.56 power draw at 5 ghz-5.1 ghz):

1520 - (1.6 * 150) = 1.280v max VR VOUT.

 

100 amps (typical light to medium gaming loads):

1520 - (1.6 * 100) = 1.360v max VR VOUT.

 

Etc etc.

 

Now, since you are using a custom loadline calibration, this throws off the entire specification. (0.2 mOhms)

Assuming your stress test was drawing 150 amps (1280mv max safe voltage here):

 

1400 - (150 * 0.2) = 1.370v.

So you were pulling 1.370v on VR VOUT when 1.280v should have been the limit, at that power draw.

Increasing to 1.425v would have just made things even worse here.  This is still ignoring the issue of transient spikes which would have been pretty huge here also (and not detectable on HWinfo sensors; an oscilloscope is needed).

 

For 150 amps, Loadline Calibration=High would have been safe at 1.40v set in the BIOS, since LLC High is 0.8 mOhms:

1400 - (0.8 * 150) = 1.280v !

Well, no. Because I can still run it at same settings with same voltage. Works brill still. Especially under 3x 480mm thickest rads from EK.

 

I also ran the CPU that high for an hour maybe in short 30 sec bursts for benchmarking purposes.

 

I'd show you TimeSpy from last month but the site doesn't show me results at the moment for some reason.

[bobhumplick]

"This is actually wrong.

The "Lowest" (Flat loadline) LLC ON your Asrock does *NOT* raise the voltage at all at load !

Elmor and another Asrock user actually tested this with Mode 1 and Mode 2, with mode1 being the completely flat LLC.

They were able to measure more accurately, but there was a LOT of noise as it wasn't properly grounded and I am still not sure if they were measuring VCC_Sense or MLCC caps!"

 

 

 

you are misunderstanding me.  level 1 on asrock boards are the highest.  level one is the "flat" one.  it will keep voltage set at whatever voltage you put in the bios (at least with an  8700k).  and yes it DOES raise the voltage, i measured it with a multimeter.  llc raises the voltage measured at the socket to compensate for vdroop.  and llc one is the "flat" one i was refering to, the highest llc setting.  by flat i mean the one that keeps the voltage reading "flat" or unchanging with no dips.  the one below it which allowed a bit of vdroop actually got me higher overclocks though which was level 2

[bobhumplick]

"Just for reference, if you are using a true "Flat" LLC (measured via VCC_SENSE)--VR VOUT is an accurate VCC_Sense measurement but VR VOUT does NOT show transient spikes or drops, and transients will increase with tigher loadline!, you should *NOT* exceed 1.28v set in BIOS unless you are under sub-ambient cooling and if you are attempting to run AVX prime95, you should *NOT* exceed 1.24v set in BIOS!  That's because at 193 amps of current (8 core), max VCC_Sense voltage (with respect to ground), with 193 amps being absolute max allowed by Intel, is between 1.212v-1.230v (depending on where it's measured)"

 

i dont use VRout.  it the voltage as way way too low at least on my board.  if i used VRout i could enter 1.5v and it would show it as much less.  i use the highest of the 2 vcore readings on the gigabyte board in hwinfo64 which seems to be more in line with what is actually happening and what i set in the bios.

 

right now i just set 5ghz and -85 dvid offset and left llc at standard.  that should be very safe and it runs very cool as well.  if any of the voltage readings are close to correct then its drooping to 1.2v or even below and still stable in everything.  the vrout is reading as low 1.12 or something and the second vcore reading is reading down to 1.15 or lower so i know that cant be right.  

[TheNaitsyrk]

You do you dude. I'll enjoy whatever I have. Don't want to be disrespectful, but I couldn't even be bothered to read this.

[bobhumplick]

but the point of what i was posting is that most boards (and i think boards with doublers especially) the llc that gets you a flat voltage curve (no vdroop) is too high and either wont do anything for your overclock or will actually make it worse.  the one below that or even the one 2 steps below that will get you the best overclock.  there will be vdroop but it will still clock better and run cooler with less voltage.  

 

and when i say "flat"  i dont mean the line on  the graph.   i jsut mean the llc setting that stops vdroop for that particular cpu.  when i used the flat (level 1 for asrock) on my 8700k and extrem4 asrock board it would do 5ghz at 1.315 after it drooped a bit it would read 1.295-1.305 which is close enough to what im calling "flat"  and it would only drop that far under heavy load.

 

now after i lowered the llc to level 2 (the one below the "flat"  highest llc) i could run 5.1gh at this same voltage with a -1 offset for avx.  so i got an extra 100mghz just by using a LOWER llc.  with llc at level 2 voltage would droop to maybe 1.27 or 1.28, it might have been as low as 1.26v im not sure.  but anyway even though it was .03 or so volts lower it was still stable and at a higher clock no less.

