i7 8700K, 8600K, 7700K Overclock results

[FireFox]

IA AC Load Line - 0.01 (These are what I needed to tame my 1.5 overshoot when booting into desktop)
IA DC Load Line - 0.01 (These are what I needed to tame my 1.5 overshoot when booting into desktop)

Was that my board had a LLC on the AC and DC IA lines, and with it set to "Auto" was overshooting (temp spikes). Setting the LLC of those to the lowest I could .1

In my System it doesn't makes any difference when set Auto or 0.01

 

[Cybrnook2002]

In my System it doesn't makes any difference when set Auto or 0.01

That's good to know. On mine it does a very big one.

Let me add to this post to try and answer a bit more. I saw all your questions @Knoxx29 and am trying not to answer them all one by one I put a link in there, and noted next to most of my line items why I did what I did, most is already answered.

But, I can see you have a similar board to me with the same model CPU. Just because they are similar, does not make them the same. If you are looking at validating your OC settings by comparing to mine, that could take a while. I only listed what worked for me, and what I found to be solid for me after a few days of rigiourous back and forth and testing for AVX vs non AVX loads and watching like a hawk my VID, vcore temps etc...... Few things to keep in mind for us would be BIOS versions, technically neither one of us listed. I am on the latest 1003, but made all my OC settings on 0802 (and updated later), that can make a big difference on how voltages and VRM's etc.. act on a board. Also our VID's are likely different, same reason some chips are great clockers, and some are not.

Nevertheless, here you are:

XMP for DDR3200 2 x 16GB, selected "No" on doing ALL adjustments, just loaded the profile.

Works fine to load the profile, notice I said did "not" click yes to load "all" adjustments

AVX Negative offset 2 (Though I could likely raise this, or seemingly remove it back to 0)

As I noted, I could remove this and go back to 0, which I will be doing in the end as I am 5.0 stable across the board, no need to offset. I read in a few threads on the interwebs as well that there is a suspicion that BIOS detection of AVX may not be the most accurate, and could be down-clocking unnecessarily.

Sync All Cores

Of course, sync all cores

50

5.0 ghz

LLC 5 (I am waffling between 5 and 6. At 5 with 1.31, I drop to 1.296 under load which runs fine. Then at 6 with 1.30, I move up to 1.32 under load, so a little bit of an over shoot with some small heat increase. If 5 proves unstable, I will drop vcore to 1.30 and use 6 since that was 6 hours OCCT with AVX stable).

this is LLC for the CPU itself (different from AC/DC IA). I am using LLC to "stabilize" my load ramps. It ramps up and down now exactly the same every time, and I know the vcore it will land on.

Long Duration PPL - 4095

Provide voltage/current for a longer duration to sustain a better "hold out". Carried over from my Z170/270 clocking.

Short Duration PPL - 4095

Provide voltage/current for a longer duration to sustain a better "hold out". Carried over from my Z170/270 clocking.

IA AC Load Line - 0.01 (These are what I needed to tame my 1.5 overshoot when booting into desktop)

Without this, LLC was ramping up to almost 1.5 on a desktop boot (which if you google is common on the Z370 platform). Removing this (setting to lowest setting), tames the "Auto" beast, and now my main LLC is where I ramp to, and it is stable and predictable.

IA DC Load Line - 0.01 (These are what I needed to tame my 1.5 overshoot when booting into desktop)

Without this, LLC was ramping up to almost 1.5 on a desktop boot (which if you google is common on the Z370 platform). Removing this (setting to lowest setting), tames the "Auto" beast, and now my main LLC is where I ramp to, and it is stable and predictable.

CPU Core/Cache Current Limit - 255.50

Without this I was noticing a fluctuation of my clocks when running a longer/sustained stress tests. It would ramp to 5, then drop to 4.5 - 4.8, then back to 5 etc. With it set, it goes up and it stays up. Different from AVX offset, this was almost like a premature down-clock.

Core voltage - Adaptive

Adaptive to get a drop in vcore states under low to no load

Eventual Turbo Voltage - 1.31 with LLC 5 (1.30 with LLC 6)

Core I set my target to be, then use LLC to bring up the read end under load

Hopefully that helps you, my hands are tired now.

 

[erocker]

@Cybrnook2002

Do you know if you have an early production 8700K? I know mine is and the earlier ones seems to have some wacky VID's that just aren't even close to what the chip actually is.
I got my other system built now, so I'm going to send mine off to Silicon Lottery for a delid and testing. Seems to be a pretty good deal.

