Im talking about package power.
Then please, for the love of all that's good in this world, stop going on about "core efficiency". Package power is only indirectly indicative of core efficiency, and to extract core efficiency from package power you must be able to reliably remove uncore power from package power. Without doing so, there is no way whatsoever of knowing how much power the cores are consuming.
Anandtech hasnt disproven anything, even if they are just checking core instead of package, they havent done so normalized have they?
Normalized for what? Your arbitrary power limits? They're running the chips as configured by Intel, allowing it to boost as high as it wants and the workload demands. And they demonstrated a wide range of power behaviours at these stock settings - in instruction dense POV-Ray, they saw a 71W increase over idle, which they estimate to be a 55-60W increase in core power. On the other hand, in the less instruction dense SPEC workloads they estimated core power at 25-30W. At (at least roughly) the same clocks. At which point it delivered marginally better performance than the 5950X, the cores in which peak at 20.6W in POV-Ray and similar to ADL likely consume a bit less across the SPEC suite.
That demonstrates that, as configured from the factory, at very similar clock speeds, Zen3 is more efficient than ADL as ADL beats it by ~5-16% while consuming notably more than 5-16% more power. Lowering the power limit will not change ADL's efficiency in this test, because the CPU is nowhere near hitting any reasonable power limit - even a 50W limit would likely deliver roughly the same performance in SPEC, and it will boost as opportunistically within this limit unless also frequency limited.
Im just trying to tell you im pretty confident it is the case.
You're so confident that you're heavily emotionally invested in the outcome, yes, I see that. Doesn't change what I said above.
And im confident cause i tested, repeatedly. Ive seen a tuned to the max 5800x score 16k in cbr23 at 150w, while 8 gc cores need.... 65 to match that. Yes cbr is a good scenario for alderlake but the differen is ridiculously big
But, even assuming that the numbers you're constantly citing out of nowhere are accurate, to repeat myself for the n'th time: this comparison is deeply, deeply flawed. Heck, this is
far more problematic than the purportedly "fundamentally flawed" testing you've been complaining about. Why? 'Cause you're comparing one clocked-to-the-rafters, pushed to extremes tuning of one chip, with a heavily power limited, and thus also clock limited, tuning of another. How does a 5800X perform at 65W? How do each of them perform across a range of sensible wattages? How do they perform outside of the one application that you love to cite because it can be run in one click and performs favorably on your platform of choice?
Τhe 12400 is a different die than the rest of the lineup and yes, it is the worst binned alderlake pretty much. The 12900ks is the best bin and should be the most efficient of them all, but havent tested it. According to igorslab though it require 124mv less than the 12900k for same clocks, so yeah, that one will knock efficiency out of the park, we are talking about numbers that zen 5 might not even be able to match
Also the review from TPUP is power from the wall, which is not really indicative. When you are testing that low wattage parts, a 5 or 10w discrepancy from the motherboard makes a huge difference. TPUP uses the maximum hero for the 12400, just the RGB and the actual screen on that motherboard throw the numbers off. You can check igorslab review which tests only CPU power, the 12400 is way more efficient than the 5600x.
It's up to 65% (that's HUGE) more efficient in lightly threaded workloads and around 20-25% more efficient in multithreaded workloads.
Today we want to question where the upcoming Core i5-12400 can still score apart from colorful gaming worlds and whether it will remain just as frugal and efficient. Gaming is actually overrated when…
www.igorslab.de
Intel's stock settings push the 12900k way way past it's efficiency point. They are trying to make it compete with the 5950x in MT performance, which it has no business doing imo. In all fairness, AMD's stock settings - as shown by the zen 4 leaks will also be out of the park, the only reason they didnt push the wattage with zen 3 is they didn't need to, Intel wasnt competing in MT performance with cometlake, so AMD decided to play the efficiency card. Now that Intel is pushing them AMD also raises the stock wattage
There is something rather iffy going on with those Igor's Lab benchmarks - at least in terms of AMD power consumption. He reports his 5950X consuming 217W, which is ... well, crazy. That's the power consumption of a 5950X tuned to the extreme, with zero regard for efficiency, and it is certainly not reflective of anything resembling stock behaviour. If Igor didn't do that manually, then he should throw his AM4 testing motherboard out that window you're talking about and pick one that isn't garbage. A stock 5950X doesn't exceed 144W whatsoever - though if measuring at the EPS12V cable you'd also need to include VRM conversion losses in that sum - but that would be roughly equal across all platforms.
Edit: looking at Igor's test setup, the motherboard is configured with "PBO: Auto". In other words, it's running a motherboard-dependent auto OC. That is
not a stock-v-stock comparison. And that is some pretty bad test methodolgy. This essentially ruins any and all efficiency comparisons based on these numbers, as the motherboard is clearly overriding all power limits and pushing the chips far beyond stock power and voltage.
It's also kind of telling that you're
very set on being as generous as possible with Intel, repeating that the 12400 is "the worst possible bin" of ADL, etc, yet when coming to AMD, you consistently compare against the 5800X -
by far the most power hungry bin of Zen3, by a
massive margin. Remember, it has the same power limits as the 5900X and 5950X, with 50% and 100% more cores respectively, while clocking only a bit higher at base. Again:
look at Anandtech's per-core power draw testing. The 5800X consumes notably
more power per core in an 8-core load than both of those CPUs, while also clocking
lower. So, for Intel, you're ever so generous, while for AMD you're consistently comparing against the worst bins. One might almost say that this is, oh, I don't know, a bit biased?
You're also wrong about your 12400/12900K binning statements - they're not the same die, so they're not comparable bins at all. They're different silicon implementations of the same architecture, and each represents a bin of its implementation. It's entirely possible that the 12400 is a low grade bin of its silicon, but unless you've got detailed clock and power scaling data for several examples of both chips, you can't make comparisons like that.
There's also the complexities of boost algorithms and thermal/power protection systems to take into account, which can throw off simple "more power=faster" assumptions. For example, my 5800X (from testing I did way back when) runs
faster in Cinebench when limited to 110W PPT than if let loose at 142W PPT. And significantly so - about 1000 points. Why? I can't say entirely for sure as I have neither the tools, skills nor time to pin-point this, but if I were to guess I'd say it's down to the higher power limit leading to higher boost power, meaning higher thermals, more leakage, and subsequent lower clocks to make up for this and protect the chip. Zen3 has a quite aggressive chip protection system that constantly monitors power, current, voltage, clock frequency, and more, and adjusts it all on the fly, meaning that tuning is complex and non-linear, and highly dependent on cooling.
