10th gen Intel a lot of beginner questions for start

[Kapucha]

Good morning

I'm trying to understand how 10th gen Intel should be overclocked. My goal in the first part is not to overclock but to understand basics and I have questions related to it so I have good starting point. I've read a lot of articled about how to overclock but everything is in my opinion difficult when You need to start doing it because everybody have different mainboards, different names in bios of similar settings. Ok but first some basics that I have problem with...

1. Where to find in Intel specs what are save voltages for CPU. I have 10th i7 10700KF to play with and I just can't find on Intel specs what is default voltage for this CPU. Looking at different models there is also totally no info about it... I'm trying just to understand what is correct vcore value. Reading from HWiNFO64 value from Vcore shows totally different value than in BIOS. Also Intel XTU is showing another different value from sensors and I'm quiet lost here. I was expecting Intel XTU will tell me what is default value for this CPU and this software is telling that this value is "default".

2. I'm trying to understand my LLC in my MSI motherboard. Currently the setting is set to AUTO which don't tell my anything. But to understand it I need to know where to look for reliable Vcore readings in idle and full load. And this is totally not obvious as I described in 1st question. Why for example in HWiNFO64 Vcore is showing me in idle ~0.833v and during load (avg) 1.255v. From this I can't say anything about this AUTO mode in LLC. From my understanding IN IDLE we should have larger Vcore than IN LOAD and here is totally different. Or maybe Vcore is showing something totally different. I have MSI MEG Z490I UNIFY. I many different programs I see totally different values:

Values IN IDLE
HWiNFO and CPUID CPU-Z MSI Vcore = ~0.833-0.980V
BIOS Readings Vcore: ~1.044V
Intel XTU VCore: ~1.22-1.23V

Also what is in the BIOS:
CPU SA Voltage which is set to 1.050V
CPU IO Voltage which is set to 0.942V

Trying to understand all this relations. The goal for now is to understand what AUTO means in LLC and how it works on my motherboard and to set some specific Vcore in bios instead of auto so I will have some point of reference... I was reading CPU data sheets but man there is so many pages in this PDF's but after one hour I didn't find anything about recommended factory vCore values...

Looking for support and good day
Cheers.

 

[Zerileous]

These haven't changed substantially since the 8000 CPUs. What you set is not what the VRM will actually give under load. You mention LLC, this is how you adjust this, in order to get the desired voltage.

In XTU I'm guessing you're also reading a set value, instead of a get value. Idle voltage will be lower as the clocks are lower. If you did not have any energy savings enabled so the CPU ran a constant clock and constant set voltage, then yes you would see the vcore droop under load. Instead the behavior you're seeing is the CPU increasing clock and voltage together with the load. That said you will note Vdroop when the voltage read out by HWiNFO etc is less than what you set in the BIOS or XTU.

In most motherboards AUTO sets the LLC to the lowest setting. LLC basically increases the set voltage under load to compensate for the droop. You can try a medium setting for starters. Some brands the BIOS will use a large number to represent MORE voltage from LLC, other will use a large number to represent LESS voltage, so here it's important to read the motherboard manual to understand what direction you're going in.

What cooling are you using? It's likely that you will run out of cooling before your voltage becomes unsafe. I don't do a lot of Intel, so I'll let someone else give you the specific numbers.

 

[EarthDog]

Here is our guide ot start you off.
https://www.overclockers.com/how-to-overclock-the-i9-10900k-a-guide-for-taming-the-beast/

As far as the questions...

1. You'll run out of cooling before you hit an unsafe voltage, really. A 3x120mm radiator can handle about 1.35V give or take while keeping the CPU around 90C in stress tests. Correct Vcore value depends on the software, sadly. Some don't read the right sensor or isn't reading the sensor right. What does CPUz say the voltage is? Remember, there isn't a default, per say... or more accurately, there is a default for each multiplier. So when it hits 52x on auto, the CPU knows what it should get... so it varies.

If you don't want to use a static value, you can leave it on auto and use an offset/adpative that adds/takes away voltage from each multiplier.

2. Auto mode generally tells you nothing about LLC. Set a specific LLC mode and see what it does to a STATIC voltage. Set 1.3V in the BIOS, boot to Windows and load it, see what XTU says (seems to the most accurate?). Does Dragon Center/MSI Command Center display Vcore? Is your BIOS updated to the latest version and you're using the latest versions of the programs, you listed?

 

[RJARRRPCGP]

In most motherboards AUTO sets the LLC to the lowest setting.