 

[Cybrnook2002]

I would assume it's a more current production line. BUT, I will say it was an "tray" CPU I bought from Provantage for about $350 when the 8700K's were going for $410 everywhere else. Probably about a month or so ago.....

And de-lid it yourself Rock-it kit seems to still work on the Coffee lake chips (so far). Nothing more satisfying than doing it yourself.

I have been tempted to do mine as well, but with my loads temps running in the mid 70's (over 20 + away from tjmax) and idle in the 25's, I am sitting pretty. If you want to do it yourself though, I bought some syringes from Amazon that have a larger diameter needle which was PERFECT for me when I laid a new bead of RTV down, just pay shipping.

 

[erocker]

Nah, not going to delid... I screwed one chip already. I can next day it to Silicon Lottery and for 60 bucks they'll do it and test it. I'm almost more interested in just getting a 3rd party to clock the chip.

 

[FireFox]

That's good to know. On mine it does a very big one.

Let me add to this post to try and answer a bit more. I saw all your questions @Knoxx29 and am trying not to answer them all one by one I put a link in there, and noted next to most of my line items why I did what I did, most is already answered.

But, I can see you have a similar board to me with the same model CPU. Just because they are similar, does not make them the same. If you are looking at validating your OC settings by comparing to mine, that could take a while. I only listed what worked for me, and what I found to be solid for me after a few days of rigiourous back and forth and testing for AVX vs non AVX loads and watching like a hawk my VID, vcore temps etc...... Few things to keep in mind for us would be BIOS versions, technically neither one of us listed. I am on the latest 1003, but made all my OC settings on 0802 (and updated later), that can make a big difference on how voltages and VRM's etc.. act on a board. Also our VID's are likely different, same reason some chips are great clockers, and some are not.

Nevertheless, here you are:


Works fine to load the profile, notice I said did "not" click yes to load "all" adjustments

As I noted, I could remove this and go back to 0, which I will be doing in the end as I am 5.0 stable across the board, no need to offset. I read in a few threads on the interwebs as well that there is a suspicion that BIOS detection of AVX may not be the most accurate, and could be down-clocking unnecessarily.

Of course, sync all cores

5.0 ghz

this is LLC for the CPU itself (different from AC/DC IA). I am using LLC to "stabilize" my load ramps. It ramps up and down now exactly the same every time, and I know the vcore it will land on.

Provide voltage/current for a longer duration to sustain a better "hold out". Carried over from my Z170/270 clocking.

Provide voltage/current for a longer duration to sustain a better "hold out". Carried over from my Z170/270 clocking.

Without this, LLC was ramping up to almost 1.5 on a desktop boot (which if you google is common on the Z370 platform). Removing this (setting to lowest setting), tames the "Auto" beast, and now my main LLC is where I ramp to, and it is stable and predictable.

Without this, LLC was ramping up to almost 1.5 on a desktop boot (which if you google is common on the Z370 platform). Removing this (setting to lowest setting), tames the "Auto" beast, and now my main LLC is where I ramp to, and it is stable and predictable.

Without this I was noticing a fluctuation of my clocks when running a longer/sustained stress tests. It would ramp to 5, then drop to 4.5 - 4.8, then back to 5 etc. With it set, it goes up and it stays up. Different from AVX offset, this was almost like a premature down-clock.

Adaptive to get a drop in vcore states under low to no load

Core I set my target to be, then use LLC to bring up the read end under load

Hopefully that helps you, my hands are tired now.

I got your point from the first post and i know very well what each setting is for, but i am still concerned about how it could be possible all those settings for a simple 5.0GHz, to get your OC stable it tooks you days and to me it tooks me minutes right after installing the CPU and setting the Voltages after that i haven't touched my OC .

Few things to keep in mind for us would be BIOS versions, technically neither one of us listed. I am on the latest 1003,

I am on the latest 1003 too.

Nah, not going to delid... I screwed one chip already.

I haven't screwed a CPU yet but this time and i don't know why i wont Delidd, weird to say that

 

[Cybrnook2002]

I got your point from the first post and i know very well what each setting is for, but i am still concerned about how it could be possible all those settings for a simple 5.0GHz, to get your OC stable it tooks you days and to me it tooks me minutes right after installing the CPU and setting the Voltages after that i haven't touched my OC .

No need to be concerned.

Nah, not going to delid... I screwed one chip already. I can next day it to Silicon Lottery and for 60 bucks they'll do it and test it. I'm almost more interested in just getting a 3rd party to clock the chip.

