Z790-H + i7 14700KF + Luiqid Freezer 3 360 - throtling on stock

[Pix2186PL]

Hallo Asus community,

I've got problem (like allmost all user of 13 th and 14 th gen Intel processors) that on stock settings processor throtle duo high temperatures. Now for some details:

- new system build by me (not the first time) - Asus z790-h, i7 14700KF, Liquid Freezer 3 360 (fans are pulling out air from case), Kingston (2x 16 GB) ddr 5 6000 mhz CL 32, be quiete Straight Power 12 1000W, case asus GT520 + 5 arctic fans (3 at bottom to suck air, one at the back to remove air and one in the middle), asus tuf rtx 4070 super

- motherboard fully updated to bios 2301  (just notice that o 12/07 there is a new one 2402)

- bios on mainly stock settings (only XMP 1 activated for memory, all fan curve on standard and Asus MultiCore Enhancement [Disabled - Enforce All Limits] )

- Arctic Liquid Freezer 3 installed with added contact frame and allready 2 time repasted with Thermal Grizzly Kryonaut to be 100% sure that there are no problems with heat dissipation. Cooler is connected by cable that allow to control separate fans on vrm, cpu and pump since when it was connected with the combo one i've got feeling that they didn't work in 100%. Pump header connected to CPU_OPT, fan to CPU_FAN and vrm to CASE_FAN

- Windows 11 pro fully updated and to read temperatures in system i've use aida64, hwinfo, intel xtu. To test system stability and load i'm using Cinebench 24

- all temperatures that i'm showing here are on cpu package so the hotes spot

So now in idle i've got somphing like 40 C - nothing to woried about since there is very hot outside and we got only one simple fan in home so not the best conditions. After 10 min Cinebench i've got somphing like 1900 points and hwinfo show 96C+ with thermal throtle (i would say that most of people got same results or the same even on aio 240.

Anyway i've try to first change SVID Behavior from auto to somphing else and notice that auto is probably setting "Intel Fail Save" since after picking it manualy i've got same vcore in hwinfo under load with is 1.480V+. On "worst-case scenario" vcore is lower 1.450V but still same thermals result and same cinebench points. Trying "typical" scenario ended with lower vcore 1.350V and not thermalthrotle (somphing like 80C-85C max temperatures) but in cinebench i've got like 1000 pointa and i can feel the lack of power even if clock are the same as the previous settings.

The second try to solve problem is to add vcore negative offset to lower that creaz power and since svid behavior on typical was too low (every next like "base case scenario" is lowering vcore even lower and that couse same low cinebench point) so i've started from leaving it to "worst case scenario and applying core voltage offset in XTU (need to turn off voltage protection in bios and turn off wirtualization so that software can work). Anyway for now even with - 0.0800 i dont have any lower temperatures. System is stable and cinebench is generating same 1900 points (more or less but still its not 1000).

For now i dont know if the direction that i'm following is ok since i can try to lower and lower that offset but still there schould be some more easy way for every user to get this to work in normal way.

If someone got any ideas of how to lower this temperatures without losing performance please write.

[cul8rmom1]

14th Gen, Sexy :D.  I read a couple of place that the intel proc mount actually bends the CPU and causes hot spots even if the cooling device is installed perfectly.  I think I saw a video about it on LTT.  This guy wrote quite a bit on intels site.  https://community.intel.com/t5/Processors/I7-14700k-strange-spikes-of-temps/m-p/1588961.  Here is a frame that supposedly corrects the issue.  https://www.thermal-grizzly.com/en/cpu-contact-frame/s-tg-cf-i13g.  I have one but have not installed it yet.  Very specific torque requirements, recommendations, they have methods to do it without an expensive torque screw driver. I just havent wanted to hassle with it yet.  Hwinfo can give you individual CPU core temps.  Work meeting I will keep an eye on this and would like to know your thoughts.

[Pix2186PL]

Hi @cul8rmom1 

In my case installing custom frame is not possible - Arctic Liquid Freezer 3 is able to mount only with added contact frame (its not perfect one but it is better then stock on motherboard). So they are allready aware of that problem and added it to product but that block option to replace it with anything else.

If i'm gonna focus only on my cooling solution and searching problem there the options are:

- bad mounting of included contact frame (ther are no information about torq power when screewing frame or even mounting pomp)

- bad paste implementation (thermal grizzly kryonaut) and it is not the first time i'm doing it - prefer to spread paste all over the cpu by thin layer

- bad fans connection (fan from radiator are connected to CPU_FAN, pump to CPU_OPT, and vrm fan to CHA_FAN2) - i'm thinking to swap that pump connector to AIO_PUMP)

- additional since my fans on radiator are stock and dont have rgp i wanna light up a bit inside of my case and allready ordered 3 pack of asus tf120 so i'm gonna replace them (from the spec they got a bit more rpm and more CFM)

[Pix2186PL]

Hi All,

Today i did some testing since when i started Forza Horizon 4 on my pc at the lounch system started to throtle but in game temps was fine (looks like work that is done when strating game is more demanding then game :p).  Anyway i found in some Youtube guide this post from reddit:

And since i've notice that on that svid behavior i've got very nice temp under load but low performance i could give it a try - i've got mainly same result without that limits and got like 1100 point in C24.

