Should I overclock or undervolt Intel i7 14700k?

[luca.fragolin15a102e4]

Hi everybody, I'm new on forum and hope I'm using the correct section.
I'm searching for an opinion from expert builders.
I built my very first PC with these components:
- MOBO: MSI Z790 Gaming Plus Wifi - Bios version 7E06vH9
- CPU: Intel Core i7 14700k
- RAM: Crucial Pro RAM DDR5 32GB Kit (2x16GB) 6000MHz CL36
- SSD: Lexar NM790 Internal SSD 2TB, M.2 2280 PCIe Gen4x4 NVMe SSD
- GPU: ZOTAC Gaming GeForce RTX 5060 Ti Twin Edge 16GB GDDR7
- Cooling: ARCTIC Liquid Freezer III Pro 240 (top exhaust) + intake/front fans from 2 140 mm fans in Fractal Design North case + 2 side intake fans at the height of the socket just above GPU / 1 rear exhaust fan, both ARCTIC P12 120 mm.
I installed the Arctic contact frame + ARCTIC MX-6 thermal paste.
On the first boot during OCCT and Cinebench 24 tests, the CPU immediately reached 100 degrees and HWiNFO reported widespread thermal throttling. So, while I was looking for some solution, I read Citay's excellent guide on this forum and planned to try the undervolt system he suggested. However, not trusting my building skills (with reasoned consideration, as I will say shortly), I decided first to disassemble and reassemble the build and all the tests I carried out afterwards showed much lower temperatures (average 85 degrees).
I noticed there were Core VID's maximum over 1.450 V, so I decided to setting only the Voltage Mode to: 'Adaptive+Offset' with Offset Mode '-' and a value of '0.050'.
I am attaching the tests I have conducted since then (language is italian, so let me know if something is not clear):

HWiNFO screenshot during Cinebench23 - 33473 points

during Cinebench24 - 1901 points

during OCCT CPU Extreme Mode

during OCCT 3D Adaptive

a configuration of CPU fans using FanControl if can be useful

I noticed that the temperatures are stable and not reaching 90°, that the CPU frequency reaches what seems to me a decent goal (in BIOS I enabled XMP Mode 1 + Turbo Boost "Enabled" and Enhanced Turbo set on 'Auto'); also I have passed some 30 minutes OCCT stability tests with no errors. However, I also noticed that during benchmarks and stress tests there are high percentages of CPU power limit exceedance and that the GPU is also affected by 'reliability voltage' limitations.
So I come to the questions:
1) Do these limits somehow affect the safety/durability of the build?
2) Am I experiencing performance drops that I could/should fix?
3) Would it still be advisable to perform an undervolt at least on the CPU Lite Load as suggested by Citay?
4) On the contrary, with these values, would it be advisable to try to increase performance with some form of overclock (both on CPU or GPU in case)?
If so, any suggestion would be valuable since I am totally clueless about it xD
Also, tell me if, from reading my data, if you notice anything unusual.
I should specify that I mainly use the PC for word processing (work+cloud+surfing) and gaming (including AAA titles), no rendering, streaming, video editing, etc.
I thank anyone who will answer me and I apologize for the length, let me know if more specific infos are needed!

 

[citay]

As the author of said guide, it's only right i give my two cents here

First, for any further screenshots, please set HWinfo to English. The terms are only ever used in English, even on forums in other languages they will say "power limits" for example, not what the translation would be. Because also in the BIOS, people usually leave it in English and then those are all the same terms there.

I have to say, your ARCTIC Liquid Freezer III Pro 240 performs above what i would expect from a 240mm AIO, that's Arctic for you, good stuff. Even still, i would go into full-blown undervolting mode with this CPU, the 14700K is an absolute monster that needs to be tamed. It behaves worse than a 13900K, they really pushed the 14700K hard from factory. Your current undervolt is too mild, your CPU can probably still be stable with less voltage.

My preferred method is always the "CPU Lite Load" method from my guide, because it's pretty easy to apply, pretty easy to stability-test (because it has the most effect at the highest loads), and gets the job done nicely.

An overshoot of the power limits, as can be seen on your screenshots, is pretty normal, those are usually just shorter spikes, the long-term power draw should be a bit closer to the limits, and 253W is indeed the most i would allow for any CPU. Simply because above this, you get too much diminishing returns, very few performance increases, but there can be still a lot higher power draw, so it turns into "junk performance", it's highly inefficient.