 

i think its the transient response on these boards, especially the ones with doublers maybe.  the extreme4 and the z390 pro both have doublers.  doublers add a bit of delay and i think by the time the controller knows its time to add llc and it tells the power stages to raise voltage, its a little slow to react and the voltage droops a long way before the voltage raises back up but this happens so fast it doesnt show up on the vcore reading.  so both level 1 and level 2 llc were dropping to about the same point before rebounding which made level1 kind of pointless.  if both drop down to 1.26 volts or something for an instant and then one jumps up to 1.28 and the other jumps up to 1.305 then it doesnt relaly matter because they both hit that 1.25v low and thats where the crash happens.  

 

this is just my theory.  but i think a "flat"  llc or even anything near flat on a doubled board (maybe any board but i dont know) is kinda not needed.  2-3 steps below the one that keeps the voltage from drooping is the one i would use.  if not lower.  

 

i may not be explaining it right, the words im using for stuff may be the wrong terms but i think its right though

 

AGAIN THIS IS JUST FOR NEWBS WHO THINK THAT EVERYBODY RUNS 5.3 ALL DAY AT INSANE VOLTAGES.  IF YOU KNOW WHAT YOU ARE DOING AND HAVE A GOOD WATERCOOLING SETUP THEN THIS ISNT FOR YOU.  

[bobhumplick]

"Just for reference, if you are using a true "Flat" LLC (measured via VCC_SENSE)--VR VOUT is an accurate VCC_Sense measurement but VR VOUT does NOT show transient spikes or drops, and transients will increase with tigher loadline!, you should *NOT* exceed 1.28v set in BIOS unless you are under sub-ambient cooling and if you are attempting to run AVX prime95, you should *NOT* exceed 1.24v set in BIOS!  That's because at 193 amps of current (8 core), max"

 

yeah for a 9900k using a flat llc with more than 1.28 being dangerous sounds about right.  i was talking about with an 8700k.  i used an 8700k and a 9900k when posting i think and 2 different boards so parts might be confusing.

 

but people need to be very carefull with llc and especially when getting near 1.3v.  1.4v at no llc is probalby fine.  1.3v at extreme llc is like having a deathwish.  when someone qutoes a vcore value to you make sure you understand what llc setting they mean.  

 

IF YOU USE A "FLAT" (NO VDROOP) LLC SETTING WITH AN 8700K DONT GO OVER ABOUT 1.3V.  MAYBE 1.32.  ON A 9900K LESS THAN THAT.  WHEN PEOPLE TALK ABOUT 1.4V ON A CPU BEING SAFE THEY ARE TALKING ABOUT NO OR VERY LOW LLC (PROBALBY STOCK LLC) AND WHEN PEOPLE TALK ABOUT THE MAX SAFE VOLTAGE FOR AN INTEL CPU BEING 1.52V OR WHATEVER THEY ARE TALKING ABOUT WITH VOLTAGE SPIKES AND OTHER THINGS THAT WONT SHOW UP ON A VCORE\VROUT READING.

 

FOR FLAT LLC 1.3V IS ACTUALLY HIGH!!!!  FOR STOCK LLC 1.4V IS PROBALBY BETTER THAN 1.3V WITH FLAT LLC ON AN 8700K AND ON A 9900K TOO MUCH.  JUST AS FALKENTYNE SAID!!

 

 and i know you already know this falkentyne but i was posting this for others.  

 

just remember, all you new guys.  someone saying that X voltage is safe,  without knowing the llc setting and which cpu they are talking about you dont know what voltage that really is.  1.3v at flat llc (level 1 asrock, turbo gigabyte) will end up more like 1.4v-ish if you read it with a multimeter.  if you see maybe .03- .07 vdroop then you are probalby about where you need to be for the best overclock.  now asus boards with full 8 or more  phases and no doublers may have better transient response but i dont have experience with 300+ dollar boards.  the best thing to do is test it for yourself.  the point is that setting llc too high can actually make things hotter, degrade faster, and you actually get less of an overclock.

 

right now, for a 9900k, i use auto voltage witha  -85 dvid and llc at standard at 5ghz and it runs beautifully.  in cinebench the vcore goes down as low as 1.2v but is still stable and doesn thermal throttle even with an air cooler and my tjmax set at 85c.  its awesome.  

