Z790 TOMAHAWK MAX WIFI + i9 13900KF. Оptimal settings?

[alex-predator]

Z790 TOMAHAWK MAX WIFI + i9 13900KF. Оptimal settings?

My configuration:
- MSI MAG Z790 TOMAHAWK MAX WIFI
- Intel Core i9-13900KF
- Liquid Freezer III 360 ARGB Black
- DDR5 ADATA 64Gb (2x32) 6400 XPG Lancer RGB (2&4 slot)
- Gigabyte GeForce RTX 4070 Ti SUPER 16Gb
- SSD M.2 ADATA SX6000LNP 1Tb
- SSD M.2 ADATA Legend 960 MAX 4Tb
- HDD Seagate 2Tb
- CASE Zalman Z10 PLUS Black
- PS Gigabyte UD Gold 1000W

Hello everyone! I apologize for my English)). I have tried a lot of BIOS settings to cool my CPU, but it did not lead to anything good, until I found the guide “Guide: How to set good power limits in the BIOS and reduce the CPU power draw”.
I would like to thank @citay for this information!

Updated the BIOS to 7E25vA7 (06.09.2024).
BIOS settings:
- XMP Profile 1 – Enabled
- Long Duration Power Limit – 253W
- Short Duration Power Limit – 253W
- CPU Current Limit – 307A
- CPU Lite Load – Mode 16 (by default)
- Intel Turbo Boost Max Technology 3.0 – Disabled
- IA CEP Support – Disabled

With these settings I started testing the computer with HWinfo64 + Cinebench R23, constantly lowering CPU Lite Load from “Mode 16” to “Mode 1”….

The optimal “Mode 3” setting was selected. Excellent temperature and power consumption, and I haven't gotten to “Guide: How to set up a fan curve in the BIOS” yet, but the 36098 pts (Cinebench R23) test result falls short of the published results of 40255 pts.

What BIOS settings can be used to increase performance in tests, at least up to 39000 pts?

 

[alex-predator]

Hi everyone!​

Had some time to make the recommended settings and run some tests.

Installed the new BIOS E7E25IMS.A80 (09/27/2024) and configured:

Power Profile: MSI Perfomance
IA CEP - Disable;
Turbo Boost....3.0 - Disable;
Enhance turbo - Disable;
Intel C-State - Enable;
TVB Ratio Cliping - Enable;
TVB Ratio Cliping Enhanced - Enable;
TVB Voltage Optimization - Enable;
C1E Support - Enable.
Room temperature - 27, AC-LL/DC-LL: 0.3/1.100

I started the tests with high values...

1. LP1: 253W; LP2: 253W; IccMAX: 400A; CPU Lite load; MODE 5.
Test results:
R23 - failed.
R15 - failed.

2. LP1: 253W; LP2: 253W; IccMAX: 400A; CPU Lite load; MODE 6.
Test results:
R23 - 38578 pts, t max - 93, A max - 224, W max - 278, V core max - 1.306.
R15 - failed.

3. LP1: 253W; LP2: 253W; IccMAX: 400A; CPU Lite load; MODE 7.
Test results:
R23 - 38307 pts, t max - 95, A max - 230, W max - 281, V core max - 1.312.
R15 - 1586 pts, t max - 90, A mah - 216, W max - 269, V core max - 1.304.

4. LP1: 253W; LP2: 253W; IccMAX: 400A; CPU Lite load; MODE 8.
Test results:
R23 - 37904 pts, t max - 95, A max - 231, W max - 279, V core max - 1.318.
R15 - failed.

It was decided to lower IccMAX, as practice has shown MODE 6 is the most stable, so I started the tests with it....

5. LP1: 253W; LP2: 253W; IccMAX: 380A; CPU Lite load; MODE 6.
Test results:
R23 - 38433 pts, t max - 93, A max - 219, W max - 271, V core max - 1.316.
R15 - failed.

Decrease: 370A.

6. LP1: 253W; LP2: 253W; IccMAX: 370A; CPU Lite load; MODE 6.
Test results:
R15 - failed.

Decrease: 360A...

