cancel
Showing results for 
Search instead for 
Did you mean: 

OC for i9 9900k

poney1991
Level 7
Hey I bought a new system and I am not a pro at OC my system is

i9 9900k with a STRIX Z390E
RTX 2080 super
16GB 3200 Mhz
PSU 750 watts

I followed a totorial witch made me set LLC on 5 and 140 percent of somthing that has to do with the power the CPU draws. At the moment I have a stable OC @ 5.1GHZ core voltage 1.30v with max temps of 80-83 degrees in Cinebench R20. I can get a stable OC at 5.2 GHZ at 1.37v with max temps in cinebench around 87. I have CPU cache at 46

I read some people saying safe voltage is 1.35 others 1.40, so I take 1.35 I guess? Can I keep the 140 percent power draw 24/7? Also should I keep it on llc5 or switch to 6? I mean is 6 maybe even more stable then 5 we never know. Last question is it normal if I oc my GPU RTX 2080 Super and my CPU to 5.1 GHz that I only gain 6 fps in game but have a lower min FPS and hight AVG fps?

Thanks

Edit: I tried to set level 6 instead of 5 but everything is way hotter I hit 91 degrees at 5.2 1.34v and 82 at 1.30v for 5.1GHz, so I am confused now level 6 makes my system more stable or is it useless for me because it only makes things hotter? Set it again on level 5 stable oc @ 5.1 and 1.30v

So the system is stable Under R20 cinebench and 4-6 hours gaming on GTA V and Warzone. Now I am confused I installed R15 cinebench and it keeps on crashing I had to bump to Level 6 llc and 1.32v to get it stable so 0.3v more from R20 to R15
986 Views
7 REPLIES 7

Carlyle2020
Level 10
R20 has been designed to be more stressful. Look at the time it takes to finish compared to R15.

So i can safely bet that you accidentally changed more than just the vcore value between runs .

You seem to be an overclocker living on the youtube fastlane.
Follwoing instructions and cutting corners by not researching the changed values is tricky business at best.

Since your data is inconclusive you will have to do the test again under more controlled cicumstances please.

Let us know how the second runs end.

ps LLC5 is the sweetspot for heat/vs vdroop with 8700k and 9900k chips i have found out.
Water cooling folk miss this quiet often because they do not really have to care due to more cooling headroom.
Air coolers find that one out pretty quick IF they are savy.

Best Regards
Carlyle

Arne_Saknussemm
Level 40
Higher LLC value...higher current/voltage....way more heat. LLC5 is a good spot

If you are on the edge of your OC you may pass or fail any given test...

"Should I...?".....you should do what works for your particular CPU...there are no rules for this stuff...only generalities that will give you a starting point...the rest is experimentation.

Try LLC5 and bump core v by 0.02v

Check what the board is setting for VCCSA and VCCIO...often more than needed and they contribute to heat...

poney1991
Level 7
Hey yhanks for the fast replies, I tested whole night yesterday. So I can confirm that with r20 my system does not bsod but with r15 it does I am confused I gad to bump my vcore from 1.30v stable on r20 to 1.34 to be able to run r15 without bsod. So I had The cpu power managment on max avx on 3 llc on 5 and vcore on 1.30v cpu @ 5.1 ghz stable on R 20 Gta v warzone. But BSOD on OCCT with AVX off and r15. I had to bump to 1.34v and llc 6 to get it stable on R15 and OCCT.

So if I get ot right staying under 85 at stress test and under 70 for gaming. And max 1.35v and max llc6 to stay safe.

How is it possible that ai optimizer sets it to 5.0ghz 1.30v while at 5.1 1.34v for me its not even sute its stable. Is AI optomisation so bad in 2020 compared to manual for an youtube ocer xD? Last question arent these 20 bsod not bad for my system?

poney1991@live.be wrote:
Hey yhanks for the fast replies, I tested whole night yesterday. So I can confirm that with r20 my system does not bsod but with r15 it does I am confused I gad to bump my vcore from 1.30v stable on r20 to 1.34 to be able to run r15 without bsod. So I had The cpu power managment on max avx on 3 llc on 5 and vcore on 1.30v cpu @ 5.1 ghz stable on R 20 Gta v warzone. But BSOD on OCCT with AVX off and r15. I had to bump to 1.34v and llc 6 to get it stable on R15 and OCCT.

