Brand new PC not sure I understand the correct BIOS tweaks, 14700k

[logan_sudopwr15d302e5]

Went way over budget and i'm excited but also stressed. I want this processor to last awhile after how much of my savings I just spent but just want to make sure I make all the right BIOS tweaks ASAP to keep new processor from degrading. If you can just bear w/ me and give feedback so I know i'm understanding what steps I need to take i would be grateful.

i7-14700k paired w/ MAG Z790 Tomahawk MAX WIFI and alot of other stuff.. way over what I intended on spending.

So my 1st step needs to be Flashing to the most current BIOS before seating the new processor. I wanted a MOBO w/ BIOS flashback so had to spend a little extra.

Then after post I need to set PL1,PL2 LIMIT 253 Watt
set ICC MAX 307a
Intel Turbo Boost 3.0 - Disabled

Set Lite Load to something like 9 and turn off IA CEP, IA CEP for 14th gen - Both DISABLED

Now I can load Operating System, Windows 11, make sure Power Plan is set to Balanced and do a couple tests like Cinebench23 and 15, monitor voltages/temps and possibly go back into BIOS and play w/ Adaptive Offset - .100 or something along those lines right? I've done so much reading this past week juggled w/ my full-time job trying to understand it all but this is alot. Seen so many different settings cause of course we all have different hardware. My main concern is getting the performance I paid for and not degrading my processor so please lemme know if i'm misunderstanding something. Thank you

 

[FlyingScot]

Awesome you've been such a big help. I've got it down to Lite Load 5, voltages look OKAY.. doing something light like watching youTube i'll be like .088 volt might jump up to 1.10 then back down under 1.0 again, alot of spikes like that but it never goes above 1.32 even during Cinebench23. Temps are always 30C idle, never above 69-70 during heavy load.

Most Watts I ever see is 233 even w/ 253 PL1, PL2. Cinebench scores never go above 34,400 which seems average for 14700k I guess, i'm not as worried about the score I care about performance but just more worried about the health/longevity of processor. I'm not sure what the next step needs to be if any. Would the Adaptive + Offset - even out the voltage readings? Does it really matter? Lol all the time i've spent reading all the stuff on here and i'm still kind of lost. Alot to take in and understand and try to apply to your specific hardware but luckily I don't have much going on besides work.

All the settings we talked about i'm still using just been lowering Lite Load, no crash or anything like that yet. Lemme know if you need a screenshot or another setting i'm not thinking of ATM. Thx again

I’ll try to think of a possible next step after sun up. But you’re already in a really safe place for CPU health. You might be able to squeeze out a little more excess voltage with an Adaptive+Offset. That might work nicely in games and low impact activities. Outside of all-core workloads, like R23, the CPU LL is far less effective. However, lots of CPU fluctuations near idle is normal. It’s due to Windows background processes. If you keep HWINFO on the screen and don‘t touch the mouse, about 20mins after a cold boot it usually settles down to a flatline MHz (e.g.800) and voltage across all the cores.

Let me know what CPU Loadline Calibration Control setting you’re running. Probably Auto, right?

Watch out for unusual voltage spikes when you wiggle the mouse when system is idling. If very high (1.4V+), let me know. It shouldn’t be the case for 97% of users‘ systems. But it’s worth checking. I think Intel fixed it.

 

[logan_sudopwr15d302e5]

Yeah I left the LL at Auto just because I didn't fully understand it. and yeah I have not seen it go past 1.31 since I got to Lite Load 6 and 5

 

[FlyingScot]

Yeah I left the LL at Auto just because I didn't fully understand it. and yeah I have not seen it go past 1.31 since I got to Lite Load 6 and 5

Those are some nice and low voltages! You might still have that Raptor Lake when you're old and gray - assuming that you're not already!

Just a tip. When discussing the CPU Lite Load setting, it's okay to abbreviate it as "LL." However, when discussing the CPU Loadline Calibration setting, it's customary to abbreviate it as "LLC". That way people won't get them confused. It does happen from time-to-time. And an LL=5 is a very different animal from an LLC=5.

BTW, are you a gamer? I assume so being that you put so much thought into your GPU purchase.

