Asus Maximus Z790 and Intel i9-13900k/14900k - An overclocking and tuning guide.

bash13r said:

Ok, let me first say thank you. & this is definitely more advanced than a beginners guide. I still don't fully understand everything, but I'm going to follow the practical part & hopefully get a slightly oc'd 13900k with better than stock temps.

So if I start at "So let's start tuning the beast!" and get to what I think is the last of the beginner's section, "Overclocking the Ring", can I stop & leave everything stable before getting into the "advanced P-core settings" till the end?

I know it may be advantageous to go further with the oc, but my initial goal was to get stock freq's with better temps. Now I have an optimal stock setup, with +2boost so a little better performance & better temps. What more could I ask for?

BTW, I haven't done any oc'n yet, just checking first if I can stop before the advanced P-Core settings? I'm really a beginner, this would be my first oc ever, I'm a little scared 😊

Click to expand...

RobertoSampaio said:

adaptive 1460mv
Rise the vf curves (8,9 and 10) 90mv
58x2-57x4-56x6-55x8 and apply +2 boost profile.

And tune AC_LL again...

Click to expand...

digitalfrost said:

I am at the moment at a point, where I can do

adaptive 1520
Rise the vf curves (8,9 and 10) 10mv

or

adaptive 1475
Rise the vf curves (8,9 and 10) 40mv

both are stable. AC_LL is already at 1, so I cannot lower it anymore. Other combinations in between also work (i.e. adaptive 1485 or 1500 with correspondingly modified VF offsets). Which would you choose and why? I figure if I don't need the high max VIDs (from adaptive) probably better not to use them on the other hand I'm slightly raising the voltage in regions that are far more used in real life. In practice due to LLC I see no difference in effective Vcore during everyday loads.

e: If I add up all that worked stable for me so far:

1475+40 = 1515
1485+30 = 1515
1500+20 = 1520
1520+10 = 1530

So I think I'll go with 1485 adaptive because it has the lowest offset while at the same time providing the lowest total max VID.

Click to expand...

themad said:

Interesting. Do you get that level of information with the Basic (free) version already or do you need the Advanced (paid) one?

Click to expand...

sunnydaler said:

Thanks, @RobertoSampaio I am new to this kind of thing and your guide helped me a lot when I tried to undervolt my potato chip.

stock: pulls 307W at VCore 1.341 and 5500MHz(p) on CB R23 multi test. V/f point of 5400MHz is fused at 1.319v on the V/f point offset table.

-1-2 cores at 58, 3-8 cores at 54
-put max ratio limits (x58 on 2 best cores and x54 on the rest).
-disabled MCE
-LLC 3
-DC_LL 1.12

Then I tested to find a stable value of AC_LL
AC_LL 0.25: Vcore 1.252 - my current setting
AC_LL 0.23: Vcore 1.243 - passed CB R23 30min x2 run but had some audio glitches sometimes on normal usage. not sure if undervolting is the cause.
AC_LL 0.19: Vcore 1.234 - failed

Should I raise it more? I guess I need to test more with other bench tools.

Click to expand...

RobertoSampaio said:

A tuning guide for beginners.​

Disclaimer !
These settings and methods
are outside Intel's specifications,
just like any other overclocking method.
----------------------------------------------------
Asus Maximus Z790 extreme & Intel i9-13900k.​

Introduction:

Once again Asus and intel surprise us with an extremely robust, powerful and efficient platform.

In the past, when we overclocked the CPU, the performance gain reflected significantly in productivity gains.

These days, processing power is at such a high level that overclocking the CPU doesn't contribute significantly, but it always yields some FPSs.
The fact is that a perfect tuning of the system guarantees good temperatures and the possibility of high impulse clocks.

When a customer buys a computer from a specialized company, it is assumed that all adjustments have already been made at the factory. On the other hand, when we decide to build our own system, it will be up to us to make all the adjustments.

Imagine that you decided to build a turbo engine for your car. It is not enough to buy a turbine, a new ECU, make all the connections and start accelerating. It is necessary to think about the air/fuel ratio, the turbine pressure, the relief system, the cooling system, and do many, many tests to arrive at a satisfactory and safe adjustment. So it is with computers.

When we buy a motherboard, memories, processor, cooler, VGA card, etc... when we connect all the components together, the first thing that happens is a message from the system BIOS warning that nothing is configured.

It is from this point that I intend to help first-timers. For experts in breaking overclocking records this guide may seem basic, but I assure you that it is always possible to learn something new.

