Extensive workstation and AI test with Intel’s i9-13900K – The power profiles Baseline, Performance, Extreme and Insane from reasonable to pointless in practice

1 - Introduction, settings and test system 2 - Easy finger exercise: AutoCAD 2023 3 - A little more power: Inventor Pro 2021 4 - Next level: PTC Creo 9 5 - Heavy duty: AI and rendering 6 - Dry food: Scientific calculations and more 7 - Power consumption, load peaks and conclusion

After we published Aris’ article about the power consumption of the Core i9-13900K in Cinebench 23 on Monday, I’m following up today with a larger program scope. And yes, I’m deliberately using a Z690 board because I also used it for the launch of Raptor Lake S and the subsequent refresh. That makes the comparison much easier. And as I already teased on Sunday: Intel’s half-hearted revelation with the new profiles for the Z790 boards also pissed me off a bit. When I explicitly asked an employee responsible for motherboards at a major Taiwanese manufacturer, all I got was a bored As of now, we only have “Intel Default Settings” for some Z790 models. When I asked why the whole thing was still dragging on, the answer was a terse Intel doesn’t have an official information about the root cause yet. Aha, no cause, no updates. So the rest is just cosmetic. Well…

Lately, people have been writing about Intel’s possible degradation problems with the high-end CPUs of the 13th and 14th generation and I’m sick and tired of all these discussions. Since I’m not interested in the rather useless gaming crown anyway, because I play almost exclusively in Ultra HD, where the CPU is of secondary importance, I’ve chosen the area where the mainstream doesn’t test: the workstation and AI area. And because some tests run for a very, very long time and involve a lot of load changes, stability is all the more important here. I also very reluctantly remember the Core i7-14700K from the launch article, which didn’t even run stably without overpowering. That was practically degradation out of the box. I have no idea where the part ended up after I returned it.

Performance profiles

I’m going to be mean today and test two profiles that Intel no longer wants to see. So there are now four. The so-called Baseline profile with 188 watts, which was still the standard in the first slides before the Raptor Lake launch, the so-called Performance profile (now “Intel Recommended”), the Extreme profile for the heating technicians among us and the completely stupid “Insane” profile of the motherboard suppliers with the no less insane 4096 watts as a placeholder for the infinite expanses of overclocking space.

And I can already spoil the fact that only two profiles are actually suitable for practical use. The move that Intel would now prefer to hide the baseline profile as “n/a” is probably also due to the fact that not every older Core i9-13900K still runs stably with it after all the energetic kicks in recent months. I deliberately didn’t use Intel’s Golden Sample for today’s test, but rather my own purchased retail model, whose average clock rates were somewhat lower in practice.

Intel therefore recommends implementing the highest power supply profile that is compatible with each individual motherboard design, not entirely without ulterior motives. What a nonsense, because I still pay for my electricity. As you’ll see from my test results, the baseline profile does affect performance from time to time, but I wouldn’t call it unusable. As long as the CPU runs stably with it, it’s a real alternative. Just not for blind marketing. I took the remaining settings from Intel’s slide, unless they were already set by default. And since the Z690 Unify used was not doped with pointless settings, customers should still be on the safe side in many respects. Things like CEP and eTVB limit are of course set as Intel would like them to be.

Test system and methodology

Let’s take a brief look at the test system, which offers no secrets. The power consumption measurement is carried out as usual using a high-resolution oscilloscope and the measuring clamps, but this time deliberately on the 8-pin EPS. I use this to measure BEFORE the voltage converters, whose maximum losses of 8 percent we still have to include here. However, I do this deliberately in order to remain comparable with Aris and also to record the actual load on the power supply unit.

Although I used a powerful cooling solution to test how high the Intel CPU can run when all restrictions are lifted, I deliberately refrained from using a custom loop solution with the IKA lab chiller for reasons of practical relevance. This should also cover over 90% of configurations. I have also dispensed with air cooling for the energetic Nimmersatt, as I don’t want to organize a BBQ.