7. LP1: 253W; LP2: 253W; IccMAX: 360A; CPU Lite load; MODE 6.
Test results:
R23 - 38161 pts, t max - 87, A max - 209, W max - 249, V core max - 1.318.
R15 - 1583 pts, t max - 85, A mah - 214, W max - 258, V core max - 1.322.

8. LP1: 253W; LP2: 253W; IccMAX: 360A; CPU Lite load; MODE 7.
Test results:
R23 - 38180 pts, t max - 90, A mah - 211, W max - 256, V core max - 1.296.
R15 - 1588 pts, t max - 86, A mah - 216, W max - 256, V core max - 1.296.
Prime95 - failed.

9. LP1: 253W; LP2: 253W; IccMAX: 360A; CPU Lite load; MODE 8.
Test results:
R23 - 38088 pts, t max - 92, A max - 216, W max - 278, V core max - 1.350.
R15 - 1579 pts, t max - 87, A mah - 217, W max - 267, V core max - 1.302.

From the above described trio the most optimal was MODE 7, but it failed Prime95 test. I decided to lower LP1/LP2 to 230W....

10. LP1: 230W; LP2: 230W; IccMAX: 360A; CPU Lite load; MODE 7.
Test results:
R23 - 37693 pts, t max - 89, A max - 208, W max - 253, V core max - 1.300, core VIDs max - 1.338.
R15 - 1570 pts, t max - 82, A mah - 201, W max - 247, V core max - 1.350, core VIDs max - 1.369.
Prime95 - passed, t max - 91, A mah - 254, W max - 255, V core max - 1.062, core VIDs max - 1.279.

All 3 tests passed, with relatively normal values, but for esperement decided to repeat the tests on pure Windows 11, with drivers only...

10. LP1: 230W; LP2: 230W; IccMAX: 360A; CPU Lite load; MODE 7.
Test results:
R23 - 37963 pts, t max - 90, A max - 211, W max - 250, V core max - 1.356, core VIDs max - 1.363.
R15 - 1583 pts, t max - 85, A mah - 213, W max - 239, V core max - 1.346, core VIDs max - 1.352.
Prime95 - passed, t max - 88, A mah - 238, W max - 256, V core max - 1.050, core VIDs max - 1.112.

I did this to understand how parallel processes from other softwareaffect the performance, the difference is not very big but it isthere.

I am still unsatisfied with the indicators because in the table at @FlyingScot has a user @NoneOfYourBeesWax with the same processor as mine and with a motherboard on the same chipset, with result R23 - 39500, but by his settings I don't get even approximate result R23. With settings like @NoneOfYourBeesWax I get R23 - 38500, R23 - failed. Prime95 - failed.

What is wrong with my settings? Maybe I'm doing something wrong, give me some advice on which way to look?

 

[MigraineFilms]

I am still unsatisfied with the indicators because in the table at @FlyingScot has a user @NoneOfYourBeesWax with the same processor as mine and with a motherboard on the same chipset, with result R23 - 39500, but by his settings I don't get even approximate result R23. With settings like @NoneOfYourBeesWax I get R23 - 38500, R23 - failed. Prime95 - failed.

What is wrong with my settings? Maybe I'm doing something wrong, give me some advice on which way to look?

My i9 14900k reaches 38500/39000 with PL230 at 85 degrees with a Deepcool LT720. It's just luck of the draw, some run hotter, some run cooler, some have 500 points more score, some 500 lower.

I think you're really good with these settings, but if you want, you can try a manual undervolt and lower Load Line Calibration (LLC) to prevent the VDROOP under load to dip below stable levels. This might push out some extra performance.

So, lets take Profile Number 2 in your list, albeit a bit toasty, if you set the LLC to mode 7, and DC value to 80 (to hopefully prevent IA CEP from kicking in), you might be able to run it stable. It might also be a bit hotter, but most likely it will use more voltage and less current to reach the same power (which in the end is the biggest heat generator). If that still isnt stable, lower LLC to mode 6 and DC to 40 to see if thats stable.

This is a lot of hassle which require A LOT of reboots to get it right, so only do that if you really want that extra performance.