The intention of this guide is to provide basic information to start tuning the Asus Maximus Z790 MB and Intel i9-13900k CPU using all the CPU and MB boost technology and power management features.

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''

* * * This procedure can be applied to the Z490 / Z590 / Z690 / Z790 Maximus and Strix MBs with some adjustments. * * *
View attachment 2577928

''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''""""""""""""""""""''''''''''''''''''''''''''''''''''''''''''''

ROG MAXIMUS Z790 EXTREME | ROG Maximus | Gaming motherboards｜ROG - Republic of Gamers｜ROG Global

The ROG Maximus Z790 Extreme is the ultimate ally for Intel Raptor Lake, driving savage overclocks with its 24-stage power solution and superspeed DDR5 support. It features five M.2 slots, one of which is PCIe 5.0-ready, Thunderbolt 4, USB 3.2 Gen 2x2, 10 Gb Ethernet, Wi-Fi 6E, AI Overclocking...

rog.asus.com

''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''""""""""""""""""""''''''''''''''''''''''''''''''''''''''''''''

''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''""""""""""""""""""''''''''''''''''''''''''''''''''''''''''''''
​

Thanks:

My thanks to Shamino, Falkentyne, Cstkl1, Nizzen, Sugi0lover and the entire Asus team.

''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''""""""""""""""""""''''''''''''''''''''''''''''''''''''''''''''​

Considerations:

People who know me know that I'm not a brutal-force overclocker, and that I don't mind a few FPSs. And to be honest, I also don't like testing software that demands something from the system that is not compatible with the real world.
My goal is to achieve the best possible result, without major changes, aiming high single-core boost clocks while maintaining stability, with a decent full load frequency and good temperatures, something within everyone's reach.

And for that I prefer to adjust the system using tools that approximate the reality of the common user, so that in these conditions the best possible performance is obtained.

The tools I usually use are:
Cinebench_R23 for full load adjustment (3 to 5 minutes is more than enough if you don't have a custom water cooler with more than 1L of water inventory).
Realbench_2.56, GeekBench_5 and 3DMARK Fire Strike (DX11) for load transients.
AIDA_64 and OCCT for memory tests.
To monitor the whole system I use HW-info.

It is worth mentioning that the intention of this work is not to provide an overclocking recipe, but to help the end user to understand and adjust the marriage between the MB and the CPU.

Asus has great features in their MBs that allow for optimization and overclocking, and they also provide a unique tool called "AI Overclocking" that helps you get the most out of your MB/CPU set.

What we'll do here is put the final touch on these adjustments in order to optimize the entire set.

To this end, I will objectively approach the concepts of voltages, frequencies and temperatures, and how all this influences the TVB configuration. In addition this guide will cover the basics of Load Lines, Usage by Core, VF Curves, TVB and OCTVB.

This guide is divided in 4 parts.
This is the first part, with some theory.
The second is about the startup process and more practical.
The third is the advanced module.
And the last is about testing.
And all this content will be constantly updated over time.

If you think that's a lot to read and learn, think about how hard it was to write all of this, and know that the best you can get from this reading is not a higher XYZ_bench score, but the knowledge that no one can take away from you!

So, enjoy reading!

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''​

Before starting:

If you've already bought an Asus Maximus Z790 card and an intel 13900K processor, I assume you already know what you want:
The best available on the market!
So it got off to a good start!

Now comes the next question:
What kind of user are you?

There are 3 types of users:
- Enthusiast: Is the one who doesn't stop fiddling with the settings and his goal is to learn more and more, always trying to get the most out of the system.
In this case, this guide will help you start the process, and your experience will probably never end. Once you've mastered one subject, you'll soon move on to another, and by the time you get the most out of one platform, you're probably already considering buying a newer MB and a newer generation CPU.
Your game is overclocking, and gaming is a way to test the system.

- Commercial user: It is the one who uses the equipment in some business to earn money. In this case, it is best to find a very stable and energy efficient configuration. The best thing would be to keep the default clocks (or even reduce them a little) and make an undervoltage without compromising system stability.

- The Gamer: Aim for fun. The gamer usually looks for a configuration with the best possible performance, but does not want to keep tweaking the system indefinitely.

Once you've defined what kind of user you are, let's start by breaking some paradigms and myths. If you are really a first-timer maybe what I'm about to tell you is easier, however if you already have some experience setting up and tuning computers it may be that some paradigms have to be broken.