But then again, you assume your getting the same chip back sure the lid may be the same......

Perhaps you can pull batch and serial with a program prior to shipping, outside of just what's on the lid...? (Tin foil hat on right now) maybe a small sharpie dot on the side of the wafer.

 

[Vayra86]

4.8 stable, 1.31v Spikes are down to 85C. LLC lv 4 instead of 5 (lowest), allowed me to push down the vCore by 0.02 - HT on as well. Removing it has a positive effect on temps (-4, -5C avg) and 'costs' about 0.02vCore.

I also now know that I need 1.36v for 4.9 non-HT, and 1.376v with HT, which is a bit too much for this cooling. 5.0 will take 1.38v non-HT and crashes due to temps.

Note: my OC is done in comfort mode, I don't go fans all out blazing. Want to always see the real deal... so while you see those temps, this is inaudible

 

[EarthDog]

Well done on the OC. Cranking fans only yield a few degrees anyway,
so, makes sense to keep it where its comfortable for the ears.

But then again, you assume your getting the same chip back sure the lid may be the same......

Perhaps you can pull batch and serial with a program prior to shipping, outside of just what's on the lid...? (Tin foil hat on right now) maybe a small sharpie dot on the side of the wafer.

Seriously??

 

[Cybrnook2002]

Seriously??

Yeah, I know I know. SL has been around for a while, and I know they are reputable. I am just a skeptical kinda guy topic is often irrelevant, hehe

Wanted to add a little screenie to the thread for you @Vayra86. I put AVX offset back to 0 (as I knew I could), and sure enough no issues. Ran a quick little 30 min run for you:

 

[Vayra86]

Since I can't seem to edit my opening post I ended up with the following settings

vCore / Offset -70 / 1.328v - 1.312v @ load
LLC 4 (2nd lowest)
VCCIO 1.000v
VCCSA 1.000v
HT ON
48x All cores

Thanks everyone for the helpful advice and sharing results!

I also added lights to make it go faster Another big thing, is that this case is SO much better for airflow than my old FD Define R4. GPU temps down by 5-7C under load; can push 2076mhz @ 70C now instead of 1974 @ 75 C. There is literally zero heat built up inside, even after hours of max load, most of it is still cool to the touch. Its also super nice to build in. Can def. recommend! (Fractal Design Define C TG)

 

[Cybrnook2002]

Good job, enjoy it!

 

[FireFox]

@Cybrnook2002 do you mind if i try your settings? i am curious to see if they work on my Machine

 

[Cybrnook2002]

Not at all, have at it. That's why I listed them out, for people to pick and choose what they want. I would recommend you clear CMOS before starting, so we know it's a clean slate.

Everything else at default/auto unless called out:

XMP for DDR3200 2 x 16GB, selected "No" on doing ALL adjustments, just loaded the profile. (Save and reboot back into UEFI)
AVX Negative offset 0 (Updated to 0 since it's fine)
Sync All Cores
50
LLC 5
Long Duration PPL - 4095
Short Duration PPL - 4095
IA AC Load Line - 0.01
IA DC Load Line - 0.01
CPU Core/Cache Current Limit - 255.50
Core voltage - Adaptive
Eventual Turbo Voltage - 1.30 with LLC 5

I set on 1.30 and LLC 5, however, this is not a guarantee here and will really determine on what your CPU wants voltage wise.

 

[Enterprise24]

What about voltage scaling ? My 8700K 5.0 = 1.31 , 5.1 = 1.37 while 5.2 need 1.48 !! temp is under control. My previous i5-6500 scale linearly until 1.52V. Quiet strange since Coffeelake is still base on Skylake uarch.

 

[Cybrnook2002]

Yep scaling starts getting heavy past 5.0. I noticed for mine I needed a hair over 1.35 for 5.2 with a higher LLC of 6, which overshot to about 1.37 under load. With a still lidded chip, I was NOT comfortable where temps were headed .

 

[Vayra86]

I would personally not go past 1.38v to be fair, even regardless of temps. Even with a tight VRM you will see it spike past 1.4v which really should be a hard limit, if this is a 24/7 CPU for ya. I have seen OC's at 1.41v, but in all fairness, I'll be surprised if they hold AND remain stable after a few years. That's over 220w getting pulled through a board.

