Z790 Chipset & Raptor Lake

[dontblink]

OK so after removing the cpu cooler, there was a huge spot in the center with no thermal paste, I guess my CPU or cooler must have a light convex shape. (first time did a middle dot of thermal paste, this time I did a vertical line)

So that would have affected the temps for sure, so hard to tell how much the CPU frame helped, but I'm sure it did, but regardless for $10, totally worth it and such an easy install.

The CPU can now handle 5.4Ghz all core without throttling, lowest core at 75C and hottest at 92C
And could push the e-core to 4.3Ghz (4.4 would trigger throttling)
Wattage is now at 268watt

Then if I push all core load to 5.5Ghz, I'm getting throttling after 10seconds, so does allow a nice little 10second boost, but doesn't really drop, just a few blips down to 5.4, until after about a minute, then it settles around at 5.4Ghz, and pulling 300watt initially then throttling down to 290watt, so in that sense 290watt seems to be my new wattage limit with the noctua, which is a huge improvement over the 250watt limit I was hitting before.

And score in windows is shy of 40000 in cinebench R23 in single pass, 10min test got a 38500 score.

Also updated the max boost based on core load to 60 60 60 60 60 59 58 55
and kept the v-core offset to -0.04

then tried 62 62 62 62 60 59 58 55, but somehow doesn't stay steady at 6.2 on single core cinebench r23, and oscillates between 5.9 and 6.2, but doesn't show any throttling in the intel utility, but most of the time seems to stay around 6.0 to 6.1Ghz, but no crash nor any sign of instability. Cinebench R23 single score single run did increase and reached 2300 (windows).
So might try that for a little bit and see if I get any instability signs.

So turns out the noctua really isn't as bad as I thought, and would recommend it for an air cooler on the 14900K + thermalright cpu frame.
I kept using the noctua NT-H1 that came with the cooler for thermal paste.

Edit:
I don't know enough to know why, I'm guessing temp spikes were the cause before, but the CPU is now more stable with lowered vcore offset, I'm now doing some test at -0.07 (which would have instantly crash before I reapplied thermal paste+cpu frame), but no issue so far, and is stable at 5.5Ghz all core, pulling about 270watt at that vcore offset (in a cinbench R23 load) which the noctua can handle without throttling settling at 96C. (from further testing 285watt is where the noctua start struggling on my setup)
I now have 61 61 61 61 60 60 60 55 for turbo.
6.2Ghz was acting strange, so I gave up on that, but 6.1Ghz seems fine at least with vcore around -0.04 (but was limited to 5.4Ghz then on all core load), giving even higher single core score around 2375, so now seeing how stable it is with this new -0.07 vcore offset. But will settle at -0.06v to play safe since temps are still fine.

I get throttling for a fraction of a second as the fan ramp up at the start of a cpu load.

What pattern did you use to apply the thermal paste? Stripe?

Edit: Doh its in the second line. My bad.

 

[eliaden]

What pattern did you use to apply the thermal paste? Stripe?

a vertical line

----

edit: I'll take the opportunity of this new post (and likely last post regarding changing max boost and v-core offset) to show my latest findings.

MacOS seems more sensitive to the all core speed change, at least Geekbench was, (multiple run at 30min of cinebench R23 and 2024 in windows was fine and was usually a good defining factor of instability at 61 61 61 61 60 60 60 55 and -0.06v offset all was great)

Geekbench 6 in macOS was very sensitive ( Geekbench 5 was too but not as sensitive ),
So now ending up with 60 60 60 60 59 59 58 55 which is seemingly stable and still have v-core offset at -0.06
A single core at 61 and offset down to -0.035 was still unstable, and 7 core at 60 was also unstable.
E-core are still at 4.3Ghz (the voltage required for 4.4Ghz is too much for the noctua, or unstable with v-core too low)

macOS with the new settings and since the CPU frame added + better thermal paste application, cinebench R23 is showing higher score around 38200 single run; over extended loads the wattage creeps up from 265w to likely 280w (same behavior I noticed in windows), which seems to trigger a few temporary throttle as the wattage then drops down to 250w, but then kicks back up to 270w. The 10min run score is 37500, which isn't bad at all.