For example: I set my PL to 230, AC/DC 40, LLC Mode 6, -0,100 undervolt. I reach 38500 R23 score, and everything is stable. When I start a adobe after effects render, which is heavy CPU bound, it will draw 188 watt, so all my tuning did exactly nothing for real world performance.

 

[alex-predator]

My i9 14900k reaches 38500/39000 with PL230 at 85 degrees with a Deepcool LT720. It's just luck of the draw, some run hotter, some run cooler, some have 500 points more score, some 500 lower.

I think you're really good with these settings, but if you want, you can try a manual undervolt and lower Load Line Calibration (LLC) to prevent the VDROOP under load to dip below stable levels. This might push out some extra performance.

So, lets take Profile Number 2 in your list, albeit a bit toasty, if you set the LLC to mode 7, and DC value to 80 (to hopefully prevent IA CEP from kicking in), you might be able to run it stable. It might also be a bit hotter, but most likely it will use more voltage and less current to reach the same power (which in the end is the biggest heat generator). If that still isnt stable, lower LLC to mode 6 and DC to 40 to see if thats stable.

This is a lot of hassle which require A LOT of reboots to get it right, so only do that if you really want that extra performance.

For example: I set my PL to 230, AC/DC 40, LLC Mode 6, -0,100 undervolt. I reach 38500 R23 score, and everything is stable. When I start a adobe after effects render, which is heavy CPU bound, it will draw 188 watt, so all my tuning did exactly nothing for real world performance.

So you think that my settings and results are fine, and I shouldn't change anything further?I just want my cpu to live happily ever after and give good performance on 3D renders. I just don't understand, will a 1500 pts difference make a big difference in rendering speed?

I just don't understand why @NoneOfYourBeesWax also has a processor with almost the same configuration getting 39500-R23 and t87 at 253/253/400 with air cooler compared to my 38500-R23 and t93 with water cooling? Something is not "clean" here, too much difference)) @NoneOfYourBeesWax share your secrets!))

 

[MigraineFilms]

7. LP1: 253W; LP2: 253W; IccMAX: 360A; CPU Lite load; MODE 6.
Test results:
R23 - 38161 pts, t max - 87, A max - 209, W max - 249, V core max - 1.318.
R15 - 1583 pts, t max - 85, A mah - 214, W max - 258, V core max - 1.322.

I'd personally go with this setup. If thats completely stable (although, if Lite Load Mode 7 isnt stable, I don't understand why Mode 6 is). Maybe lower IccMAX to 307, and see if it makes a difference. Your readings show A Max 214, so lowering ICCMAX to 307 would make no difference (At least, thats how I interpret it).

My Deepcool LT720 doesnt seem to be able to cool more than 260 watt either, at 230, it might even hit 85 degrees. Reviews are saying it could cool 330 watts, but not for me. I don't know why, I changed the cooling paste, put on a contact frame, tuned my CPU, yet it still thermally throttles at 253 PL 1/2.
We cant all be winners with that, but here's the thing:

An R23 score of 39000 is only 2,6% more than 38000. When gaming, the CPU will use about 100/120 watts, never the 250. That little bit extra performance only affects the compiling of data and rendering of content. And even if that's the case, it would be AT BEST 3% faster. So, if you're shader compiling at the start of your game, and it takes about 35 seconds, that would take 34 seconds if you max out the CPU.

During gaming, make sure that your cores reach the 5,6ghz boost. If your voltage is too low, those frequencies will not be reached. Thats the most important thing to look for during gaming.

You undervolted it, you're making sure that the temperatures do not top 90 degrees, and you get a better score than my i9 14900k. You're good in my opinion, and did everything to raise the lifespan of the CPU.

As the temperatures go: Maybe its "bad" soldering in the CPU itself, maybe the heat spreader doesnt make the most efficient contact on the cores, maybe the cooling block of the pump isn't as smooth as it could be, maybe the cooling paste is too thin or too thick, maybe there is too much hot air in the PC that's being pushed through the radiator fans so the water doesn't cool as efficient as it could. There are so many factors in play that I've just taken the loss and let it be. The AMD R7 9800x3d is way better in that regard, intel is just too hot to handle.