 

[Vario]

Anyone do any ram overclocking on this platform? I managed to get this validation yesterday but probably would really need 1.4v Dram for stability.
https://valid.x86.fr/qpyumu
DDR4 4300 CL19 1.35V.
After microcode 80 update, it does seem like a need a lot more voltage just to get to windows for 50X multi (5 ghz) so I will just work with 4.8 GHz and see how optimized I can get it since I can do that at 1.2V and below
Working on a 24/7 overclock right now maybe DDR4 4000 19-19-19-39 @ 1.375V, VCCIO 1.120, VCSST 1.1V. Vcore offset 30 Multi 48X AVX offset multi 1. Running AVX offset to reduce temperatures. Under load VID is 1.260 and actual cpu vcore is 1.168 to 1.20V for 4.8 Ghz, 4.5GHz uncore. I have also found setting base clock to 100 Mhz seems to improve stability but I can't quite prove it yet. I'd like to tighten up my ram timings a bit I guess reduce the CL first, I am not the best with ram OC I just looked at some of the factory settings for the higher binned G Skills. I would also like to reduce the VCCIO and VCSST, they are set to 1.1V in bios. I don't know how doable that would be.

CPU-Z validation at same time as Aida:
https://valid.x86.fr/7pptqb

What about voltage scaling ? My 8700K 5.0 = 1.31 , 5.1 = 1.37 while 5.2 need 1.48 !! temp is under control. My previous i5-6500 scale linearly until 1.52V. Quiet strange since Coffeelake is still base on Skylake uarch.

From what many say on Overclock.net, delid is necessary to get past the 5.1-5.2 wall.

Not at all, have at it. That's why I listed them out, for people to pick and choose what they want. I would recommend you clear CMOS before starting, so we know it's a clean slate.

You should set your uncore aka nb frequency up, it should reduce latency a tiny bit and improve performance. A good processor might get you 4.8Ghz uncore, 4.5 should be doable on any I think.

Silicon lottery has some rip off prices on their site. It would be foolish to buy any of the processors that are under 5.1 binning because pretty much all coffee lake should reach 5.0Ghz at 1.4v or less. You would be better off buying it unmodified with intel warranty intact. In fact, a 5Ghz Coffee Lake that requires 1.4v is a very bad sample! A good binned 5 Ghz should be 1.25V.

 

[fusseli]

On my 7700k I've gotten 5.0-5.2ghz marginally stable, but vcore as high as 1.4v didn't solve infrequent crashes. 4.9 on all cores however is rock solid all the time.

I have been very lazy and have only tried asus optimal and upping the cpu power targets, everything else is auto.

 

[Vario]

Nice result, can you get lower vcore on the 4.9? You could try also raising your uncore frequency.

 

[Vayra86]

Anyone do any ram overclocking on this platform? I managed to get this validation yesterday but probably would really need 1.4v Dram for stability.
https://valid.x86.fr/qpyumu
DDR4 4300 CL19 1.35V.
After microcode 80 update, it does seem like a need a lot more voltage just to get to windows for 50X multi (5 ghz) so I will just work with 4.8 GHz and see how optimized I can get it since I can do that at 1.2V and below
Working on a 24/7 overclock right now maybe DDR4 4000 19-19-19-39 @ 1.375V, VCCIO 1.120, VCSST 1.1V. Vcore offset 30 Multi 48X AVX offset multi 1. Running AVX offset to reduce temperatures. Under load VID is 1.260 and actual cpu vcore is 1.168 to 1.20V for 4.8 Ghz, 4.5GHz uncore. I have also found setting base clock to 100 Mhz seems to improve stability but I can't quite prove it yet. I'd like to tighten up my ram timings a bit I guess reduce the CL first, I am not the best with ram OC I just looked at some of the factory settings for the higher binned G Skills. I would also like to reduce the VCCIO and VCSST, they are set to 1.1V in bios. I don't know how doable that would be.
View attachment 96181

View attachment 96180

CPU-Z validation at same time as Aida:
https://valid.x86.fr/7pptqb


From what many say on Overclock.net, delid is necessary to get past the 5.1-5.2 wall.



You should set your uncore aka nb frequency up, it should reduce latency a tiny bit and improve performance. A good processor might get you 4.8Ghz uncore, 4.5 should be doable on any I think.

Silicon lottery has some rip off prices on their site. It would be foolish to buy any of the processors that are under 5.1 binning because pretty much all coffee lake should reach 5.0Ghz at 1.4v or less. You would be better off buying it unmodified with intel warranty intact. In fact, a 5Ghz Coffee Lake that requires 1.4v is a very bad sample! A good binned 5 Ghz should be 1.25V.