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''​

Breaking paradigms:

Let's now clarify some points that many people have doubts and that are somehow controversial.

• Does the I9-13900K consume more than 300W at full load?
• The answer is NO!

If you are running cinebench_r23, and you are getting readings above 250W, with the system in stock, something is probably wrong.
Always compare the CPU power reading to the VRM power reading. They must be compatible. For that use the HW-Info.
Although manufacturers provide motherboards and processors operating at higher voltages than necessary (to ensure system stability under adverse conditions).
If the power reading at full load (r23) is above this limit, it is very likely that an adjustment in the load lines is needed.
The 13900K operating at stock frequencies, as long as the load lines are minimally adjusted, will present a consumption of less than 250W on full load (r23).
Later we will address the load lines and everything will become clearer.

• Do I need a custom cooling system for the 13900K?
• The answer is no!

A custom cooling system is only needed if you intend to push the system to the limit (or beyond). A 360mm water cooler and good thermal paste will probably do the cooling, keeping temperatures in the mid-90s for extremely heavy tasks. I venture to say that if you live in a cold place, even an air-cooler might be enough.
There are several air-coolers that handle this level of power, such as the AK620 (maximum heat dissipation power of 260W).
What we have to keep in mind is that no everyday application will put a load on the CPU like the R23 or P95. Obviously there are exceptions, as in the case of Flakentyne which uses stockfish, which is an “Open Source Chess Engine”, and which is extremely CPU-heavy. In these cases, it is recommended to adjust the MB/CPU set for your daily reality.

• Do I need to do long tests at full load to ensure system stability?
• It is not necessary.

The test duration should be long enough for the temperature of your cooling system to come into equilibrium. Once equilibrium has been reached (water temperature has stopped rising) the test can be terminated. I particularly don't like this type of test that puts all cores on 100% load for long periods. Many times your system can withstand more than 30min in this condition and fail in a simple processing load transient.

• Will voltages above 1.5V deteriorate my CPU?
• Once again the answer is NO! But be careful !

Just like a car engine, where wear is due to excess power, temperature, and torque, CPU wear comes from excess power, temperature, and electrical current.
Intel's data sheets put the maximum allowable voltage at 1.72V. So once the power and temperature are controlled, there's nothing wrong if Vcore reaches 1.65V at light loads. But be care. One thing is a voltage spike or a high voltage at a very, very light load, the other is trying to run a heavy load at a high voltage. If you try to run a heavy load with a voltage that exceeds the CPU power capability, you will have problem. For instance, 1,2V (a relative small voltage) with 250A will be enough for throttling or even a degradation.

• If I do the delid, use liquid metal or change the IHS will I degrade the CPU?
• The answer is YES and NO. I'll explain...

If the changes are only made to lower the temperature, there is no harm. The problem is when we take advantage of this modification to increase the CPU voltage to reach higher frequencies at full load. In this case, the CPU power will increase, and its temperature will still be low. As CPU protection is done thermally, there is a chance of degrading the CPU due to excess electrical current and/or power.
It is worth remembering that applications that put all cores operating at 100% load constantly are rare. Rest assured that 30 minutes of r23 or P95 will wear out the CPU infinitely more than Vlatch reaching 1.60V at “idle”.
The fact is that it is uncomfortable to see that the voltage (VLatch_max) has reached 1.60V. But people don't feel any discomfort when the CPU reaches 300W of power. Very strange, don't you think? What you need to know is Ohm's law won't forgive you if you do something wrong.

• What can degrade my CPU?
• The answer is POWER!

Power (Watts) is the product of Voltage*Ampere, and it generates heat. So, power and heat will degrade your CPU.
If you have a high voltage at light load, you will have no problems. Imagine your CPU is running a load at 1.48V and consuming 40A, this means you have less than 60W of power to dissipate.
Now let's do the same, running a heavy load at 1.2V that draws 250A. Now we have 300W to manage...
Can you imagine running a load from 250W to 300W for 30 minutes on a CPU designed for 125W TDP?
It is worth remembering that the maximum power allowed by intel is 253W.

This is from intel datasheet:
Maximum Turbo Power: 253W
The maximum sustained (>1s) power dissipation of the processor as limited by current and/or temperature controls. Instantaneous power may exceed Maximum Turbo Power for short durations (<=10ms).

So, there is no problem to overclock the CPU. An overclocked CPU in real world (like gaming) will draw less than 200W for sure.
The major issue is submitting the CPU for a long period of testing that will test more your cooler solution than the stability of your overclock.
"It's not worth blowing up the engine to know its limit".

