Help overclocking F4-4000C19D-32GTZR on asus maximux XI extreme with i9 9900k

[hq-hq]

memtest will run on windows 10,

i give you an example picture (i have six cores 12 threads, i use 6x memtest for high load stability test and 12x memtest for extreme load stability test -> you have 8 cores so you can check with 8x or 16x)

[marcolisi]

Please bare with me because this is all new to me. Most of what u are saying is like a different language for me.

what do you want me to do next?

What should I set for 4.9 GHz and start the first testin with prime/ycruncher ?

I installed ycruncher. Where do I set the options of your screencap?
The program asks me this

[marcolisi]

As for cooling, I do not use water cooling. I have the noctua nh-d15

[marcolisi]

For Vccio and vccsa, they were set to auto. So I think you saw those high value because maybe the values change based on the usage of the pc? Higher use of the pc, higher number ? Could that be a possibility?

I have set these now:

VCCIO = 1.20V
VCCSA = 1.30V

and the pc boots

[marcolisi]

How do I use memtest?

does it run In windows ?

[hq-hq]

so one more thing, cpu load-line calibration, asus uses from level 1 to 8

standard LLC means let`s say 1.20V at low cpu load (not zero) and 1.10V at full load
for oc it is better to use LLC where you have (example) 1.20V at low load and 1.16V at full load
maybe your bios gives you some explanation, i think LLC lvl 8 = flat line means you set 1.25V VCore you get 1.22V at low load and 1.22V at full load
this is ok if you stay away from 1.40V VCore, so i think with soft oc up to 4.9GHz you can use LLC 5-7, maybe LLC 7 with a small voltage drop

you can test VCore @ 1.30V in bios and use a LLC which has a small drop - LLC 7 would be great
then let`s start with 4.9 GHz and check stability with prime/ycruncher ( let`s hope your cooling can handle real 1.25V and 4.7-4.9 GHz @ full load ) and log the sensors please - if you can see core temps like 95°C you need to reduce voltage. You can`t burn the cpu, it will throttle down at high temperatures. Maybe 1.5V VCore and >90°C will kill it over time)

[hq-hq]

if you are sure that 4133MHz @XMP Timings are running well, we can try to get higher RAM OC and CPU OC, i would say CPU is more easy so let`s start here:

AVX offset = 0

i have it at zero too, AVX workload needs always higher voltage, that is why most people use AVX offset >0 with coffee lake CPUs
let`s say 1.30V VCore @ 100% Load is good for 5100MHz allcore but at high AVX workload you need to lower your clockspeed to 4900MHz
so you set AVX offset = 2 (51x 100MHz @ standard // 49x 100MHz @ AVX)
i don`t like to see my cpu downclocking, so i have 5200MHz and AVX offset = 0 but i need to use higher voltage...
that is your decision what you want, i would say use AVX offset = 0 ( for video encoding x264 x265 avx performance is important for you )
so you need to check stability with avx workload at desired OC clockrate

you can use prime with CPUSupportsAVX=1 and ycruncher (my favourite) too http://www.numberworld.org/y-cruncher/


y-cruncher use benchmark, multicore and 2,500,000,000 (will run up to 3 minutes) and while it is running log your sensor data please

prime - please don`t use small fft

do it like this
and if you like read the article https://overclocking.guide/stability...with-prime-95/

[hq-hq]

ah, ok,

so let`s check the data:

CPU Clock = 4.7GHz allcore (never drops so i think AVX offset = 0)
Uncore = 4.3GHz

As i can see at ~74% cpu load:
CPU max temp = 87°C ( no water cooling ?)
CPU package power = 148Watt ( no TDP limit in bios active )
VCore @ high (not full) load = ~1.25V ( i would say a small "overvoltage" (above stock) from the board maybe because of AVX offset=0 )


DRAM Voltage = 1.4V -> ok (you can go to 1.45v without risking anything, some would say till 1.5v)
VCCIO 1.30V -> very high! ( we should reduce it)
VCCSA 1.40V -> extreme high! (we should reduce it)