I run my VCCIO, VCCSA at 1.000, with 2x 8GB 3200mhzCL16 sticks. You could probably drop to 1.050 on both. I also lowered my CPU PLL from 1.2v > 1.150 with no negative effect on stability.

AVX offset is broken, you're better off going for 47x multi instead then and leave it at 0. Keep in mind, our boards are near identical Still if you get 4.7 all core stable on 1.2V, that is quite a feat.

As for 5.0 -for most CPUs the voltage needed ramps up very quickly after 4.8... so it seems.

 

[Enterprise24]

Anyone do any ram overclocking on this platform? I managed to get this validation yesterday but probably would really need 1.4v Dram for stability.
https://valid.x86.fr/qpyumu
DDR4 4300 CL19 1.35V.
After microcode 80 update, it does seem like a need a lot more voltage just to get to windows for 50X multi (5 ghz) so I will just work with 4.8 GHz and see how optimized I can get it since I can do that at 1.2V and below
Working on a 24/7 overclock right now maybe DDR4 4000 19-19-19-39 @ 1.375V, VCCIO 1.120, VCSST 1.1V. Vcore offset 30 Multi 48X AVX offset multi 1. Running AVX offset to reduce temperatures. Under load VID is 1.260 and actual cpu vcore is 1.168 to 1.20V for 4.8 Ghz, 4.5GHz uncore. I have also found setting base clock to 100 Mhz seems to improve stability but I can't quite prove it yet. I'd like to tighten up my ram timings a bit I guess reduce the CL first, I am not the best with ram OC I just looked at some of the factory settings for the higher binned G Skills. I would also like to reduce the VCCIO and VCSST, they are set to 1.1V in bios. I don't know how doable that would be.




CPU-Z validation at same time as Aida:
https://valid.x86.fr/7pptqb


From what many say on Overclock.net, delid is necessary to get past the 5.1-5.2 wall.



You should set your uncore aka nb frequency up, it should reduce latency a tiny bit and improve performance. A good processor might get you 4.8Ghz uncore, 4.5 should be doable on any I think.

Silicon lottery has some rip off prices on their site. It would be foolish to buy any of the processors that are under 5.1 binning because pretty much all coffee lake should reach 5.0Ghz at 1.4v or less. You would be better off buying it unmodified with intel warranty intact. In fact, a 5Ghz Coffee Lake that requires 1.4v is a very bad sample! A good binned 5 Ghz should be 1.25V.

I am sure that RAM on Taichi can do the following...

3600 14-15-15-28-278-1T @ 1.45V
3866 15-16-16-28-278-1T @ 1.45V
4000 16-17-17-28-300-1T @ 1.45V
4133 17-18-18-38-3XX-1T @ 1.45V
4266 18-19-19-39-3XX-2T @ 1.45V or 16-17-17-37-3XX-2T @ 1.55V
4400 19-20-20-40-3XX-2T @ 1.5V (if IMC is strong enough).

All of this base on at least 400% HCI Memtest.

 

[EarthDog]

You run AVX loads vaya? Otherwise it wont change temps..

I also doubt -1 is going to yield much cooler anyway considering i doubt it changed vcore much.

 

[Vario]

I run my VCCIO, VCCSA at 1.000, with 2x 8GB 3200mhzCL16 sticks. You could probably drop to 1.050 on both. I also lowered my CPU PLL from 1.2v > 1.150 with no negative effect on stability.

AVX offset is broken, you're better off going for 47x multi instead then and leave it at 0. Keep in mind, our boards are near identical Still if you get 4.7 all core stable on 1.2V, that is quite a feat.

As for 5.0 -for most CPUs the voltage needed ramps up very quickly after 4.8... so it seems.

DDR4 4300 requires 1.1V for the VCCIO/VCSSA. It is only just stable enough to run some benchmarks. Maybe if I was less conservative I'd push 1.45V VDRAM into it and 1.2V VCCIO VCSST and it would probably stabilize no problem. However, I have found on benchmarks though there is almost nothing to be gained over 3200 CL14 so I am running 1 volt for VCCIO VCSST as Intel specifies. AVX offset works perfectly fine for me. I am all core stable 48X/47X with under load peak voltage 1.168V with transient voltage peak 1.2V on transition idle/load. Clearing my CMOS after the bios update fixed the excessive vdroop issues I was encountering from bios 1.30 to 1.40, so now I am running the following settings.
Bios P1.40
All Core Multi 48x
Cache Multi 45x
Load Line Calibration: LLC2
Vcore Voltage Offset: +10
AVX Offset: 1 ( to reduce temperatures and keep the chip under 65*C while AVX, not really necessary but I am air cooled and small case so I am being conservative)
VCCIO Voltage: 1 volt
VCSST Voltage: 1 volt
Ram: 3200 14-14-14-34-480 2CR (XMP speed and timings)
DRam Voltage 1.330 (gives 1.350 in hw monitor)

