[SilenMar]

13900K is more heat-resistant. It can run 1.44V 6.0GHz all-core gaming when it is below 65C. It can also handle 1.42V 5.7GHz all-core around 100C at 350W. For the last two months I have been doing 350W rendering around 100C with this CPU and it handles the heat just fine.

350W Photoshop high resolution image rendering

350W Handbrake E-core video rendering + 3 x 8K 60fps P-core video playback

350W P-core +E-core Handbrake thermal throttling at the set 105C limitation. The same goes to Cinebench23.

The problem is the current motherboard doesn't have sophisticated voltage control to cover the full range on both the heavy load and the lite load. The CPU is either undervolted or overvolted.

If the CPU is undervolted with lower LLC then it doesn't have enough voltage for high frequency 6.0GHz during the lite load.
If the CPU is overvolted with higher LLC then it doesn't have lower voltage for low frequency during the heavy load. It will have thermal throttling at the set temperature limit.

The default setting of 13900K out of most motherboard can already run 40K in Cinebench23. Not everyone runs rendering 24/7 for that undervolted 40K-47K heavy load bench performance. It's more beneficial to run 1.44V 6.0GHz all core lite load for gaming.

 

[SilenMar]

The current power limit or limit on power consumption doesn't make a difference in terms of how stable the CPU is around 220-230W, at 85C, all-core, 5.9GHz. A 105C limit with a bit of overshoot at 110C is perfectly fine.

LLC Saturation only helps to boost voltage a bit higher to 6.0GHz for undervolted LLC, such as LLC8, at 1.25V all-core, 5.4GHz. However, it requires both a very good CPU and good VRM, otherwise, it won't work. My CPU crashes straight up with LLC8, LLC7, LLC6 with 1.44V overdrive for 6GHz. The voltage curve for other frequencies is too low. When the CPU runs with borderline LLC5 + Saturation above 40A, the 350W 105C thermal throttled voltage is 1.40V, which is even higher than Adaptive voltage 1.33V.

It's still the combination of LLC, TVB, and Adaptive + Offset voltage that makes a difference. The CPU cannot be stable when the voltage is lower than 1.42V at 5.9GHz, 85C. But there are only two points of effective offset voltage: the point of 6.0GHz, 1.44V, and the point below 5.7GHz, 1.35V. Other frequencies, such as 5.9GHz, 5.8GHz, will be affected by LLC. 5.4GHz is completely ineffective, no matter what the offset value is. LLC5 gets 1.412V at 5.9GHz, which is not stable at 85C, while LLC4 gets 1.426V at 5.9GHz. This also results in 1.41V at 5.7GHz, which can be offset to a lower voltage, but I would rather use a constant high voltage instead of a constant undershoot and overshoot due to higher LLC.

There are only three points of 8-Pcore TVB as well. Anything higher than 85C, I can only set 5.7GHz TVB, or I could have set more points, such as 5.8GHz at 85C-100C, then 5.7GHz 100C-105C.

 

[SilenMar]

The PC is just a rendering machine with an old Corsair H115i 280mm AIO with 2 NF-A14 and 2 NF-A12x25.
If the CPU is delided with a custom loop the frequency under heavy load can go higher.

 

[SilenMar]

Hmm... would recommend investing in a custom loop. I'm not saying you're wrong here but I'm not convinced you've tested this long enough to make any definitive conclusion. At least I'm not ready to risk my $600 investment

I also think something is buggered with your volt tuning. I used this Guide and tuned my LLC ( AC_LL & DC_LL) and VF curve in accordance and my my volts are at 1.5 but my R23 watts are 290W, occasionally peaks to 300W. I'm rock steady form 6.2 light load to 5.7 heavy load. IIRC, settings were 60, 60, 60, 60, 59, 58, 58, 57 but I would generally get a 200Mhz boost at light and medium load even though TVB was disabled. Custom loop with direct die block.

View attachment 2604903

But I'm now just running stock with LLC curve optimized IAW the above guide, Asus MCE disabled and with a 100W and 10*C drop, I'm still boosting very close to the same clocks as the OC with R23 only dropping 1k points.

With this voltage you cannot run 6Ghz all-core in games. You got 5.7GHz all-core with undervolted voltage instead.

 

[SilenMar]

I couldn't. Maybe adjusting the LLC tune but it failed at 6.0 all core R23 (posted though). However, I agree, I have a lot more headroom in the CPU and cooler. Unfortunately not in the PSU. My STRIX 3090 has the XOC bios and when overclocked, pulls 600W continuous and transient spikes up to 1000W so with the CPU overclock and GPU overclock (2130 core) I get random PSU trips so I decided to stop there.

But when I found out that I get basically the same performance with a lot less power and heat, I saved my OC profile then returned to stock settings and still get game clocks in the 5.8ish range. At 4k on a 3090, I don't think I need a 6Ghz CPU.