Sadly, on the first three screenshots, you expanded the sensors in such a way that the board's sensors like temperatures and fan speeds are sometimes cut off on the right. But it looks like you have good ventilation through the case. BTW, hide T0 and T1 temp sensors, they always show nonsense.

The power limits, you can keep at 253W, your cooling can cope with it. I would now dive into CPU Lite Load undervolting as per my guide. You can reset your VCore undervolt back to Auto, CPU Lite Load will take care of it on its own, plus your undervolt was not doing that much anyway.

You will see that, purely from finding a good Mode for CPU Lite Load (the lowest stable mode, then one step higher for stability headroom), the performance should increase a bit, because now it has more "breathing room" for clocking higher within the same 253W power limits.

I would not overclock a 14700K whatsoever, the days of overclocking are behind us with such CPUs, you can hardly get a few hundred MHz more out of them because they come so aggressively set up from factory. And then you would need more voltage, so you'd sacrifice any undervolting headroom for a minuscule performance improvement that is almost not worth mentioning, if you can even utilize it with your current cooling. Because more voltage means more power draw and a hotter CPU, until it hits the power limits, and then it has to clock down, so nothing is gained. So, overclocking such a beast is rather useless.

The best thing you can do is to fully exploit any undervolting potential the CPU has, while staying fully stable. This is the safest and most efficient way to run such a CPU.

 

[luca.fragolin15a102e4]

Hi and thank you for the answer!
I'm posting right now HWinfo reports adjusted in the way you suggested (or I hope so, let me know if they are expanded bad again) and during week-end will go on changes in Bios+tests as for your guide: I hope I can bother you again to look at them after I’ve posted them.

This is the system with only background processes+Edge open

Here is during Cinebench23 (concluded with 33481 pts - had a max of 84° CPU Package / 1.458 Core VIDs / 267W CPU Package Power)

I guess what those sensors indicate?

 

[citay]

Ok. All looks pretty much as expected, with minimal undervolting and power limits in place. Fans are also ramping up nicely, and again, for a 240mm AIO, this level of cooling is impressive.

There's still room for improvement in the fan curves though. Looking at your initial post and what i see now in the sensors, you can set them to lower speeds in idle, my Noctua fans are at 400-500 RPM in idle. Right now, you're cooling the CPU unnecessarily much in idle, it doesn't really care if it's at under 30°C or closer to 40°C there, but you are getting higher noise and more dust accumulation into the system if you run fans at 900 RPM in idle.

Your fan curve, despite using a dozen points across the curve, is more or less a straight line. So all those small adjustments are not amounting to much in the end. Try making it more of an "ascending dominant" curve as i explain in my fan curves guide. And the lowest point of the curve should be at 35°C for example, and then the curve should rise much slower and more gradually, to stay in the lower RPM ranges for longer. The fans only really need to kick in when the CPU starts to put out tons of heat. In low- to mid-load scenarios you can prioritize noise. But approaching 90°C, the cooling performance obviously has top priority.

 

[luca.fragolin15a102e4]

Thanks again for your reply, waiting to have time to start testing on undervolt I'm following the "ascending dominat" curve suggested in your guide (1st pic attached) but as I'm writing now I have Package Temperatures between 28-30° (full idle apart from HWinfo and Edge), could be useful to set the lowest point on 10% at 25° for example? Consider theese approximate max RPM values (2nd pic): CPU Fan 2500, Pump Fan 2800, 2x120mm Exhaust 3000. I have them now at 15% (CPU package 30°) but Fan Control show me they are running respectively at (approximatively) 700, 1100 and 800 RPM (3rd pic). Maybe I'm not good in math but that doesn't seems to be much more than their 15% potential so I wonder if I'm messing with it or not understanding some notions about!

 

[citay]

Each fan model responds differently to the same RPM value it gets from the board, because each model has a different RPM range to begin with, as well as a different mapping of the PWM value to that range. Some fans have the same lowest RPM from 0-20% RPM range, because some motherboards have 20% as the lowest PWM value you can choose. Some may have the fan turn off at 0%, but then lowest RPM from 1%. Things like that. You need to find the lowest RPM of these fans.

And then 25°C makes no sense as the starting point, because even in idle on the desktop it's not that low. Of course, in that program it doesn't matter that much since you have lots of adjustment points, but in the BIOS this matters more. So there i always start at for example 35°C, that's the point of the lowest RPM. And then try to keep it quite low for a long enough time, like on your latest screenshot, that's pretty good.