 

if you want manul voltage i run mine at 1.285 with high llc.  thats the one right below turbo which is flat on my board.  it droops down to about 1.24v in cinbench but still is stable.  when i tried flat llc, which is turbo, i had to set 1.3v and it was hot and loud and didnt help overclocks at all.  

 

with the high llc i can run 51 with no avx offset at 1.325 and it droops down into the high 1.2-ish range (1.27 or so) but its totally stable even without an avx offset.  if i set turbo (flat) llc then i would have to go up to 1.335 or more.

 

flat llc does not help overclock.  lower the llc and raise the voltage a tiny bit and youll be better off.  most times you wont even have to raise the voltage, itll run at the same voltage or even less.

 

 

AGAIN THIS IS JUST FOR NEWBS WHO THINK THAT EVERYBODY RUNS 5.3 ALL DAY AT INSANE VOLTAGES.  IF YOU KNOW WHAT YOU ARE DOING AND HAVE A GOOD WATERCOOLING SETUP THEN THIS ISNT FOR YOU.  

[Falkentyne]

2 hours ago, bobhumplick said:

 

you are misunderstanding me.  level 1 on asrock boards are the highest.  level one is the "flat" one.  it will keep voltage set at whatever voltage you put in the bios (at least with an  8700k).  and yes it DOES raise the voltage, i measured it with a multimeter.  llc raises the voltage measured at the socket to compensate for vdroop.  and llc one is the "flat" one i was refering to, the highest llc setting.  by flat i mean the one that keeps the voltage reading "flat" or unchanging with no dips.  the one below it which allowed a bit of vdroop actually got me higher overclocks though which was level 2

Please quote the messages correctly.  Not just with "  "; that's confusing.

You measured at the socket, right?  That's the problem.

The socket is the 'MLCC' reading.  This will have higher readings because there is impedance between the socket and the actual CPU VRM supply line.

 

Did you look at the charts I linked?
look at this one again and see "Socket MLCC"?  There's your v-rise.

You need to measure directly from the VRM line that feeds VCC_Sense.  This is difficult to do and very risky.

 

https://www.overclock.net/forum/27686004-post2664.html

 

Take a look at this video.  Look at 8:00.

 

 

Notice that he has wires soldered directly to the VRM?

That's where you have to read to not have the voltage "rise" (which is caused by impedance).

That is very dangerous to do however.  You touch the wrong thing and you can kiss your motherboard goodbye.

[bobhumplick]

"Please quote the messages correctly.  Not just with "  "; that's confusing." 

 

ill quote however i want.

 

"You measured at the socket, right?  That's the problem."

 

ok so what voltage do you get when you measure with llc under load?  if it doesnt go above what you set and its medium or higher then its not reading right.  if you are just trusting sensors then just keep doing that. 

 

but think about it. if normally at 1.3v, lets say it droops to 1.2.  now with turbo llc all of a sudden it only droops to 1.29v.  the reason its only drooping to 1.29v instead of 1.2v is because its adding .09v to it to offset what it lost in droop.  thats the only way it could do it.  if the board kept the same voltage at full load with llc and the load stopped it would go up to 1.39.  in effect your voltage IS 1.39v for the time that it is at load. 

 

you might as well just set the higher voltage and not use llc.  buildzoid even says the same thing,  you know the guy who made that video?

 

i mean how do you think it counter acts vdroop?

[bobhumplick]

and also when it comes to buildzoid, i have a lot of respect for him but theres no way the VRout is the correct sensor for all the z390 aorus lineup.  cause if it is then my 9900k is running 5ghz stable at 1.15v.  i doubt thats right.  theres more to this than just what you are saying

 

just because buildzoid does it one way doesnt mean thats the only way to do it.  llc adds voltage to counteract vdroop.  if you are arguing with about how much voltage it adds then fine, it could be more than i said or less.  but it DOES add voltage.  thats how llc works.  if you think i measured it at the wrong point or whatever is irrelevant.  the point is that llc adds voltage and if you are already at a high voltage and turn up llc you have to be very very careful.  most people apply way more llc than they need and would be better off with less llc and a tad bit more voltage.  

 

plus you cant just go by a video.  i have a aorus pro.  he has the pro wifi.  they look almost identical.  except pro has a totally different controller.  not the IR one shown in the video.  it has a renesas.  and that was an asrock board anway with an 8700k

 

but what i saw when that cpu was under load was llc raising the voltage.  period.  thats how llc works.  it raises the voltage when the vcore starts to droop.  and it WAS raising it the amount i stated.  period

 

i dont even know what your original point was.  just a bunch "youre wrongs"  my point was that voltage goes up with load.  thats it.  and it does

 

 

[TheNaitsyrk]

Time to call a mod to close this thread...