So if I get ot right staying under 85 at stress test and under 70 for gaming. And max 1.35v and max llc6 to stay safe.

How is it possible that ai optimizer sets it to 5.0ghz 1.30v while at 5.1 1.34v for me its not even sute its stable. Is AI optomisation so bad in 2020 compared to manual for an youtube ocer xD? Last question arent these 20 bsod not bad for my system?


BSOD on R15 but not on R20?
That's because you're using an AVX offset.
Remove the AVX offset and R15 will not BSOD anymore.

This is tricky because R15 does *NOT* use AVX. However the AVX offset gets repeatedly "triggered" off and on by other things in the background (system libraries, drivers, etc) which causes the voltage transients during the R15 run to drop through the gutter and you crash, because the CPU downclocks randomly and then re-upclocks, causing a massive loss of stability. R15 isn't triggering this random downclock. A random downclock during a non AVX application by a background AVX application forces the CPU "PLL" to fall asleep and re-awaken at the lower clock frequency, and the same thing when it clocks back up. The VRM's are not able to handle this "max load-no load-max load" gracefully and the voltage drops drastically but too fast for your onboard sensors to pick up. An oscillioscope would show a massive voltage dip however. That's why you crash.

The reason why this does not happen with R20 is because R20 actually uses AVX, so the CPU downclocks and remains downclocked during the entire test.

Arne_Saknussemm
Level 40
When AI optimizer is better than manual OC....we are all done for...

BSODs can be bad for OS...depends

Run "sfc /scannow" in command prompt....open cmd as administrator...this will give you a clue....if anything important is corrupted

Watch der8auer's 9900K on Asus overclocking guide, it's quite useful.

https://www.youtube.com/watch?v=95Ujni7-fVM&t=627s

I run my 9900K with


  • Load Line Calibration of 6 or 7 in the BIOS (my Hero set itself to 6 based on auto calibration after the initial overclock, and I've left it as Auto)
  • all CPU core ratios to 49
  • manual CPU core voltage of 1.28 V
  • AVX offset of 3 (4.9 -> 4.6 GHz during AVX workloads)
  • RAM (2x 16 GB Vengeance 3000 Mhz) at XMP defaults with a slight voltage bump to 1.35 V.


With a Corsair H100i I get idle and general usage temps between 39 and 45 celcius; gaming or rendering takes them to mid 60 or low 70s.

This is slightly conservative relative to what this CPU could achieve. I prefer to prolong CPU life instead of burn it out or lose the OC down the line!

Synthetic tests like OCCT or Cinebench R20 give me temps around 75-82 all cores and package; Folding@Home will take it up to 85-88 package but I throttle that back to 6 cores. Cinebench R15 runs about the same temps as R20. At my level of OC, I have no instability from any AVX offset and I prefer a small downclock for heat management.

Load Line Calibration is an important concept to understand - configuring it properly helps compensate for Vdroop during higher overclock workloads. Improperly calculating it can prematurely degrade / damage your CPU.

Also watch this GamersNexus explainer: https://www.youtube.com/watch?v=NMIh8dTdJwI
MSI published a decent blog post about it: https://www.msi.com/blog/LLC_what_is_it_and_why_are_MSI_Z370_motherboards_the_best_choice_for_overcl...
A guy from ASUS ROG also filmed a GN primer about transients, VRM and LLC: https://www.youtube.com/watch?v=tqvj4CQhRmg


From what I've been advised by people overclocking these chips all the time, anything up to 1.45 V is safe for the CPU - it'll thermal throttle before it damages itself permanently. However you'll have a massive cooling issue to deal with first. Do not push the LLC to more extreme levels once you start to get to the 1.4 V CPU core voltage level as you will have load voltages towards 1.5 V which is dangerous for the CPU and stresses the VRM.


Take a note of your per-core temperatures while rendering, if you're unlucky (or your thermal paste is badly applied) you may have some cores which are heating faster than others. If so, consider setting per-core overclock ratios and adjusting the CPU voltage down a little.