On the windows side with those new settings I expected single core to take a hit, but single run in windows cinbench r23 are 2355(single core)/39951(all core) which are some of the best I got (and roughly lines up with score posted online on review sites), and I now know it's super stable or much closer to being super stable. I'm sure 10min run or so will also take a score hit, but an expected one, as my settings are ridding the limit of the noctua cooler.

Hopefully the journey I went through will be helpful. But the lesson I learned is play with boost speed, I think lowering the e-core by 0.1Ghz was the best move overall with the noctua for sustained all core loads, which in my case is important as I do a lot of 3d rendering.

edit: had one crash, WindowServer in MacOS during 3d rendering, so I changed my core settings to 60 60 60 60 59 58 58 54, and lowered core offset to -0.05v to play it safe, now I'm well below the max wattage the noctua can handle, maxing out at 250watt (the noctua seem to handle 270~280watt)
My rendering time in Fusion360 was similar within a second, actually better at 5.4Ghz vs 5.5Ghz boost within a second on a 30second render, which is what I actually care (as opposed to cinebench results).

 

[WoodCabin]

Hello @eliaden, I have the same motherboard and microprocessor as you, it would be great if you could take some screenshots of your bios settings or save the settings in a .cmo file (I think) so I can load it on my system and experiment with my equipment In my previous X299 system I understood certain overclock and voltage parameters, but with this system I didn't know how to start, I see that you have achieved good results, greetings.

 

[eliaden]

Hello @eliaden, I have the same motherboard and microprocessor as you, it would be great if you could take some screenshots of your bios settings or save the settings in a .cmo file (I think) so I can load it on my system and experiment with my equipment In my previous X299 system I understood certain overclock and voltage parameters, but with this system I didn't know how to start, I see that you have achieved good results, greetings.

For anyone else reading this, this is for a Gigabyte Z790 AERO running F9 BIOS, and a 14900K.
Has a Thermalright CPU frame and Noctua NH-d15 (using both fans).

Also keep in mind every single CPU is different even if sold as a 14900K, and some may not like these Core boost settings and vcore and fail to boot, and some can be pushed further than mine.

USE AT YOUR OWN RISK, there is nothing bad per se in this BIOS settings, like no increased in Vcore value, so nothing should melt, but under voltage can cause instability and fail to boot. This board has so far always allowed me to get back into the BIOS when I played with unstable settings.

Some BIOS settings, need a corresponding EFI settings to match in order to work with macOS, of the top of my head, ResizeGpuBars and VT-d setting are two important ones.
Also the iGPU is disabled.

---

Here is my current BIOS (F9) saved setting I'm using (there's a beta BIOS F10b, but would avoid that for now if loading my bios settings), tweaked for macOS but works great in Windows 11.

In the Advance menu, you'll see Vcore offset, set at -0.06. You can try as is, otherwise go down to -0.02 and then test in -0.005 increments. The first time I had the cooler installed and no CPUframe, the system wouldn't even boot at lower V-core offset.
In the advance CPU settings, you'll find the boost value 60 60 60 60 59 59 58 55, and the e-core there are 8 places set to 43 (as in 4.3GHz).

I highly recommend using the Intel Extreme Tuning Utility in windows (I didn't install any gigabyte utilities, only drivers one by one from gigabyte) and live test v-core offset and max boost speed in case things aren't stable, and check temps, then manually set those in the bios as value tested in the Intel utility won't stick. that will get you closer, then like I did last was Geekbench 6 in macOS, since it appeared to be the most sensitive to max boost/v-core offset tweaks (it simply stops, so no score displayed, but doesn't close fully, and if you check the console.app under crash report, you'll get confirmation that it did crash during cpu benchmark)

I have my fan profile set to 100% fan speed once reaching 60C (you can likely push that to 70C or even 80C), and just progressively slower under that, so that it's silent at 40~45C (like right now basically idle, hovers around 33~40C), none of my fan are high rpm even at 100%, I have two Noctua 140mm 1200RPM for my front intake, and the included fan in the NZXT H5 flow also 1200RPM (120mm) for bottom intake and rear exhaust.