Like i said, in idle, with under 10W CPU Package Power, the CPU could care less if it's at 25°C or at 40°C. The cooling only becomes important once the CPU has way higher power draw. In idle it has zero benefits to cool it better, only downsides, in noise and dust accumulation.

 

[luca.fragolin15a102e4]

Found that Arctic LFIII 240 has those minimum RPM at 1%: 400 x CPUFan, 750 x Pump, 600 x Exhaust Radiators so I'll set the Bios with an average 30°-15% / 50°-30% / 70°-70% / 90°-100% and will try to desing a curve with RPM values (instead of %) for each AIO's component using FanControl so to stay near minimum in idle.
Here is an example:

 

[citay]

That about checks out, the VRM fan is specified for 400-2500 RPM, the water pump for 800-2800 RPM, the radiator fans for 600-3000 RPM. So they all go decently low. I see you're going to full speed at 80°C, is the noise still tolerable? Otherwise you can go as high as 90°C.

 

[luca.fragolin15a102e4]

As the max temps reached during stress tests even before doing any undervolt are at 80-85 I tought to use max fan speed at that point (as I understand, should be a temp not even approached during normal PC use or gaming), maybe I'll up if I see the noise is annoying. Meanwhile, thanks to your suggestions, it feels like the PC isn't even on because of the fans quietness

 

[luca.fragolin15a102e4]

Meanwhile I managed to have some tries in undervolting:

1) VCore undervolt on Auto + LiteLoad Mode 4 --> crash on Bios

2) LiteLoad Mode 6 --> had a CB23 score several thousands points less than previous

3) LiteLoad Mode 6 + IACEP Support / IACEP Support for 14th disabled + Undervolt Protection Enabled + PW1/PW2 253/253 gave me those results while doing CB23 (score 33454 pts, same as before I modified LL Mode):

Here are WHinfo during OCCT stress test (concluded with no crash, no errors):

4) Follows CB23 with LiteLoad Mode 7 (33053 pts):

5) And LiteLoad Mode 8 (33240 pts):

Noticed there are no Core power limits (I guess is because power doesn't reach 253W, right?), but Performance Limit Reasons are always at 100% (am I losing some performance from CPU?). As I can see, there is a very bit increas of average core clocks and max voltage from Mode 6 to Mode 8. I don't understand why with higher Modes benchmarks scores are worse.
Do you think this results are good (temperatures, power consumptions, voltage, performance, etc.?), if so what config should I choose? Or should I try to push onto higher LL modes or doing any other change in bios? Consider that I let Intel Turbo Max 3.0 Active and Enhanced Turbo on Auto, didn't touch either PL1/PL2 and UV protection is still enabled.

 

[citay]

When undervolting, why use UV protection? Scores are roughly the same as they were without serious undervolting, but at much lower power draw and temperatures, so for sure this is a success, it's become much more efficient.

As for the Performance Limit Reasons, there will always be one of them you hit under load. Usually, if the CPU can avoid the power limits altogether (due to undervolting), then it will be "Max Turbo Limit" for example, it's simply hitting the maximum turbo multiplier.

I don't understand why with higher Modes benchmarks scores are worse.

If you calculate the score differences into percentages, there's almost no difference, it's fully within the margin of error. But of course, higher modes tend to have a bit worse scores. Remember, lower modes mean less voltage per frequency, which means less power draw per frequency, so if it's quite close to the power or current limits, it may be able to clock slightly higher in some situations.

If Mode 4 was unstable even in the BIOS, then 5 is also out of the question, even 6 might be too low. Something like 7 or 8 might be viable, just make sure to test the stability thoroughly.

 

[luca.fragolin15a102e4]

Remember, lower modes mean less voltage per frequency, which means less power draw per frequency, so if it's quite close to the power or current limits, it may be able to clock slightly higher in some situations.

That's at first something strange for me to understand, but it's definetively true!
Seems that disabling UV protection made the system stable, I was able to down LL to Mode 1 without any crash. Best CB23 score was on Mode 3 (33630). I'm going to test LL3 with OCCT to see if it's stable and in contrary I'll upgrade back to Mode 6 but yeah, seeing those temps not even turning over 70° make me feel very satisfied!
I have to thank you again for your patient guidance