I would do something like run an OCCT stress test for 1 or 2 hours; if it passes that without BSOD or thermal throttling, your machine is probably stable enough for normal use. Don't rely solely on synthetic benchmarks. Running small FFT Prime95 BSODed my PC almost immediately (before I adjusted my previous OC a little) but I don't care specifically about Prime95 outcomes because it's not representative of real world use. Unless your sole intention is to run Prime95 🙂


The 990K is a HOT chip, and unless you're going to delid and spend $$$ on liquid metal cooling with IHS removal to do direct-die cooling, managing temps is going to be a pain. Your mileage may vary...

Personally I wouldn't bother going beyond 5 GHz unless you have a decent watercooling setup.
Hi mum!

Synergist wrote:
Watch der8auer's 9900K on Asus overclocking guide, it's quite useful.

https://www.youtube.com/watch?v=95Ujni7-fVM&t=627s

I run my 9900K with


  • Load Line Calibration of 6 or 7 in the BIOS (my Hero set itself to 6 based on auto calibration after the initial overclock, and I've left it as Auto)
  • all CPU core ratios to 49
  • manual CPU core voltage of 1.28 V
  • AVX offset of 3 (4.9 -> 4.6 GHz during AVX workloads)
  • RAM (2x 16 GB Vengeance 3000 Mhz) at XMP defaults with a slight voltage bump to 1.35 V.


With a Corsair H100i I get idle and general usage temps between 39 and 45 celcius; gaming or rendering takes them to mid 60 or low 70s.

This is slightly conservative relative to what this CPU could achieve. I prefer to prolong CPU life instead of burn it out or lose the OC down the line!

Synthetic tests like OCCT or Cinebench R20 give me temps around 75-82 all cores and package; Folding@Home will take it up to 85-88 package but I throttle that back to 6 cores. Cinebench R15 runs about the same temps as R20. At my level of OC, I have no instability from any AVX offset and I prefer a small downclock for heat management.

Load Line Calibration is an important concept to understand - configuring it properly helps compensate for Vdroop during higher overclock workloads. Improperly calculating it can prematurely degrade / damage your CPU.

Also watch this GamersNexus explainer: https://www.youtube.com/watch?v=NMIh8dTdJwI
MSI published a decent blog post about it: https://www.msi.com/blog/LLC_what_is_it_and_why_are_MSI_Z370_motherboards_the_best_choice_for_overcl...
A guy from ASUS ROG also filmed a GN primer about transients, VRM and LLC: https://www.youtube.com/watch?v=tqvj4CQhRmg


From what I've been advised by people overclocking these chips all the time, anything up to 1.45 V is safe for the CPU - it'll thermal throttle before it damages itself permanently. However you'll have a massive cooling issue to deal with first. Do not push the LLC to more extreme levels once you start to get to the 1.4 V CPU core voltage level as you will have load voltages towards 1.5 V which is dangerous for the CPU and stresses the VRM.


Take a note of your per-core temperatures while rendering, if you're unlucky (or your thermal paste is badly applied) you may have some cores which are heating faster than others. If so, consider setting per-core overclock ratios and adjusting the CPU voltage down a little.

I would do something like run an OCCT stress test for 1 or 2 hours; if it passes that without BSOD or thermal throttling, your machine is probably stable enough for normal use. Don't rely solely on synthetic benchmarks. Running small FFT Prime95 BSODed my PC almost immediately (before I adjusted my previous OC a little) but I don't care specifically about Prime95 outcomes because it's not representative of real world use. Unless your sole intention is to run Prime95 🙂


The 990K is a HOT chip, and unless you're going to delid and spend $$$ on liquid metal cooling with IHS removal to do direct-die cooling, managing temps is going to be a pain. Your mileage may vary...

Personally I wouldn't bother going beyond 5 GHz unless you have a decent watercooling setup.


The AVX offset is crashing him because his Non-AVX stability is already borderline (if he were not using an offset to begin with).
AVX offsets need to be improved by Intel and even they admitted it needs to be improved. Currently the only point of an AVX offset is to reduce temps and force a lower ratio to improve stability when in AVX instructions, not improve global stability, because it hurts stability in non AVX applications.