Second attempt was to apply this settings from new official information from intel (i did load defoult setting and apply this):

That ended with standard thermal throtle after 1 min of C24 and i didnt even wait till the end of test since dont wanna cook procesor but point looks like allways (near 1900).

In that case i'm still at the point of start and dont have more ideas - tomorrow i will try that setting that posted previously here.

[Silent_Scone]

Launching FH4 would have caused higher thermals as the game conducts shader compilation. This process is quite taxing on the CPU and can pull more current than typical gaming workloads. Any behaviour witnessed under Intel Profile settings is due to following Intel's spec more closely, including loadline parameters. This results in higher temperatures than what users were accustomed to previously as motherboard vendors were under-volting the CPUs to increase thermal margin and performance.

[Pix2186PL]

Hi,

I'v manage to do some testing today and i think it's going in the right direction. My goal was to test different bios settings and compare temps, Cinabench score and looking on voltage on cpu. Bios is fully updated to 2402:

1 - Bios load to defoult  (no XMP and MCE disable all limits from the stock) everything not touched:

after 3 min of Cinabench 24 (1829 points) i've allready notice hight temp and throtle so didn't continue to proceed with test

2 - added XMP I and changed all fans from automatic to pwm since i wasn't sure if that is any different:

again after 3 min of Cinabench 24 (1874 points) same results so it looks like xmp profile is not doing any special changes that resolve cpu temps

3 - next i've added power limits and going back to autodetect fan settings:

it looks like power limits is not the problem and it is not the source of the constant thermals but at least this time non of the p cores was on 96C but overall cpu temp spike to 100C - Cinabench got after 3 min 1873 points

4 - next one was to check svid behavior settings since it direcly changes voltages so i wanna check how this work - first was change it from auto to "worst case" :

still the same and Cinabench after same 3 min got 1833

5 - svid behavior "trained"

this one started to do somphing since temps got way lower but Cinabench got poor 855 points

6 - svid behavior "typical"

looks like the same as "trained" but got a little more points 974 - still it's way to low but at least temps are nice

7 - svid behavior "intel save fail"

on that one we are back in hight points 1862 but temps are allso back - and that vcore value is soo hight

At that point i wanna start with somphing that i can work to apply negative offset in order to keep permofmance and lower temps. Since on svid behavior set to auto i wasn't able to do anything with it (even value like - 0,160 didnt do anything) i started to work with "worst case" scenario since it was the last hight score setting with relative temps:

8 - Svid behavior „worst case” and offset – 0.0500  and since i wanna try later to modify that in XTU i've remove voltage protection and turn off vmx since without it XTU wont work:

at this point it was the best results for me - after 3 min of Cinabench 24 got no thermal throtling and got 1896 points - so next i've try full 10 min test on Cinabench 23 and see how it go:

Again it looks fine to me even if there was some thermal throtling - 33531 points was allso nice even if i think there schould be like 35000 if i'm correct.

So since the offset was only 0.05 i was thinking to go more....

9 - Svid behavior „worst case” and offset – 0.0800:

After full 10 min on Cinabench 24 i've got 1921 points but pc throtle and got some hight temps but everythings spikes like after 7 min of test when on previous settings it was like this after 3 min.

So my big question to someone that know this better - schould i continue to test with more offset till the time that system will be not stable or schould i go in some other direction. Looking forward for some guide 🙂

[Pix2186PL · Accepted solution]

Hi,

I've ended up with setting offset to - 0.1100   (on 0.1200 Cinabench stoped in the middle of test and crash so it looks like it that is all for me). So to summarize everything and write SOLUTION for my problems:

1) Bios - load to defoult

2) enable xmp I

3) CPU Core/Cache Current Limit Max "307"

4) Long Duration Package Power Limit "253"

5) Short Duration Package Power Limit "253"

6) set svid behavior to "worst case" scenario

7) Global Core SVID Voltage to "Adaptive Mode"

😎 offset mode sing to "-"

9) offset voltage to "0.1100"  (for others better start with "0.050", test stability with C24 and then lower more by another 0.010 and repeat till error/crash

Additionaly i've replaced stock fans on my Arctic Liquid Freezer 3 360 to tf120 argb (from the spec they got better air flow so i wanna give radiator some more options to cool down. And spare fans was added to the case (Asus GT520) where on the bottom 3 and 3 side are intake and the rest (1 on back and 3 on radiator) are exhaust. I coud change it and mount radiator on side in intake configuration that could lower my CPU temps but after that the temps in case will be highter and GPU will have worst temps.

After all that changes and can do full C23 test without any thortling with score 33849 (if i'm correct it schould be around 35000 for i7 14700k)

If someone got some advice please reply

[Pix2186PL]

Hi,

Today i've manage to update Bios to lates version that include new microcode 0x129 and actually i'm fine with changes that it made:

I've lost a bit of performance but the temps are now looking ok - at least after full 10 min of C24 it didnt throtle.

Note for all that even saving all settings in bios to separate profile it didn't keep it after bios update so its better to write it somewhere to apply it again.