 

[alex-predator]

Я бы лично выбрал эту настройку. Если она полностью стабильна (хотя, если Lite Load Mode 7 нестабилен, я не понимаю, почему Mode 6 стабилен). Может быть, снизить IccMAX до 307 и посмотреть, будет ли разница. Ваши показания показывают A Max 214, поэтому снижение ICCMAX до 307 не будет иметь никакого значения (по крайней мере, так я это понимаю).

With Mode 6 this setting also fails the R15 test. Thanks for the tip! I will try this setting with IccMAX 307A.

 

[MigraineFilms]

Try LLC (in DigitALL settings) mode 7 or 6. It will "remove" your undervolt during heavy load (less heavy vdroop). Maybe it gets a bit hotter, but it will become stable again.
This worked best for my CPU.

 

[FlyingScot]

if you want to stick to AC-LL/DC-LL: 0.3/1.100 then just keep walking down the LLC until you get it stable. But don’t go lower than 3.

When you hopefully find stability, then we’ll need to lower that DC_LL setting from 1.100 to more closely match your new LLC setting.

I’d keep power limits conservative but still high-performance for this test, like 253/253/307A. My guess is that things should stabilize at LLC5. If that doesn’t work, then try with C1E=Disabled. If nothing works then we need to step back and look at the whole picture. You have relatively high room temps, which doesn’t help.

 

[alex-predator]

Попробуйте LLC

если вы хотите придерживаться AC-LL/DC-LL: 0,3/1,100, то просто продолжайте идти вниз по LLC

My head is still a mess, misunderstanding all the BIOS settings.)) Where these “AC-LL/DC-LL: 0,3/1,100” settings can be changed? I got them from OCCT. If I understand you correctly, I need to lower/change lite load calibration (LLC)?

 

[MigraineFilms]

My head is still a mess, misunderstanding all the BIOS settings.)) Where these “AC-LL/DC-LL: 0,3/1,100” settings can be changed? I got them from OCCT. If I understand you correctly, I need to lower/change lite load calibration (LLC)?

In the BIOS, if you set Lite Load to "Advanced" you can manually input the AC/DC Load Lines. There you can manually set AC-LL to 30 and DC-LL to 110. This would correspond with LL Mode 6 on your board if I understand correctly (This is different on every board).
If you then set Load Line Calibration (in DigitALL settings) to Mode 7 or 6, you should (if you have IA CEP enabled which I dont think you have) lower the DC-LL to match the LLC Mode (also different on each board, nobody knows the exact numbers)
This is very fiddly, but we can guide you in the right direction.

Set Lite Load to Advanced
Set AC to 30
Set DC to 80
Set LLC to Mode 7

or:
Set Lite Load to Advanced
Set AC to 30
Set DC to 50
Set LLC to Mode 6

this should roughly (VERY ROUGHLY) match the correct settings.

 

[FlyingScot]

My head is still a mess, misunderstanding all the BIOS settings.)) Where these “AC-LL/DC-LL: 0,3/1,100” settings can be changed? I got them from OCCT. If I understand you correctly, I need to lower/change lite load calibration (LLC)?

No worries. It's very easy when starting out to get confused by all these similar sounding names. EDIT: MigraneFilms is on top of it!

 

[FlyingScot]

My head is still a mess, misunderstanding all the BIOS settings.)) Where these “AC-LL/DC-LL: 0,3/1,100” settings can be changed? I got them from OCCT. If I understand you correctly, I need to lower/change lite load calibration (LLC)?

If you can stand the extra complexity of reading the backstory behind these settings you are playing with, try to read my write-up here.

 

[alex-predator]

If you can stand the extra complexity of reading the backstory behind these settings you are playing with, try to read my write-up here.

Thank you very much for your write-up! I read your write-up for the second time the other day, everything is clear and clearly described, but still for me many things are not clear and confusing, especially everything related to electricity)))

 

[FlyingScot]

Thank you very much for your write-up! I read your write-up for the second time the other day, everything is clear and clearly described, but still for me many things are not clear and confusing, especially everything related to electricity)))

I'm not surprised. Intel's Loadline settings are quite confusing. You would think that we would just have one single setting to control voltage to the CPU from the motherboard. But instead we have three. That's three levers that we push and pull on until we find a combination that unlocks lower power usage and lower temperatures while still remaining stable. In fact, think of it as just that, a combination lock!