All of these motherboards don't have voltage control to cover both overvolt and undervolt simultaneously.

You are not rendering 24/7 like I am and are more likely playing games. In fact, I don't even need to do anything for heavy load rendering. The default BIOS setting is already at 40K. You undervolted it to get a 2.5% increase 41K with a lower temperature. However, the CPU is designed to handle 100C just fine.

So the headroom is meant for high voltage and high frequency at light loads.

 

[SilenMar]

Says who? And what difference does it make? I'm just saying I think you have a lot more voltage tweaking to go. I ran my overclock at 5.7 all core for over 24 hours of R23 and barely broke 300W peak power and still was stable at 6.0-6.2GHz under light and medium load. I'm pretty sure you can bring that 350W power down a bit.


This is where we're going to have to agree to disagree. I honestly don't think there is enough testing to validate that claim.

I already said your voltage is undervolted that won't be stable for 6GHz all-core light-load gaming. This is the difference. You have 5.7GHz in games instead. If you want 6Ghz all-core gaming then the LLC needs to be raised to the stage of overvolt instead.

I have been running like this for the last two month including several days of continuous rendering. There is no problem with it.

You want the even lighter heavy load like smallFFT Prime95 330W instead of Handbrake 350W then I can show it to you. It just hit the thermal wall to have less voltage.

 

[SilenMar]

100C on stock clocks and volts may be "fine", not overclocked and overvolted.

The CPU doesn't request specific "low" default voltages. It's always the motherboard doing the adjustment.

He must have watched that YT video that der8auer did with some Intel engineer in which he said that "100C is perfectly fine as long as you let your CPU cool off for a while" or something along those lines. Total marketing bullcrap even a 80IQ simpleton shouldn't fall for.

What video? I've been doing 100C for long since 9900k. Maybe somebody IQ lower than 80 don't understand the heat-resistance of this gen 13900K.

 

[SilenMar]

SAYS WHO?


Let me try one more time.

I
AM
STABLE
FOR
6.0
GAMING


Overclocked, my gaming clocks were between 5.8 low, 5.9 - 6.0 average, and 6.1 peaks (and 6.2 light load/medium load).

I am NOT stable at 6.0 R23 all-core but I don't think you are either. Whats your point?

Says me of course. Games use all P-cores. You have a setting of 60, 60, 60, 60, 59, 58, 58, 57 then you have 5.7GHz in games. You should have a setting of 60x8. This is the point.

 

[SilenMar]

Wrong. I hit 5.7 in all core heavy load such as R23. Look back to what I said. My CPU boosts 200MHz over the overclock settings even though TVB is disabled. My game clocks are NOT 5.7, they are 6.0 with small drops to 5.9 and small peaks to 6.1. Look at my attachment. My clocks are sitting at 6.2 all day under light load.

You're saying I can't setup a V/F curve that allows the CPU to run at 5.7 heavy and 6.0 light load, yes? I disagree.

Your attachment picture is meaningless in desktop with no game to use all p-cores. In reality it is 5.7Ghz.
Better show the screenshot of an actual game.

 

[SilenMar]

Show me a game that loads 16 threads to 100% and I'll show you my gaming screen shots.

As long as a game uses 8-Pcore you will have 5.7GHz. Don't fool yourself.

 

[SilenMar]

Good. Show me a game that uses 8 cores.

Easy. Any game can be optimized to use 8 P-core and even 8 E-core.

I don't run 5.7GHz instead.

 

[SilenMar]

What that Intel engineer said is nothing new under the sun
View attachment 2604957

"All there in the manual": if Tjmax for a 13900KS is 100C then 99C is long term reliable, which in turns makes 100C long term reliable and irrelevant too from a scientific standpoint since that's the temperature at which the CPU will regulate. That's all like he said "as long a you're under the stock voltage and PL1" which is said here too: voltage and frequency modifications are straight up considered overclocks (their long term reliability is out of the bargain) and here's mentioning processor base TDP which is PL1 for all intents and purposes.

Like it or not an stock Intel CPU can run at 100C during heavy load no problem and still last as intended. Another case is an overclocked CPU: that's actually when nothing is a given

So stop spreading statements out of context: if unsure double check the sources.

As for games thread usage that's too much work to know for certain: but nowadays little things are single threaded or even use a few cores.

Almost any game using DX11 uses at least 8 cores: depending on how demanding they're(and year of release) and target frame rate they may get away with 4 cores or 6 cores. Albeit strangling core count is may hurt stable frame rates: but even this can be countered with configuration, allow the game to prepare more frames on the CPU ahead of the GPU and there's a trade between more stable frame rate, but higher latency. But fixes it.

DX 12 use even more cores with more evenly spreaded load than DX 11: that's the main reason why DX 12 games(most modern ones) are actually able to render well past 200 FPS assuming the GPU doesn't botteneck. Again configuring frames rendered ahead by the CPU serves as a software workaround to fix unstable frame rate at the cost of increased latency.

