i7-8700k 4.9Ghz 90ºC improvement advice?

[WooT-PT]

Hi, i'm new on overclocking, but i have watched several videos on youtube and saw a lot of threads on this forum, but i'm still afraid of touching something i'm not supposed too. So i came to ask your advice.
I would like to overclock my CPU, GPU and RAM, but not to the extreme, just some improvement without taking much of the life-span.
So i'm currently working on the CPU and I'm with a stable 30min test each (RealBench, Cinebench, 3DMark, Aida64, Prime95) 4.9Ghz 1.320v.
I created a table for me to easily see what i have set, where tests failed and what were the results. You can see it here:

https://docs.google.com/spreadsheets/d/1-onDFtN3Rz3EJaM6TSlnTPa0pOy26dikh2XIzL2TlF8/edit?usp=sharing

And this is my PC built:

Case: Nox Hummer​ TGX
OS: Windows 10 Pro
MotherBoard: Asus - ROG STRIX Z370-E GAMING ATX LGA1151
CPU: Intel - Core i7-8700K
CPU Cooler: Corsair - H115i
Thermal Paste: Cooler Master - MasterGel Maker
GPU: Asus - GeForce GTX 1080 Ti 11GB STRIX GAMING OC
Memory: G.Skill - Trident Z RGB 32GB (2 x 16GB) DDR4-3200 CL14
Storage: SSD - Corsair - Force MP500 480GB M.2-2280 + 2x HHD - Seagate - BarraCuda Pro 2TB 3.5" 7200RPM
Power Supply: SeaSonic - PRIME Platinum 750W 80+ Platinum Certified Fully-Modular ATX

So my question is:

I managed to get it to 4.9Ghz, 1.320v stable, but its getting to 90ºC... Is there anything i could change to make it stable but not reach those temperatures?

Edited: I just remembered to make a Small FFTs test on Prime95 and it 10 out of 12 Workers crashed on the first 5 seconds, besides reaching a temperatures of 95ºC, so apparently not stable at all.
Can't get more voltage or i'm going to get higher temp... What should i do next?

[Carlyle2020]

Hi WooT-PT,

concerning temps you should be aware that XMP tweaks some voltages to ensure maximum compatibility. Many users state that they can run lower voltages by manually setting the values lower than XMP would.

The safe route would be getting an (old) usb stick, formated to fat32 and placed into the "Bios" marked usb slot for the capability to make screensshots within bios with F12 i believe. Kindly post those screenshsots. They will reveal the running voltages as well as your present config. Both helps us giving you a hint or two without you having to play with values you are not yet comfortable tweaking alone.

ps a pic of your case will help as well
My stomach tells me you may want to check thermal paste application due to the fact that the temps ramp up a little too fast for at least my taste
Check for air bubbles as well if you are at it please

For temp analysis i suggest HWinfo

Thanks in advance

Kind regards
Carlyle

[WooT-PT]

In Response to Carlyle2020 and NemesisChild, since you have some similar questions:

I forgot to say that i'm using HWMonitor for Temp info.
I applied TIM by placing tape on all 4 sides and creating a thin perfect square layer covering the full size of the CPU (and of course removed the tape after), i did this with a guy that worked on a PC store before and is used to build up PC's more the fact that i like to make this things perfect (perfectionist?). So i'm quite sure that is well applied but if you think i should, i will do it. Any specific way to apply it?
My ambient temp can be bettween 10ºC to 35ºC, but at this moment it is at 23ºC.
I know my case isn't the one with best airflow especially since i need to replace the front fans of the cooler (push) because i can not control them, i can just control the pull ones. I'm gathering money to buy a full kit of corsair fans with controller for my pc and that way be able to control the push fans at the same time as the pull fans.
The CPU Cooling temp with PC at idle normally is at 33ºC and during stress test stabilises at 38/39ºC max with the current config.
I have to be honest, those high temps normally are at the beginning, like a spike of temp and then it goes back down and goes really slowly from a +/- 70ºC temperature to the high temp and it normally does not even reach the high temp registered before (at least on 45 min tests).
About the Delid i have already studied that possibility and i'm thinking of doing it, but i can not sent it to EUA to make it since i'm from Portugal or it would cost the double or triple of buying the materials and doing it myself. But i simply can not do it right now.

Here are the BIOS screenshots as requested:

]





Case Pics:



CPU Cooling Airflow Config:

[NemesisChild]

What is your ambient temp?
How did you apply the TIM?
What is your case and CPU cooler airflow configurations?

This is a great way to drastically improve CPU temps and it's on sale now:

https://siliconlottery.com/collections/sl/products/delid

[Carlyle2020]

Thanks WooT-PT,

I will split in things you can switch to right of the bat and things to set and then tweak and test:
Things to change for good:
svid behaviour to best case
CPU Core ratio to sync all cores
The option you chose here “syncs” as well under load. But it starts sync by reading the lowest CPU and syncs the rest to it and THEN ramps up – “somebody correct me if I am wrong”
CPU current capacity 140% (‘cause there are other saveguards in play)

Same goes for the three Turbo mode parameters: type 99999 each and see what happens
Talking about Turbo mode parameters:
The last two are managed and already set by the above mentioned “svid behaviour”
:cool: But let’s ocd the XYZT out of it and put: 0.01 in both

--------------------------------------------------------------------------------------------
CPU: manual is not bad. But half of your setiings are set to power saving and with 32 Gig you will have a lot of benefits using adaptive mode and putting in your value under "additional turbo mode", which will only appear after changeing from manual to adaptive mode.