For Asus, "Auto" seems to be the same as disabled.

 

[EarthDog]

For Asus, "Auto" seems to be the same as disabled.

Auto I feel is disabled if it doesn't actually have a disabled option. I also feel this way because vdroop is actually a part of the Intel specification.

 

[Kapucha]

Here is our guide ot start you off.
https://www.overclockers.com/how-to-overclock-the-i9-10900k-a-guide-for-taming-the-beast/

Thanks. I'm reading it right now...

If you don't want to use a static value, you can leave it on auto and use an offset/adpative that adds/takes away voltage from each multiplier.

That was it. I just wanted to set Vcore to static but on MSI MB there was no such option. There was option name Override but I was not sure if this is dynamic or STATIC? I tried it and suddenly Intel XTU show me default factory Vcore for my cpu which was: 1.331v

From this point I understand that the CPU in adaptive mode was deciding how much volts it needs and in idle this value was very small... OK Override option seems to be STATIC option.

After this is set I started to looking in HWiNFO similar value in idle as BIOS is displaying and I found it. There are two values named Vcore and one from VRM named VR VOUT. Vcore is showing a lottle bit more but VR VOUT was showing exactly what BIOS was displaying. So this it the right CPU voltage and it is changing as expected according to load. The more load the voltage is dropping.

I was reading that this is VRM reading and it is called DIE SENSE. There are also VCC SENSE and SOCKET SENSE. From what I understand DIE SENSE is the most accurate and this is VR VOUT reading. Also there is a lot of other info from this VRM like power out and current which is freaking awesome.

So everything now is working as I expected. I just needed to disable AUTO in CPU Voltage Mode and switch it to Override and then I discovered the rest for my own. Vcore is dropping as expected with the load. No Overclocking now...

Auto mode generally tells you nothing about LLC. Set a specific LLC mode and see what it does to a STATIC voltage. Set 1.3V in the BIOS, boot to Windows and load it, see what XTU says (seems to the most accurate?). Does Dragon Center/MSI Command Center display Vcore? Is your BIOS updated to the latest version and you're using the latest versions of the programs, you listed?

MSI really like to use [AUTO] almost everywhere and it is not described which option is hidden under this AUTO which is quiet problematic. I discovered that [AUTO] in LLC means option 3 which is almost flat with little drop. I was checking it by using default voltages and I was observing vdroop in full load in Cinebench R23.200 2 minute test.

Results in LLC set to AUTO and 3 are the same:
Vcore in Bios set to Auto = 1.331v
IDLE VR VOUT = 1.327v
LOAD VR VOUT = 1.310v (POWER approx: 211W, CURRENT approx 161A, Max Core Temp after 2m Cinebench test 96C !!!)

So FYI if somebody is looking for info. In MSI LLC1 is Extreme and LLC8 is a lot of vdrop under load. LLC3 seems to be very neutral and default.

Now please tell me if my readings are fine. Temperature is really high 96C after 2m stress test?

I tried to set clocks to 49x on all cores without any AVX offset (0x) and I found the lowest voltage that is needed under load to be 1.181v. Below this area under load system is not stable during load stress test. So now my settings are

Multiplier 49x, AVX Offset 0x
Vcore in BIOS = 1.200, LLC3
IDLE VROUT = 1.195v
LOAD VROUT = 1.181v (POWER approx: 154W, Max Core Temp after 2m, 10m and 30m Cinebench test = 81C)

Are above readings looks good? Are they fitting in statistics?

The cooling that I'm using is Noctua NH-D14.

I'm not sure if this is good idea to try to go to 50x without AVX offset at all with such temperature. I think I will need to raise voltage and temps will be much higher... I will try but everything above 90C will be not good sign for me.

Thanks a lot for suggesting to set Static voltage. It really helps me out start understanding everything.

 

[Zerileous]

The cooler is about as good as you can get with air. 81c is fine for CB. During a stress test up to 90C is okay, Your 96c at 1.331v, I'm guessing you're cooler can handle 1.25v roughly, but go ahead an experiment until you find what you like. Multiplier won't increase temperature on it's own, it's the voltage that does that. Voltage is required to be increased in order to stabilize the CPU at a certain multiplier, the higher the clock the more voltage required.

CB isn't commonly used as a stress test. I don't know if it has robust error checking, or if it just makes the CPU really hot and possibly crash. Other stress tests will detect more nuanced errors without having to force the PC to crash necessarily, and you may find a more reliable overclock from that. You can download AIDA64 and use it during the free trial to stress test, and then decide if you want to buy it after you have dialed in your OC. Another option is Prime95 small FFT runs. However this will make more heat, especially when using AVX, and is a less realistic workload. But it's free. It is somewhat challenging to disable AVX (I believe either running an old version or modifying a config file will get you there).