If you have a specific question please feel free to ask away. You can cut and paste text from my guide if you want to add context to your question.

 

[alex-predator]

@FlyingScot @MigraineFilms How long does it take to do the Prime95 test?

 

[MigraineFilms]

20 to 40 minutes is usually enough, I let OCCT run for an hour mostly

If there is instability, it will often show up in the first five minutes, the rest of the time is just to make sure its really stable

 

[alex-predator]

@FlyingScot @MigraineFilms I have been testing my computer for half a day and noticed that LLC 7 is the most stable, when I start lowering LLC (6, 5, 4, 3), my performance gets worse, namely R23 pts decreases, temperature increases, A and W increase, V increases. Why should I not start with LLC 8?

 

[alex-predator]

..... And within what limits is it best to stick to V core and core VIDs?

 

[FlyingScot]

@FlyingScot @MigraineFilms I have been testing my computer for half a day and noticed that LLC 7 is the most stable, when I start lowering LLC (6, 5, 4, 3), my performance gets worse, namely R23 pts decreases, temperature increases, A and W increase, V increases. Why should I not start with LLC 8?

It all depends on your objectives. If the most stressful thing you usually do is playing games rather than heavy all-core workloads (like R23) then there is a benefit to a lower LLC setting because you can more easily combine it with a stable manual voltage offset (with or without CEP enabled). This setup will also restore some of that performance you see with LLC8 when running R23. So, this would be part 2 of your more advanced tuning.

If however, you want to keep things more simple then CiTay’s guide is where you start and end, which just has you stick with LLC8 (Auto) and lower CPU Lite Load (to lower the AC_LL). Your voltages (and temps) in games might be a little (or a lot) higher, but the trade-off is that you kept life simple.

 

[alex-predator]

It all depends on your objectives. If the most stressful thing you usually do is playing games rather than heavy all-core workloads (like R23) then there is a benefit to a lower LLC setting because you can more easily combine it with a stable manual voltage offset (with or without CEP enabled). This setup will also restore some of that performance you see with LLC8 when running R23. So, this would be part 2 of your more advanced tuning.

If however, you want to keep things more simple then CiTay’s guide is where you start and end, which just has you stick with LLC8 (Auto) and lower CPU Lite Load (to lower the AC_LL). Your voltages (and temps) in games might be a little (or a lot) higher, but the trade-off is that you kept life simple.

That's the thing that I almost do not play games, I work with 3D programs (3Ds Max), engaged in 3D rendering, and for me is important performance in this process, the faster I will make a render, the more I will earn and spend less time.

 

[FlyingScot]

..... And within what limits is it best to stick to V core and core VIDs?

Good question. This is the heart of what we are trying to do when it comes to Raptor Lake CPUs and their unfortunate history of self-destruction due to flirting with high voltages. We are trying to lower voltages as low as we can get them without getting too deep into the weeds.

The first step is to use CiTay’s guide, as that is a relatively quick way to bring down voltages with minimal stability testing. However, while CiTay’s approach is great for getting results in those stability tests (e.g. Cinebench R23, Prime95, OCCT) it is not quite as effective at bring down voltages for normal daily activities like web surfing, MS Office, video watching, and of course gaming.

To your specific question: You might establish a baseline by using your computer as you would normally, including playing games (if that’s what you do) while you keep HWInfo running in the background. I usually advise keeping it running for a couple of hours or more of normal work, or at least 40mins of game play with several different titles. Then note down the peak VID and Vcore values from the Maximum column.

As a general guide, IMO, here’s how your voltage question can be answered:
Voltage peaks above 1.5V = potentially bad for the life of a RPL CPU.
Voltage peaks between 1.45V - 1.49V = possibly okay, but still in the historically uncomfortable zone.
Voltage peaks between 1.40V - 1.44V = likely okay, but not ideal.
Voltage peaks between 1.35V - 1.39V = should be safe
Voltage peaks below 1.35V = very safe (historically) and easier on your cooling solution.