One thing I would like to make clear here. Intel has taken the 13900K to the extreme, which can be verified by the temperatures of the CPU running in stock. Therefore, this processor will not allow an overclocking margin like the previous ones. With a normal cooler and without big gimmicks, like the delid, full load frequencies of 55x to 56x are to be expected.

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''​

Some conventions:

Let's face it, when we follow a pattern things get easier...Imagine two people talking about a subject where one insists on referring to temperatures in Fahrenheit and the other in Celsius. Or one referring to the opening time of an injection nozzle (informing the volume of the cylinders and the air pressure) and the other talking about the air/fuel ratio. The same goes for the CPU and MB.

It doesn't make any sense to inform minimum voltage VID (telling which LLC is being used) if we want to know the minimum voltage of Die-Sense for a given frequency.

So we'll do some conventions:

- Whenever we want to know the minimum voltage for a given frequency, we will refer to Vcore (Die-sense).
No matter what your LLC, AC-LL or DC_LL configuration, for minimum voltage we will always use VCore voltage. (eg.: Full load P55x/E43x: Vcore=1.137v)

- Whenever we refer to a core frequency configuration we will use the following nomenclature:

12900K (Stock):
Max turbo - 5.2GHz
P-cores - 3.2GHz/5.1GHz
E-cores - 2.4GHz/3.9GHz

P: 52x1 – 51x2 – 50x4 – 49x8
E: 39x4 – 37x8
Full load @ P-49x/E-37x - VCore=1.18V
---------------------------------------------
13900K (Stock):
Max turbo - 5.8GHz
P-cores - 3.0GHz/5.4GHz
E-cores - 2.2GHz/4.3GHz

P: 58x2 - 54x8
E: 43x16
Full load @ P-55x/E-43x - Vcore=1.137V

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''​

Some concepts:

So we can talk and understand each other, let's make sure that some concepts are known, they are:

• “By Core” and “Sync all cores”:

The 13900K has 8 performance cores (#0 to #7) with independent frequency controls.
This means that we can assign each core a completely independent frequency limit, and they can each operate at different frequencies, depending on the processing demand.

On this page below, you can set the frequency and the number of cores that can run at the same time at that frequency.
View attachment 2579292

Above we have the following configuration:

P-61x2 - 59x4 - 57x6 - 55x8.

This means that if all cores is loaded the frequency will be 55x.
If 7 cores are loaded the frequency will be 55x
If 6 cores are loaded the frequency will be 57x
If 5 cores are loaded the frequency will be 57x
If 4 cores are loaded the frequency will be 59x
If 3 cores are loaded the frequency will be 59x
If 2 cores are loaded the frequency will be 61x
If 1 cores are loaded the frequency will be 61x
*Later we will see how to limit the frequency of a specific core.

Similarly, efficiency cores can also receive independent frequency assignments, but in groups of 4 cores.
Cores #8 to #11 form the first group.
Cores #12 to #15 are the second group.
Cores #16 to #19 are the third group.
And finally, cores #20 to #23 form the fourth and final group.

So you'd better set frequencies for a group of four cores.
View attachment 2579295

Above we have the following configuration:

E-48x4 - 47x8 - 46x16

And we have the same logic for the number of cores loaded.


If you set all performance and/or efficiency cores to the same frequency value, this setting is called “sync all cores”.
Then we can synchronize all performance cores to the same value (eg. 55x) and all efficiency cores to another value (eg. 43x).
This setting can be expressed as “Synced-P-55x/E-43x”.

Below we have all P-Cores synced to 55x.
View attachment 2579296

This setting (sync all cores) is still widely used today, although it no longer makes sense. When we do that, it's like we have one big performance core and another big efficiency core, each locked to a specific frequency. This configuration is valid to try to get a high score in some benchmark software. For everyday use, it's like locking your car's engine to a fixed RPM.

On the other hand, the “by core” configuration allows the CPU's internal algorithms, together with the operating system, to decide which core to assign the processing load to, as well as and control their frequencies within limits that we can stipulate. This setting is valid for both performance cores and efficiency cores.

The 13900K in stock configuration operates at:
P-58x2-55x8
E-43x16
In this configuration above, if the workload requires only 2 performance cores, they can run at 5.8GHz.
If all cores are required, the frequency will be 5.5GHz.

The advantage of the “By core” configuration is that we can create different usage rules for the cores, depending on workload, temperature and frequency.