DRAM Temperature = ~40°C -> good


hm, i think you don`t have much headroom for OC and higher cpu voltage,

first we should get the VCCSA and VCCIO voltage a bit down,
so please try

VCCIO = 1.25V
VCCSA = 1.35V

if it is working please go a bit lower to

VCCIO = 1.20V
VCCSA = 1.30V

and check then with 8x memtest till 100%

lowering these voltages can lower your cpu max temp a little bit

[marcolisi]

This is the HWiNFO64 results of the pc how it is right now

https://www.mediafire.com/file/ukhiazrb0fpk74c/TEST.CSV/file

This is while I am running video conversion and downloading

[hq-hq]

in the first step i would check if cpu without oc and ram @ 4133@XMP Profile is stable

for a quick check you can open 8x memtest, set each to 2000mb and start them, let them all run to 100% and check the physical memory load in HWiNFO64 - should be 80-90 % (in my experience that is enough to be sure your games will not crash, countercheck with linx it should run at least 15minutes without error, better is 30min)
hardcore check is 16x memtest each 1800mb and physical memory load ~99% an let it run to 200%

and post your memory benchmark scores with aida64 [ Tools / Cache and Memory benchmark ]

[hq-hq]

very good!

you need:

HWiNFO64 for sensor monitoring (load of mem controller, voltages (vcore, vccsa, vccio, vdim), temperatures (voltage regulators on your board, ram sticks temperature), wattages ... https://www.hwinfo.com/download/
CoreTemp for simply CPU core temperature and wattage monitoring https://www.alcpu.com/CoreTemp/
Aida64 for benching your RAM (latency & bandwith) https://www.aida64.com/downloads [use google for registering]
Memtest (to create mem controller load for stressing and stability check) https://www.hcidesign.com/memtest/

Prime95 for stressing cpu, cpu&ram&uncore, testing avx & fma3 impact https://www.mersenne.org/download/
(use newest version, you can activate and deactivate AVX & FMA3 this way:
edit local.txt in the prime folder and add following lines to disable:
CPUSupportsAVX=0
CPUSupportsFMA3=0
-> you can test now with or without AVX & FMA3, =1 means enabled)

maybe linx, good tool for checking long term cpu/ram https://linx.en.lo4d.com/

corona benchmark for comparing your build with other i9 9900k CPUs https://corona-renderer.com/benchmark

example: https://corona-renderer.com/benchmark/results/cpu/8700k

in your case: https://corona-renderer.com/benchmark/results/cpu/9900k

so we aim for 1:30

[marcolisi]

ok. I am wrong. It worked ! I set the bios to xmp I and it works with all the auto settings :-)

Would you help me to optimize it please and then maybe to overclock my cpu ?

Thanks!

[marcolisi]

I have got the new ram.
I have set it to 4133 mzh and everything else in auto and it is not booting

[marcolisi]

I have ordered the Let's hope it will run as advertised.

G.SKILL TridentZ RGB Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 4133 (PC4 33000) Desktop Memory Model F4-4133C17Q-32GTZR

As soon as we confirm all is good, I will return the other ram

[hq-hq]

in the past i did some tests in cpu limiting settings, with Gigabyte Z370 HD3P i was limited to DDR4@3866MHz so i did some comparison between 2666MHz stock clock (Cl 16 16 16 30) & 3866MHz (Cl 16 16 16 30)

here is one example with my i7 8700k (4800MHz uncore), as you can see there is a high benefit on RAM OC and CPU & RAM OC together makes a big difference in Farcry 5, average fps and min fps


(as far as i can remember i used 5 runs and arithmetic average, GTX 1080Ti @ 2076/6264MHz)

one more example between DDR4@3866MHz (Cl 16 16 16 30) and DDR4@4133MHz (Cl 17 17 17 32) in compressing rate with win rar (5.50)


i7 8700k@5300MHz DDR4@3866MHz:

i7 8700k@5300MHz DDR4@4133MHz:


3800MHz VS 4133MHz
workload which needs high bandwith you gain some percent
games which benefits from low latency it will push the min fps some percent if they were caused by cpu limitation

your i9 9900k has 2 more physical cores so it will benefit more than my old i7 8700k from higher bandwith ( it is harder to feed 8 cores with the same bandwith than 6 cores )