Yes of course im talking about n core workloads, in other cases no cpu really pushes consumption that high to be an issue.
Except that Intel's aggressive turbo clocks cause them to draw 50+W per P core while boosting, while Zen3 tops out at ~20W/core. So even while ADL is faster at those clocks, it needs tons of power to get there. That doesn't mean that it
can't run more efficiently - but it would be slower than Zen3 in that case.
I still disagree, no core takes a lot of power. Every core in existence uses as much power as it is given.
That is a gross oversimplification, and essentially entirely untrue. All CPU cores have a base power level required for them to function properly at all. They also have a lower threshold of clock speed and power where they become able to perform efficiently - where clocks and performance rise high enough to overcome the inherent disadvantage of that base power draw. For most modern CPU cores that clock speed is quite low, but it's still a threshold. And, crucially, all cores have some form of clock/power ceiling. For Zen3 that is ~21W, where scaling essentially stops past this. You
can push more power into them, but clocks will stagnate and performance will drop. Exceeding 30W/core for Zen3 is essentially impossible (possibly outside of LN2). For ADL, it's more like 70W, though no stock CPU reaches those levels - but it scales much, much further in terms of clocks and power.
From what I've seen, it seems that Zen3 cores generally scale very well downwards - shown by their use in very low power mobile implementations where base clocks are still decent. Intel P cores seem to have a higher base power requirement, but in desktop CPUs this is offset in low threaded tasks by MCM Zen3's high uncore power (due to through-package IF).
The question is if it performs well enough with that power. We know that pcores are more efficient than ecores at everything when it comes to 5w and above per core. Therefore i really fail to see how a big full pcore cpu will have any problems with power draw.
Do we know that? Also, if that's the crossover point, your "16p > 8p8e @50W" thing seems rather shaky - that would give each of those P cores just 3.125W after all, assuming
zero uncore power (so obviously less than that). That seems like it would be cutting things pretty close in terms of which is more efficient if that's the case.
Also: a big core only CPU would have the potential for problems with power draw as it will tend to boost too aggressively, even for small tasks. Most CPU boost algorithms, particularly Intel's ones, are tuned for race-to-finish efficiency. This is great for mobile tasks with moderate boost power, but falls apart for desktop parts with massive per-core boost power budgets. After all, a P core boosting to 50W would need to be 5x faster than an E core at 10W for it to match its efficiency. And we know a P core doesn't come close to 5x the performance of an E core, even at peak boost. This is one of two scenarios where E cores shine: they're fast enough and plentiful enough to handle potentially disturbing background tasks without consuming too much power. If you only have P cores, are running a high performance but low threaded foreground task, and a background task needs some work, one of two things happen: either you're power limited and the new core boosting will force the foreground cores to clock down, or you aren't power limited and the new core will increase power consumption for however long the background task takes to finish.
Is this a massive problem? No. Does it make an all-P core design less efficient outside of nT tasks? Yes. (And, of course, in nT tasks you run into the question of how many E cores you could have gotten for the same silicon cost within the same power envelope as those P cores, and how that would balance out - but that gets complicated very quickly due to boost powers, cache access, etc.)