For example, we can change your configuration to work like this:

P-59x2-57x4-55x8
E-45x4-44x8-43x16

In this case if the demand for performance cores requires only 2 cores, these will run at 5.9Ghz.
If the workload requires 4 cores, these will run at 5.7GHz.
And if the load requires all cores, the frequency will be 5.5GHz.

The same reasoning applies to efficiency cores.
As your demand for cores increases, the frequency decreases.

Maybe it's still not clear, but if we think about processing load, the rule we created makes tasks with less demand to be executed with higher frequencies and consequently faster.

Did you see the advantage over “sync all cores”?

What if we create a configuration as follows?

P-63x1-62x2-61x3-60x4-59x5-58x6-57x7-56x8

No problem if your CPU quality allows it!
If all cores are demanded, the frequency will be 5.6GHz, and will increase by 100MHz for each core that is not being used, until a task that requires only a single core is executed at 6.3GHz.

• “Adaptive voltage” and “fixed voltage”:

Once you understand the previous frequency dynamics, it becomes much easier to understand the difference between fixing the voltage and letting the voltage be applied adaptively.

Fixed voltage should be used when synchronizing all cores to a certain frequency. The adaptive will vary according to the number of cores used and the frequency applied.
The higher the frequency, the higher the voltage applied. There is a frequency/voltage table for this, the so-called “vf curve”.

• “VF Curve”:

The VF curve is nothing more than a table internal to the CPU that determines the voltage that the CPU must request from the VRM according to the operating frequency of the cores.
This table has a series of offsets that allow us to make voltage corrections for each specific frequency, allowing us to calibrate the voltage for each frequency.
The interesting thing is to find the minimum voltage for each frequency, so that the cores can vary their frequencies using the lowest possible energy level, generating less heat.
View attachment 2579291

It is also this table that gives rise to the famous SP number.
CPUs that have a table where higher frequencies demand a lower voltage have a higher SP number.

The “AI Overclock” BIOS page indicates what voltage is needed for each core to operate at 5.8GHz, and it is this set of information that is compiled into the SP number.

View attachment 2579290


Another great advantage of configuring the CPU in “by core” and “adaptive voltage” is that we can configure an independent adaptive voltage for each core, according to the table mentioned above. This operating configuration allows us to put the best cores to operate at higher frequencies.

On this page below we can limit each core frequency and assign it a unique adaptive voltage.
View attachment 2579299
In this example above we have cores #0 and #1 limited to 60x with an adaptive voltage of 1.446V. We can do this for each of the 8 cores.

• Load Line

Many like to tinker with their motherboard load-line settings to achieve better overclocking results. But how does this setting really work and how does the voltage output change with it? Check below to find out.

What is Load-Line?
The load-line setting, normally in mΩ (milliohms), determines how much the output voltage decreases when loaded.
This is derived from Ohm’s Law U = R*I. The drop in output voltage is calculated as load-line * Iout (output current).
For example a load-line of 1 mΩ and output current of 100 A, dU = 0.001 Ω * 100 A = 0.100 V.
At 1.300 V set-point output voltage, when loaded with 100 A the output would really be 1.300 – 0.100 = 1.200 V.
The primary reason for using a load-line in modern systems is to reduce voltage spikes (overshoot) when going from high to low output current and achieve a more predictable behavior.

Load-Line Levels or similar are profiles created by motherboard manufacturers to obfuscate and “simplify” different load-line values for users.
Another reason for these profiles is because additional VRM (Voltage Regulator Module) settings may need to adjusted along with the load-line value to keep it operating within spec.

The captures below show the output voltage transient behavior when loaded with about 70 A for ~150 μs.
The LLC1 capture illustrates ideal load-line behavior.
As the load-line value decreases (higher level), the line flattens and the under/overshoot spikes at start and finish become more pronounced.
The lowest voltage point at the beginning of the load transient does not improve much.
In this case, using a Load-Line Level of above 3 seems questionable.
The load voltage would increase meaning higher power consumption, but the worst case lowest voltage would stay the same.

Credits: ElmorLabs

• Understanding LLC, AC_LL & DC_LL:

Let's first understand load lines:

For this, didactically, we will exchange the electric current for a flow of water.

Our goal is to adjust the circuit so that the tanks remain at the same levels at all times, regardless of LOAD.

The resistance to the passage of water in the “Load Line” piping is physical and intrinsic to its construction.
(This is the physical wire from the VRM to the CPU).

As the LOAD varies all the time, the CPU tank level tends to get higher or lower than the VRM tank in an uncontrolled way.