The board will thank you.

XMP
Since your goal is to lower temps I strongly suggest setting up your RAM manually and therefore ditching XMP which also ramps up “CPUVCCIO voltage” and “CPU system agent voltage” a little too much in many cases.

SetRAM manually to 14 /14/ 34
your DRAM voltage is already manually set to 1,3530 i believe
If the setting is stable you will still have to further check I suggest.
Lowering this setting in your case is almost impossible due to the 32 Gig.
You may even have to bump it up if nothing else gets the system stable.

Now we get to things you set manually and have to test:
a) For stability
b) test with cinebench to find the sweet spot (run CPU bench 3x)

I personally have found out that “CPUVCCIO voltage” and “CPU system agent voltage” have sweet spots that let my cinebench score soar upwards up to 60 points.
Any miniscule change in voltage (up or down) and I loose 60 points.

CPUVCCIO voltage 1.10-1.25 max / try: 1.2

This setting concerns the RAM controller, which is located in the CPU.
You will need a lot due to the 32 Gig.
Any small voltage setting drop here that is still stable will lower temps.

CPU system agent voltage 1.10-1.25 max / try: 1.2 and work your way down

Just because I told you to same min/max levels and starting value does not mean you will keep both settings the same!
I have tested these settings extensively in many small variations to achieve a final score of 1678.

! Be aware that cinebench scores vary pretty quickly if one does not have exactly the same background tasks and foreground tasks open.
Basically, you will not be able to install stuff or test once fresh after restart and once after using an email software or playing a game.
Monitoring software lowers and sometime screws up the results by using up resources inconsistently during the tests. So best shut down everything you can and always use the same routine until having found the sweet spot.
Be also aware of nasty automatic updates in the bachground after a fresh restart. Best to disble the network while testing.

Thermal paste application:
Any technique has other variables to consider.
Your technique is perfect for coverage. It is also vulnerable since getting the thickness right beforehand is far more important than the old school method.
Stick to it if you like but remember that thermal paste is rather a “filler” and will underperform as a “layer”. Do it again until you mastered it with only a drop.

IF you want to compare:
Put a drop in the middle. Screw the heat sink on about 60-80%. Now the idea here is to have pressed the drop into a pancake. Coverage is done. Now you want to expel as much excess paste as possible, since a full layer would rather isolate due to the fact that metal-on-metal action transfers heat better than any metal-on thermal paste-on metal – action could do.
We therefore have to give the paste a “way out” by screwing it out.
Check your placement a last time since the first screw gets screwed in to its fullest. Now go round the clock and screw slowly tight.
Go around one last time, but be very very sensitive and accept the fact that a screw may not move this time.
That’s how I do it.
The applied drop has only to cover the real Cores in the middle, not the whole lid btw to work optimal. Freaks people out but hey 😉

Now I’m hungy so I’ll end here.

[NemesisChild]

I set VCCIO at 1.15v and SA voltage at 1.20v for all overclocking profiles so far.
I've been doing the grain of rice size TIM application in the CPU center for many years, always produces favorable results.
Airflow is critical and is the primary reason why I don't run ACX graphic cards in my builds. All they do is circulate hot air within the case.

[WooT-PT]

Thank you Carlyle2020 for all the tips

I ran RealBench, AIDA64, Prime95 and CineBench. It failed at RealBench and on worker of Prime95 (Table updated)
https://docs.google.com/spreadsheets/d/1-onDFtN3Rz3EJaM6TSlnTPa0pOy26dikh2XIzL2TlF8/edit?usp=sharing

What settings should i change?

This is the new configuration as you told me (Tell me if you see something wrong):

[Carlyle2020]

Well next you have to make a choice.
Your data is not conclusive (before and after) concerning stability of:
CPU overclocked
RAM overclocked from JEDEC 2133 to XMP of 3200

the rulebook would state to set default/auto for CPU (most stable setting normally/ reading out VID under load to have a setting to start manually with and then dial in the RAM with default/non OC settings which are still only 2133.
then dial RAM to 3200 stable. After that one may touch the CPU.

So, one tweak at a time normally. That would be the sane choice.





... .. . 🙂
And now let's make it run and then worry about perfection:

CPU:
Bump it up to 1.4, which many advise as max for 24/7. Just to see if one change can let you pass your tests.
If yes: work your way down
If no: change to first setting and leave until you check your RAM

RAM:
step one: double checking if somebody else has your RAM (its description in full as a search word in this forum, after that manufacturer name and Gb size)
check their settings with yours. save your profile/change settings according to the users profile/test and write down stuff/maybe revert
If above don't help:
step two: the RAM can be tricky with CL14 at this level. Bump voltage ever so slightly to 1.3550/ maybe even 1.3555 - quickly double check with manufacturer pages/forums if you want to go higher +
Give VCCIO a bump to 1.25 / 1.30 / 1.35 - those would be numbers you will have seen from other 32gb users as well at this point
Use with caution due to great heat generation.
If above don't help:
Time to remove one stick and start from the first bios save. rinse and repeat.
If above don't help:
food, a bit of social contact & afterwards a hardware sacrifice to ze tech gods!
Throw away one of those wonky fans that you haven't thrown away but never will use again. Or an old cable for example 🙂
Reseat RAM. Start again with the safe step-by-step overclocking procedure i mentioned in the beginning.

Rgds
Carlyle

ps you can do this with all cores synced, except for reading out VID with std settings