 

[EarthDog]

CB doens't error check. It runs in seconds (multi thread) and isn't really a stress/stability test... just a benchmark.

 

[Kapucha]

Based on the observation I know that CB is making my CPU much hotter than Prime64 for example. What stress/stability test do you recommend? I just can't find anything that is stressing this CPU more than CB. CB also is using a lot of AVX instructions.

 

[EarthDog]

If CB is running hotter than P95, you're likely running the blend test (tests some CPU, a lot of memory) or disabled AVX. I keep AVX enabled as there are many things that can use those instruction sets....

 

[RJARRRPCGP]

Hotter than Prime95?=That metal, usually goes to Linpack!

 

[Woomack]

If CB is running hotter than P95, you're likely running the blend test (tests some CPU, a lot of memory) or disabled AVX. I keep AVX enabled as there are many things that can use those instruction sets....

Someone said that new games, like even highly bugged cyberpunk, are starting to use AVX.

Based on the observation I know that CB is making my CPU much hotter than Prime64 for example. What stress/stability test do you recommend? I just can't find anything that is stressing this CPU more than CB. CB also is using a lot of AVX instructions.

CPU stress test:
Prime95, any new version - small FFT will give you about max temps
AIDA64 stability test - CPU+FPU test, about the same as Prime95 but usually gives slightly lower temps

I'm typically running 1-1.5h CPU stability test as usually after about 30-40 mins nothing is changing (there are exceptions). After that I run multiple tests based on 3D or mixed load. It helps if you have paid version of 3DMark as there is stability test which can be run in loop. You can use something else or just manually run more benchmarks based on games. Final Fantasy XV benchmark is free and reacts well to any instability (even faster than some stability tests).

 

[Zerileous]

A stress test isn't about how hot you can get your CPU. In real world application you'd never reach those temps unless you're doing something like CPU folding or rendering. You're not testing the CPU cooling. You're testing the stability of the overclock (the ability of the CPU to perform accurate calculations with the clock and voltage settings applied to it), and monitoring the CPU cooling for adequate performance. A hotter stress test is not a better stress test.

 

[RJARRRPCGP]

At least a while ago, Linpack was the best test for Intel CPU core stability. It would bomb out in a short time, usually, if unstable at all!

While Linpack is likely to have a false negative for finding non-CPU-core-stability issues, especially related to bus ODT. Prime95 in blend is what you would use for testing bus-ODT-related stability.

 

[EarthDog]

I stay faaaaaaaaaaaaar away from Linpack. If the other stress tests are overkill, why go harder? AIDA64 and some Realbench does what I need in a timely manner.

 

[RJARRRPCGP]

I stay faaaaaaaaaaaaar away from Linpack. If the other stress tests are overkill, why go harder? AIDA64 and some Realbench does what I need in a timely manner.

I only use Linpack for a short time to test core(s) in the past on Intel chips. It's known for failing extremely fast when there's not enough Vcore. Prime95, at least in the past, takes a lot longer to find anything wrong, but got 0x00000124 BSOD in not too long on a Q6600 G0, because of an FSB ODT issue, by the looks.

Linpack is super when it comes to testing core-only OCs. (with CPU multi or bus and RAM underclocked or at stock)

But I'm aware of Linpack possibly having false errors on AMD CPUs!

 

[Kapucha]

Thanks guys for help.

I finished to test stability with Prime64 torture test and running in the background Heaven Benchmark. You were right. I was using blend test in Prime64 before. Now I'm using torture custom test with min/max FFT size 32. It really kick *** to the CPU.

My settings was of course not stable because right from the beginning I got calculations errors... So after several approaches/blue screens my settings looks like stable after 12m Prime95 + Heaven Benchmark Looped.

Prime alone is stable even under 1.185v but when I add Heaven to it it is not stable at all in 1.180-1.190 area...

LLC 5 & Multipier 49x without any AVX offset.
1.280v set in bios. Voltage mode Override.
1.267v IDLE Vcore
1.197v LOAD Vcore <- I found this is stable after 12m of tests.
Max Power: 233W
Max Temp: 94C <- This is high I know. Just one core reports this the rest are under 90C. I'm thinking what can I do with this...

Thanks Guys again for help. You are great.

Merry Christmas!