 

[WoodCabin]

Thanks, yes, I have a configuration the same as yours (motherboard and BIOS). Right now I have a 13900K,and I will buy the 14900kf soon. I have a Thermal Grizzly frame with an Artic Freezer 420 AIO. When I get the processor, I will compare the stock BIOS with yours, and I will study the values to try to adjust them to the silicon of my processor. There is a lot of work done with your bios, I will test your values and proceed with caution to try to achieve an ideal voltage for the processor. Thanks again for this fantastic work. I will report back as soon as I get results. Greetings.

 

[Feartech]

Right now I have a 13900K,and I will buy the 14900kf soon.

is it really worth the upgrade?

 

[WoodCabin]

The truth is that I think there is only a 10-15% improvement. But I have to build another team, and I will try the 14900KF on this occasion. The price in Spain is also similar.

 

[Feartech]

The truth is that I think there is only a 10-15% improvement. But I have to build another team, and I will try the 14900KF on this occasion. The price in Spain is also similar.

There doesn't seem to be that much of a difference:

Core i9-14900K vs. Core i9-13900K: what was Intel thinking? | Digital Trends

Intel's new Core i9-14900K is the fastest desktop CPU from Team Blue, but is it worth upgrading from the Core i9-13900K? We break down the differences.

www.digitaltrends.com

 

[WoodCabin]

According to the article there is not much difference, I don't know where I read that there was more difference between the two but hey, considering buying a new one, the price difference is not big, plus it is good to "scratch" a little more performance, I don't know the time that Apple will allow us to continue enjoying its operating systems with Intel processors, what is a shame is that there are no drivers for AMD 7900 graphics cards, that would be a great advance...

 

[schreiberstein]

Built a new system just now, have not touched macOS on this one yet, as I was still tinkering with optimization and stability on Microsoft Windows.

My 14900K on Z790 Aero G, with Noctua U12A and Thermalright CPU frame, thermal throttles on Windows after 10-15 seconds on stock settings. I have now manually locked the frequency to 5.3 GHz for the performance cores and 4.3 GHz for the efficiency cores. Results are good. CPU never thermal throttles, clocks are always 100% stable, temperatures rarely go over 90+ in Cinebench. And ironically, I get around the same Cinebench score (38000 - 39000) as I would with all this "boost" stuff (I mean the out of the box behavior) enabled. Because this is ultimately where the CPU will settle anyway after some time..

As one user stated before, shaving 100 MHz off of the efficiency core clock (locking them to 4.3 GHz) really made a significant difference.

I think all this 'boost up to 5.7 GHz if X amount of cores are in use' stuff only matters if you are on Windows, where the scheduler takes advantage of all that. macOS does not, as far as I know. And so, I think it would just be better to have a fixed setting of 5.3GHz/4.3GHz. The CPU cooler does not howl anymore just because one core boosts to 5.7 GHz, as some random task decides to run at 100%. Does this task *really* need the additional 300-400 MHz, at the cost of the other cores lowering their clock-speed significantly? Seems like diminishing returns to me. The CPU is still mighty fast. For perspective, I come from a Z390 Designare / 9900K setup. Boosting to 5 GHz was rare and difficult to achieve on air cooling, so I never bothered with it. Most cores ran at 4.4-4.8 GHz if I recall correctly.
Aside from all the other useful platform improvements (DDR5, PCIe 4/5, etc.) introduced in recent iterations, we essentially get an all-core stable 5.3 GHz 9900K, with 16 additional 4.3 GHz cores (so basically similar to 9900K cores that are currently thermal-throttling), which hover around 90 Celsius under full load.

I think it would have been more honest to advertise this CPU with these clock rates (5.3GHz/4.3GHz) instead, and everything else is just "overclocking headroom" for those who invest a lot into cooling, optimizing or sacrificing noise levels. (i.e. enthusiasts and gamers)

BTW. undervolting worked well with this configuration in Windows, stable until -75mV with Intel Stress test and Cinebench. Temps in Cinebench went down to just 75-80 Celsius, compared to 90-95. But when I tried to wake from S3 sleep, it just booted into BIOS (that is, it crashed). So, I guess there is a certain amount of covert instability involved. I wonder what the sweet-spot is here, if there is one at all. I will first evaluate this system on Microsoft Windows and macOS without undervolting for some time, and then optimize once I know the system is otherwise stable and any other malfunction introduced at a later point can be attributed to the negative core voltage offset modification.

Maybe my configuration / conclusion is flawed, so let me know if I have missed something here.