So let's take control !

For that we have 2 pumps: the LLC pump and the AC_LL pump.

The first thing to do is to choose an LLC pump that will compensate for losses in the load line pipe.
(This is the VRM impedance characteristic, which determines the voltage drop as current flows).
Ideally, the pump should be neither too strong nor too weak.

We have 8 LLC pumps to choose from.
Pumps #7 and #8 are very strong and are not viable for daily use. So we have six pumps left.
It seems to me a good idea to choose an intermediate pump, #3 for example, but we can choose any one of these 6 pumps, as long as we make adjustments in the control circuit.

All this control will be done by the CPU, so we must inform which pump we choose through the DC_LL parameter. This way, the value of DC_LL (milliohms) must match the value of LLC (milliohms) chosen so that the CPU does all the calculations correctly.

The next pump to choose is AC_LL.
(This is the load variation compensation component).

This parameter makes the CPU, upon perceiving an increase in the water flow to the LOAD, to increase the VID value sent to the VRM, in order to anticipate the losses that this flow increase could cause. Therefore, the VRM uses the LLC and AC_LL pumps to fulfill the CPU VID request.

So if we have a stronger LLC pump, we can use a weaker AC_LL pump and vice versa.

Some combinations are not recommended, for example: two weak pumps or two strong pumps.

All this game can be done according to the desired goals.

Comparison with fixed voltage:

When we decide to use "voltage override" we turn off all this controls described before.
The selected LLC pump and the VID manually set will feed the level of the CPU tank without a control.
In this case, when the flow to the load rises, the level of the CPU tank goes down... And when the flow to the load decreases, the level of the CPU tank goes up...
So, bye bye level control.... 😂

You will need to run all the time with more voltage than you need, for that specific frequency, just waiting for when the heavy load comes.

If you think you can do a better job than the CPU algorithm, use fixed voltage, don't worry about AC_LL and DC_LL. Set an LLC #5, 6 or 7...
But I think it's a better idea to use this extra voltage to reach high frequencies...

• The LLC, DC_LL and AC_LL numbers:

This Is the Board Maximus Z-790 Extreme LLC Impedance table*:

LLC1: 1.75 milliohms
LLC2: 1.46 milliohms
LLC3: 1.1 milliohms
LLC4: 0.98 milliohms
LLC5: 0.73 milliohms
LLC6: 0.49 milliohms
LLC7: 0.24 milliohms
LLC8: 0.01 milliohms (flat).
*Some adjustment may be necessary.

The impedance values of the DC_LL shall be used according to the LLC chosen, so that the CPU performs its internal voltage and power calculations accurately.

Impedance stake:

DC_LL=LLC: The CPU performs correct VID and power calculations;
DC_LL<LLC: The CPU performs higher than real VID and power calculations;
DC_LL>LLC: The CPU performs lower than real VID and power calculations.

So, rule is: ALWAYS TUNE The DC_LL according to the LLC chosen.

And here I have very good news for you !!!!
Asus linked the LLC impedance to DC_LL automatically in the Z790.
So when you set a LLC#, if you let the DC_LL in AUTO the BIOS will adjust it for you !!!

The LLC controls the output impedance of our VRM, and MB vendors allow us to change and control this impedance.

DC_LL is the parameter that informs the CPU the VRM impedance.
If you decide to use LLC #1, the impedance is 1.75mohm, then you need to inform the CPU of this impedance using the DC_LL parameter.
If you use LLC #4 then the DC_LL should be 0.98.

To find the LLC impedance, you need to test (under load) one by one LLC and change the DC_LL until the VRM power matches the CPU power and the VID matches the Vcore. Once they match, you found LLC impedance.

AC_LL is a parameter that compensates for voltage loss due to your load line impedance, and you need to guess and test a different number for each LLC.

If you use a high impedance LLC you will need a higher AC_LL, on the other hand if you use a low impedance LLC you will need a low AC_LL
Note that on Asus MB, LLC # high means low impedance. And LLC # low means high impedance.

So never use a low impedance LLC with a high AC_LL... This will result in a very high voltage....
That's why I recommend an IA VR voltage limit of 1700mv...
So if you commit an error, the voltage will have a limit.

Simplifying the formulas:

You have the VF curve VID... Let's call it raw-vid.
That VID you see in hw-info, let's call it VID
You have the LLC impedance, let's call it LLC#
You have the DC_LL impedance parameter, let's call it DC_LL
You have AC_LL impedance compensation, let's call it AC_LL.
You have the CPU current, let's call it Amp.
You have the offsets and temperature components that we're not going to use to make things easier right now.
You have the Vcore, and let's call it Vcore.