 

[FlyingScot]

Hey logan,

I'm helping another forum member with tuning his new 14900K (that has just replaced a degraded 13900K under warranty) with the primary goal of lowering voltages to increase longevity - just like you are doing. I think he's a regular gamer, too. So your two usage cases should be very similar. Who knows, perhaps you can help each other out down the road.

Give my last two posts (link is to 1 of 2) a read. Perhaps it will get you thinking, and might even prove helpful with your setup. You'll just have to change values where appropriate so they relate to your 14700K. Enjoy!

https://forum-en.msi.com/index.php?...-so-bad-after-bios-update.407767/post-2308628

 

[logan_sudopwr15d302e5]

I am a little bit of a Gamer when I have time but I want to break into IT career even tho I am almost in my 40's so I am self-taught programmer and doing alot of cyber-security related stuff right now so I'll have some virtual machines running and my own DNS and Mail server running - all kinds of stuff that can get intensive. I almost went w/ the cheaper Nvidia RTX 4060 but based on everything I was reading the CPU would've been bottlenecked w/ that GPU. 4070 Super seemed like the much better fit. I had the money for it but no way I could've done 4080 or 4090 unfortunately. Guess i'll meet you guys in the other thread - Thank you

 

[FlyingScot]

I am a little bit of a Gamer when I have time but I want to break into IT career even tho I am almost in my 40's so I am self-taught programmer and doing alot of cyber-security related stuff right now so I'll have some virtual machines running and my own DNS and Mail server running - all kinds of stuff that can get intensive. I almost went w/ the cheaper Nvidia RTX 4060 but based on everything I was reading the CPU would've been bottlenecked w/ that GPU. 4070 Super seemed like the much better fit. I had the money for it but no way I could've done 4080 or 4090 unfortunately. Guess i'll meet you guys in the other thread - Thank you

Well, I wish you a lot of success in your new career. Someone famous once said “If you turn your hobby into a career, you’ll never work a day in your life.” Given your level of enthusiasm, I would guess computers are a bit of a hobby, too.

I’m glad you took the time to explain your usage case. You may well have different needs from the average Joe. You’ll definitely want to stability test using a variety of applications to ensure that you don’t get crashes or WHEA errors. The more negative offset you combine with a CPU Lite Load undervolt, the trickier it can be to stability test. Sometimes it takes several months of monitoring to be sure you are truly stable. For example, I have run into a negative offset that only caused WHEA errors during intense Windows Updates. Everything else was rock solid. The VM plus business apps is going to be interesting as to how it responds to all your tuning.

By the way, feel free to keep this thread going if you have any more question for the community.

 

[logan_sudopwr15d302e5]

By the way, feel free to keep this thread going if you have any more question for the community.

I am glad you said that. I was hoping I wouldn't have to ask anymore questions but i'm just not understanding the AC/DC loadlines trying to set my own no matter how many guides I read here on these forums. I have spent so much time messing w/ these BIOs settings questioning everything and in the end it feels like i've just wasted my time because I keep having to go back to the Easy, simple Lite Load 4.
No matter how much I play w/ the LLC AC/DC I just can't get my Vcore and VID to align. The averages will line up but the voltages are like 0.8 one sec then right back up to 1.2, 1.3 the next second back and forth and the VID, Vcore really only align during load but any other time they are both all over the place - like I said its really just the averages that line up doing the more advanced settings like picking my own AC/DC and LLC w/ adaptive offset + anywhere from -0.020 - -0.060v it don't make a difference. Also I noticed no matter what I do w/ the curve once I start altering it instead of Auto LLC it draws wayyy more power and VID/Vcore high/lows don't even out at all. I'll still be under 1.3v but the Package Power watts go past the 253w limit pretty quick and it throttles itself.. No bueno