I have found that LLC2 is essential for this clock as LLC3 requires a lot more offset for stability. The LLC on this board does not feel very powerful. Very slight Vdroop occurs even under LLC1 (maximum) while on other boards I owned in the past, it would over volt under load on the highest LLC.
LLC2 requires offset +10. LLC3 requires offset +40.
At only 1.2V Vcore (the max ever reported), I can't imagine it has a particularly high unreported transient Maximum Positive Overshoot as a consequence of LLC2.

You run AVX loads vaya? Otherwise it wont change temps..

I also doubt -1 is going to yield much cooler anyway considering i doubt it changed vcore much.

AVX draws something like 20 more watts reported in HWmonitor, using the AVX offset to reduce it to 4.7 drops temperatures about 10 degrees for Aida FPU test. I'd like 5+ years out of this chip so I am being real conservative.

I am sure that RAM on Taichi can do the following...

3600 14-15-15-28-278-1T @ 1.45V
3866 15-16-16-28-278-1T @ 1.45V
4000 16-17-17-28-300-1T @ 1.45V
4133 17-18-18-38-3XX-1T @ 1.45V
4266 18-19-19-39-3XX-2T @ 1.45V or 16-17-17-37-3XX-2T @ 1.55V
4400 19-20-20-40-3XX-2T @ 1.5V (if IMC is strong enough).

All of this base on at least 400% HCI Memtest.

Seems very likely it can, I was doing the 4300 with 1.35V to post and it was not stable, I don't remember if I ran 1.4V, I think so but I do remember it failed again after I ended benchmark. The AIDA memory benchmark didn't look that much better than the 3200 CL14 to make it worth it.

 

[Enterprise24]

DDR4 4300 requires 1.1V for the VCCIO/VCSSA. It is only just stable enough to run some benchmarks. Maybe if I was less conservative I'd push 1.45V VDRAM into it and 1.2V VCCIO VCSST and it would probably stabilize no problem. However, I have found on benchmarks though there is almost nothing to be gained over 3200 CL14 so I am running 1 volt for VCCIO VCSST as Intel specifies. AVX offset works perfectly fine for me. I am all core stable 48X/47X with under load peak voltage 1.168V with transient voltage peak 1.2V on transition idle/load. Clearing my CMOS after the bios update fixed the excessive vdroop issues I was encountering from bios 1.30 to 1.40, so now I am running the following settings.
Bios P1.40
All Core Multi 48x
Cache Multi 45x
Load Line Calibration: LLC2
Vcore Voltage Offset: +10
AVX Offset: 1 ( to reduce temperatures and keep the chip under 65*C while AVX, not really necessary but I am air cooled and small case so I am being conservative)
VCCIO Voltage: 1 volt
VCSST Voltage: 1 volt
Ram: 3200 14-14-14-34-480 2CR (XMP speed and timings)
DRam Voltage 1.330 (gives 1.350 in hw monitor)

I have found that LLC2 is essential for this clock as LLC3 requires a lot more offset for stability. The LLC on this board does not feel very powerful. Very slight Vdroop occurs even under LLC1 (maximum) while on other boards I owned in the past, it would over volt under load on the highest LLC.
LLC2 requires offset +10. LLC3 requires offset +40.
At only 1.2V Vcore (the max ever reported), I can't imagine it has a particularly high unreported transient Maximum Positive Overshoot as a consequence of LLC2.


AVX draws something like 20 more watts reported in HWmonitor, using the AVX offset to reduce it to 4.7 drops temperatures about 10 degrees for Aida FPU test. I'd like 5+ years out of this chip so I am being real conservative.


Seems very likely it can, I was doing the 4300 with 1.35V to post and it was not stable, I don't remember if I ran 1.4V, I think so but I do remember it failed again after I ended benchmark. The AIDA memory benchmark didn't look that much better than the 3200 CL14 to make it worth it.

You need to tighten sub timings. At 4000+ some timings will go very loose.

Try looking at this. Credit : https://bbs.io-tech.fi/threads/inte...otukset-kokemukset.41829/page-53#post-2000999