So....

VID = raw-vid + (AC_LL * Amp) - (DC_LL * Amp)

Vcore = raw-vid + (AC_LL * Amp) - (LLC# * Amp)

This is how I found the LLC impedance.
If DC_LL = LLC than VID=Vcore at full load.

The Intel recommendation is to use AC_LL = DC_LL and LLC#3 (for asus MB)
So the intel recommendation is:

LLC = 1.1 mhom
AC_LL = 1.1mhom
DC_LL = 1.1mhom

It's a very conservative setting that will work with any poor MB.
Using AC_LL = LLC #, the lost voltage caused by VRM impedance will be compensated by AC_LL.

But we want to UNDERVOLT the CPU, right?

So, we will use AC_LL < DC_LL

But how to find the correct numbers?
Testing the combinations.
Each CPU will respond depending on the silicon lottery.

This is the page where you set the LLC#
View attachment 2579308

And this is the page where you set tha AC_LL, DC_LL and IA VR Voltage Limit:
View attachment 2581254

• C-State

C-states are states when the CPU has reduced or turned off selected functions. Intel processors support multiple technologies to optimize the power consumption.
C-States range from C0 to C10. C0 indicates an active state.
All other C-states represent idle sleep states where the processor clock is inactive (cannot execute instructions) and different parts of the processor are powered down.
We need to have C-States enable to use OCTVB.
I like to set the limit at C8, but you can let it ENABLE in AUTO.

These are the possibilities:
View attachment 2578492

This is the page where you set the C-States:
View attachment 2579310

• TVB:

The typical CPU has a standard clock speed and a turbo boost speed. However, CPUs with Thermal Velocity Boost have two additional Boost speeds. Also known as Intel TVB, the feature is available on 10th, 11th, 12th and 13th generation desktop chips.

TVB is a technology that takes advantage of the processor's thermal "opportunity" to increase its working frequency.
Thermal Velocity Boost allows these CPUs to achieve even higher boost speeds than their typical turbo boost.

This means that in addition to its standard clock speed and Boost of all cores, an Intel CPU can have four additional speeds.

Turbo Boost 2.0 is a single-core Boost available if the CPU is running to your power, current, and temperature specifications.
The velocity of turbo boost max 3.0 applies to two favorite cores. It is only possible if the CPU is running below its power, current, and temperature specifications.
Thermal Velocity Boost takes the fastest of the two favorite CPU cores at a higher speed than it gets with turbo boost max 3.0.
This is only possible if the CPU is running below 70 degrees Celsius and if the CPU is running below its power, current and temperature specifications.

The thermal speed increase of all cores refers to the reachable speed if all cores are active and the CPU is operating under its respective temperature limit (70 degrees Celsius).

• OCTVB:

The TVB overclock consists of changing boost patterns to achieve higher frequencies than standard when there is a thermal opportunity.

E.g. You can change the Boost of the 13900k to work this way.

Before +2Boost Profile:

P: 58x2 – 55x8
E: 43x16
Full load @ P-55x/E-43x

After +2Boost Profile:

P: 60x2 – 57x8
E: 43x16
Full load @ P-55x/E-43x

Note:
The Boost Profile will never be applied to the E-Cores.
The Full Load frequency will be the raw "By core" full laod due to temperature.

• ASUS OCTVB:

This is the page where you eneble the Asus OCTVB
Here you can find some other features like:

• Cache Dynamic OC Switcher - where you can define some rules for Cache Gear 1 or Gear 2 frequency;
• Voltage Optimizations - which is used to decrease voltages as per CPU temperature;
• Enhanced TVB - that use a especial algorythm to control the TVB voltage limit ;
• Overclocking TVB - where you can define the boost level;
• The TVB temperature offset - where you can apply a positive or negative temperature (in ºC) offset.