So yeah.. frustrating. These settings 0.150 / 1.10 Ohm LLC Auto, Lite Load 4 and the normal 253w limit, CEP disabled, 1.325v limit (even tho it never gets there) LLC auto 307A ICC MAX, Dynamic Tuning enabled, Undervolt Prot. Disabled seem to be my safest I can do, temps idle in 25C-30C no more than 69C under heavy load. If you look at the graph in my screenshot should I worry or try to spend any more time in working on these spikes I see in my voltage? I can't figure out the LLC to save my life and I'm assuming its because I can't get the right AC/DC on my own. Just hoping maybe you can put my mind at ease if I have nothing to worry about here.. thank you the screenshot is taken while doing a benchmark pretty heavy OCCT load so thats why the 1.27v right there. I'd rather it be about 1.23v max but I just don't know what todo to get it lower and thats the max, idles in 0.8v no higher than 233watts.. this is the best one I can come up w/ honestly. Hope this all makes sense. Is it even possible to get Vcore/VID in sync at all times? I've read that you want them close together but not sure if averages count. Under moderate heavy load mine stay close, something light or idle my Vcore will be a bit higher briefly then they level out back and forth. Maybe someone can explain what I should be looking for when monitoring VID/Vcore and what I can be happy with

 

[FlyingScot]

I’m kind of following you, but kind of not. You raise lots of potential questions, so we need to break your problem into smaller pieces to make the answers more digestible.

First, I’m confused in your description of VID and Vcore not lining up. Can you give some numerical examples? And explain what activity you see or would expect with those kind of numbers?

Here’s what I can say:
[1] Core VID will only align with Vcore under load. But even then, it may be out by up to 30mV just due to error in the Vcore reading. Vcore will tend to read high. Even in a perfect configuration, you could see a difference of 15mV. The final goal of tuning is to just get them as close as you can by using the DC_LL value. But this is pretty advanced stuff and you should not necessarily be focusing on this at this stage in the process.

[2] VID and Vcore can change very very rapidly, as does Core Frequency that initiates the change in voltage. What you see in HWInfo at any given point in time is a very tiny slice of what is going on inside the chip. When looking at the Current values it’s not unusual for them to be misaligned in terms of time. That’s why the other columns (Min, Max, Average) are sometimes more trustworthy. The longer you accumulate data that goes into the Average, the more the time differences between sensor readings are smoothed out.

Question: How have you configured HWInfo polling/sampling settings? I believe the default is 2000/0.

Ultimately, I would suggest that we take a step back so that we can properly define our goals. So, that’s another question. At the end of the day, what is your ultimate goal? What is most important to you? With every setting you change, there are usually tradeoffs. For example, what’s more important: raw performance or cool running? Video rendering and multi-core benchmarks or games, or none of these? Either way, I won’t let a specific voltage be your driving goal but rather be the byproduct of your other priorities, including what applications you intend to use in your day to day activities.

 

[logan_sudopwr15d302e5]

as always thanks for the quick response. I have my polling set to 1000s for some reason I thought it would be better through my lack of understanding lol. I'm "highly functioning autistic" (aspergers) .. so my communication sucks. Goals are multi-core loads. Not worried about gaming. Doesn't mean if I get an opportunity to waste an hour in Call of Duty once a week I won't do that but thats it as far as gaming. More worried about longevity, stability, and being undervolted.. getting the same or better performance w/ less power draw so I don't ever have to deal w/ RMA in the future (hopefully) I took another picture during Cinebench this time and I think this might be the right settings to be honest.

You really answered what I was trying to ask tho .. about the Vcore/VID that your really looking for it to be aligned during heavy load. During heavy load (CB23) we're looking at about 1.2 for both, idle is mainly .8 - little spikes to 1.1, 1.2 .. I've been taking information from here, (some) reddit threads, overclockers, etc. honestly probably a little to fixated on the whole thing but was getting hung up on Vcore/VID being around the same but I guess I misunderstood thinking they always needed to be about the same so that was mainly was I was asking about.

I also was worried about the what I think you guys call "transient spikes"? all the kinda crazy drastic up and downs while idle and I guess maybe I thought that wasn't good

This is the graph during Cinebench23 I should've just done this and looked at it from the beginning instead of doing OCCT.
Averages after 3 hours uptime :
Vcore - 1.056
VID - 1.064 and I think I read from multiple places that some of the MSI mobo's can be -0.013 on Vcore sensor readings? Alright well looks like maybe I did abit of panicking for nothing. Lol sorry about that my friend. Thx as always for taking the time to help me out.