But in general this is all too a complex scenario with many knobs to touch in order to get it working the best possible way in any hardware/software combo: so it's really way more complex than some arbitrary number of cores needed to run games.

Intel only said "may" result in damage when running out of specification. Since this is overvolted OC at 350W. It's already far from specification. What is the concern? Warranty? Degradation?

The CPU runs just fine. All it can be debated is something like "time will tell" several years later. But a year later a 14600K can run just as fast as 13900K. The OC is over then.

 

[SilenMar]

So the purpose of OC is the hope it can last longer without upgrading while new chips can run much faster regardless.

CPU is one of the most resilient components especially this gen. Just sell the old to get a new one.

 

[SilenMar]

I am not sure what the point is here.

You have an over volted and over LLC'd chip that runs at 100C and renders. Ok. StAndrew has a chip that clocks just as well and pulls less power.

I am struggling to see the point here.

CPU has been and probably will be the most resilient component for quite a long time, but not due to high temps, high volts and no vdroop.

The purpose of OCing used to be to get a processor at half the cost, have good cooling and OC the snot out of till it surpasses the most expensive thing that could be bought.

Have to admit, I am still missing what the bread and butter I am supposed take away from the OP is.

These motherboards do not have control over both undervolting and overvolting. It is only possible to choose one. That's why StAndrew, with an undervolted curve, can only achieve 5.7GHz in games with a slight load. If he wants to reach an overclock ratio of 8-core 6GHz, he needs a higher LLC to overvolt.

My point is that it's more beneficial for performance to run an overvolted 1.44V 6.0GHz all-core ratio for light loads. Heavy loads are rare, and even at an overvolted 350W and 100C, the CPU is still fine. Therefore, there is no point in having an undervolted setting for heavy loads that you don't use often, but instead limit the CPU to 5.7GHz for light loads where you use the CPU the most.

Regarding cost, I don't even need to delid this generation. I just put in a 280mm AIO, and that's it.

 

[SilenMar]

How do you know it is fine? It hasn't been out that long. What is your testing period length that you feel confident to state this? Overvoltage and heat are the exact things that degrade a cpu over time. Sure you could buy another cpu in a year. Why should anyone listen to your advice if your unconcerned about the longevity of your components?

You stated yourself that you have a 5.7 all core load. So does StAndew. I feel english is not your first language so maybe I will just chalk this up to that.

I have 6.0GHz all-core at light loads in games, while I achieve 5.7GHz at 350W during heavy loads. StAndrew doesn't have 6.0GHz in games because his setting is 5.7GHz all-core, while mine is always set to 6.0GHz all-core. This is why he's too shy to show his game screenshots.

I've been running overvolted like this for two months, doing 350W heavy rendering each session for about 6 hours. When I run Prime95 smallFFT, it doesn't even reach 350W; it only hits 330W.

As I've mentioned before, the only thing up for debate is something like "time will tell," and that will be several years later.

 

[SilenMar]

That's not what StAndew said. He hits between 5.9 and 6.1 in games with TVB and a 5.7 all core load. Yes time will tell. So why should anyone listen to this advice? How long have you been OC'ing? Jumpers ring a bell?

He only said so to just show a desktop screenshot that runs nothing. Then where is the screenshot in game? He has 5.7GHz in games.

As I always said I've been running overvolted like this for two months, doing 350W heavy rendering each session for about 6 hours. The CPU is just fine.

I know these motherboards don't have full range of voltage on the both end in the very beginning.

 

[SilenMar]

I'm not sure why the exposure, maybe its the Auto HDR feature but here is Warhammer II, my heaviest CPU intensive game.

View attachment 2605020

So much for your all-core 6GHz. Only a few 6Ghz bouncing for a few seconds.

 

[SilenMar]

You do you


Two months? That is nothing.

Just pump volts and heat into it, I am sure it will work out fine.

I probably should have stopped responding a post ago.

What happened to this forum?

I like how you imagine motherboard works without actually testing it yourself instead of letting all the imagination fly off the cliff.

 

[SilenMar]

I can't take you seriously, and you can't even address counter points that were brought up to you.

What counter points you have given? Running the CPU 24/7 for a year instead? I've been running this CPU for two months.

The only thing up for debate is something like "time will tell". Time is on my side I'm afraid.

 

[SilenMar]

your power is so much higher cuz the thermals are bad resulting in massive leakage, i have seen shocking reductions in power draw from going down 30C in load temps

People like to jump on when I just use an old Corsair H115i 280mm. My setup can outperform most 360mm out there. You put a 1.42V and see how the 350W temps go on a 360mm on prime95. It's still the similar temperature. The point is the CPU runs just fine. You put a 330W power limit in motherboard then it runs 99C. Suddenly CPU doesn't boil like water anymore.