View attachment 2579313

• Explaining Asus OCTVB:

I'll try to make OCTVB easy...
Let's use my 12900K manual OCTVB settings and Asus OCTool as an example:

First line is for only 1 active P-core.
So this line will be used when only 1 core (anyone) is active, and the others are parked or not loaded.
In this condition, If Core temp is < 60 this core will run 57x.
If temp is 60 to 69 this core will run 56x.
If temp is >= 70 this core will run 55x

The second line is for when 2 cores (anyone) are active and the others are sleeping or not loaded.
If temp < 56 these 2 cores will run 57x.
If temp is 56 to 65 these 2 cores will run 56x
If temp is >= 66 these 2 core will run 55x

The third line is for when 3 cores (anyone) are active and the others are sleeping or not loaded.
If temp < 52 these 3 cores will run 57x.
If temp is 52 to 61 these 3 cores will run 56x
If temp is >= 62 these 3 core will run 55x

The fourth line is for when 4 cores (anyone) are active and the others are sleeping or not loaded.
If temp < 66 these 4 cores will run 55x.
If temp is 66 to 75 these 4 cores will run 54x
If temp is >= 76 these 4 core will run 53x

So let's go to the last line....

The last line is for when all cores are active and loaded.
If temp < 72 these 8 cores will run 53x.
If temp is 72 to 81 these 8 cores will run 52x
If temp is >= 82 these 4 core will run 51x


Once understood lets try some tricks:

You can use "8 Active Core tempB" = 100C to change the full load logic.

This way your full load will be 52x, because when 8 cores are active and loaded, and temp hit 72 the freq. will drop from 53x to 52x... and the next temp step is the TJmax.


Another trick:

You can chage the BinA (or BinB) to force 2 drops:

So when 8 Active cores are loaded and temp hits 72C, freq. will be 51x, and when they hit 82C, freq. will be 50x.

TempA is linked to BinA and TempB is Linked to BinB

You can change BinA and BinB changing the frequency more than 1 step in any position.


Adding few "º C" to the table:

This table below is an Asus +2Boost profile automatically calculated by the Asus algo for the following "by core" configuration:
56x2 - 55x3 - 53x5 - 51x8. (+2 Boost added)

You can edit all the temps adding a few degrees..

+15C to the 53x8 (6,7,8 active cores)
+ 5C to the 55x5 (4 and 5 active cores)
And keep 57x3 and 58x2 untouched.

And you can try any kind of combination that you can boot... LOLOLOL

I use to test with the Asus OCTool and when I find some nice setting I write to the BIOS.

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''
Here you can find all about Intel 13th CPU

Dropbox

www.dropbox.com

'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''


I believe we already have enough theory, what do you think?

Shall we move on to the practical part?​

Click to expand...

Bradicardia said:

E-cores 4x56 8x55 16x54

Click to expand...

Bradicardia said:

Hi All, amazing guide Roberto.

First time doing overclock, trying to get familiar with all the concepts and settings.
Actually configured with LLC4 & IA AC: 0.17
P-cores 8x55
E-cores 4x56 8x55 16x54

My motherboard is the Asus Prime z790-A WIFI, so I hope that the automatic value of IA DC is working in this motherboard too.
Now I wanted to continue with the "P-core overclocking" section, but I see that it is talking about modifying the VF curve. For now I didn't setup the VF curve, how should I configure it? trying offset until I find the sweet spot for each section? I know how to test for maybe the full Px55Ex44 runing cinebench with static vcore, but how I can know the voltage for the other frequencies?

Another question, on my non-premium motherboard, I cannot know the SP values, any other way to know this?

Thanks!

Click to expand...

mattxx88 said:

do you know why it makes like this, changing ring values?

Edit: i noticed i launched cb23 on first screen single core, but it doesnt effect the boost issue

default:
View attachment 2588221
View attachment 2588218

after:

View attachment 2588220
View attachment 2588219

Click to expand...

RobertoSampaio said:

I would leave the min ring on auto.

Click to expand...

Darccon said:

Amazing detailed guide. You might needed hours of work put into it.

Last time doing oc was on x99 and an intel I7 6800K. Thank you very much, explaining 13th Gen OC to me.

My current config:

Maximus Hero z790 - I9 13900 K; running at following config:

disable all limits
LLC4 & AC: 0.04 / DC 1.02
VF 8,9,10: +0.083
P-cores 2x58, 2x57x 2x56 2x55
E-cores 4x46 8x45 16x45
Rest running at auto

The final cinebench R23 and Geebench Score always is nearly the same result, no matter if running at stock or overclocked.
During cinebench, I am hitting 253watt and vocre of 1.285. (will share some screenshots, when I am back at home)

What could be the the reason of not getting increased numbers while overclocked? Low voltage?

Many thanks for any response.

greets Rene

Click to expand...

pipes said:

Which is min value for ring?

Inviato dal mio MI 9 utilizzando Tapatalk

Click to expand...