 

[FlyingScot]

No problem! I still think you are doing great. Your questions at this stage of the game are not that unusual. Plus, I’m glad we were able to get “aligned” even if we can’t get VID and Vcore aligned! Maybe I can offer a little clarity now that I have a better understanding of your goals and concerns.

HWinfo64 polling at 1000 and 0 average samples (so you get a continuous averaging until you manually reset the counter/sensors or close the application) is fine for what you are doing with voltage tuning and stability testing. It will just hurt your benchmark scores a little more than say 2000.

Since you are concerned about voltage spikes (only visible on an oscilloscope) and voltage peaks (seen and unseen by HWinfo) we should expand our testing to some additional applications. Cinebench R23 is a great tool, but it does have a couple of limitations. First, it only produces all-core workloads. Second, it produces a steady workload with very limited transient fluctuations by comparison. What we need to do is include some tests with applications that are much more dynamic in their workloads, which will give you more insight into what VID requests your CPU will generate under normal PC usage, e.g. Internet Browser, Youtube videos, Office applications, games, Windows Update, zipping and unzipping files, antivirus scans, and all other Windows background processes.

For at least the first round of these tests, you will want to reset your voltage related settings. I doubt there is any point in testing with the default BIOS voltage related settings because they are crap. So, let’s make our default settings the kind of settings that CiTay likes because (a) they keep things nice and simple, and (b) they usually get you good and quick results. These settings would be:
CPU Loadline Calibration Control (“LLC”) = Auto [close in behavior to user mode 8], CPU Lite Load Control = Normal and CPU Lite Load Mode = lowest value you think is stable. You mentioned 4. Incidentally, if you are jumping between CPU Lite Load Control Normal and Advanced, you need to be aware of this BIOS navigation issue. It’s always best to confirm what AC_LL and DC_LL values are active by using these instructions, otherwise you could end up with invalid data and/or be driven crazy.

For these tests, I would suggest downloading and installing the following apps and give them a run.
All of these are good tests of transient workloads that will test your loadline settings:
Free on Steam: 3DMark Demo (includes Time Spy and Firestrike)
Free on Steam: Black Myth: Wukong Benchmark Tool
Optionally: Geekbench N (N = latest version).
And for a more serious stability test of transient behavior versus your voltage (and frequency) settings, the Extreme Mod version of Cinebench R15 (downloadable from here).

Note that the highest VID requests (and therefore Vcore) usually happen when you first open an application. Transient spikes (not visible) usually happen at the beginning and end of an application run. R15 is a perfect example of this. In other applications, like 3DMark, transient spikes can happen in the middle of the run. Even though we cannot see the transient spikes, we do have some data to fall back on thanks to YouTuber BuildZoid and his trusty oscilloscope investigations. I think a good general rule of thumb is to add 50mV or, for extreme safety, 100mV to any peak voltage you see in HWInfo. So, if you see 1.27V peak, a transient spike might be somewhere between 1.32V and 1.37V. As long as your calculation is below 1.45V, I think your CPU should last a long time. If you are above 1.45V in your estimation of transient spikes then you might want to look at additional mitigation strategies.

There are other applications and tests you will probably need to run to do a full stability test (once you have some favorite settings) but the above should give you lots of data to study. Incidentally, this is the kind of data you need to be collecting for a better understanding of what’s going on: post
[EDIT: Here’s another user’s use of a spreadsheet (my all time favorite tool) to keep track of his many tests of different loadline settings and power limits.]

After you have run these tests, while monitoring with Hwinfo at polling rate 1000/0, let’s have a conversation about another set of voltage settings you could run the same tests through so you can have a more fuller understanding of the strengths and weaknesses of these different voltage related settings. For the purpose of any future conversations about voltage settings, I think we can categorize them as follows:

Auto LLC, CPU Lite Load Control = Normal, CPU Lite Load Mode (1 - 23)
Above + negative Adaptive+Offset
——
Manual LLC, CPU Lite Load Control = Advanced, user entered AC_LL / DC_LL values
Above + negative Adaptive+Offset
——