Thank you for posting this here directly — this is exactly the mechanism-level explanation I've been missing. The point about the engagement condition being "unreachable by construction" under the current regulation puts into precise words what I could only describe empirically (system_profiler showing No, sustained scores stuck at Automatic-equivalent levels, no combination of testing changing that). And referencing FB23754032 with 10+ similar reports attached should make it a lot harder to treat this as environment-specific. Really appreciate you bringing this level of instrumentation to the thread.

To rule this out as thoroughly as I can: I've tested with the official 96W MagSafe charger (Apple's MagSafe cable, not USB-C), the 96W USB-C Power Adapter via USB-C, and a 140W-rated Thunderbolt dock via USB-C. All three give the same result — system_profiler reports the same wattage (~98W) in all cases, and sustained power/score during the benchmark is the same (~27-30W / ~3070-3430) regardless of which one is connected. I've also swapped wall outlets/circuits during this testing, and confirmed other high-draw devices (other laptops, including ones that pull 100W+) charge normally on the same outlets and cables, so I don't think the wall or circuit is the limiting factor either. I did this deliberately because I'd already considered the "something in the power delivery chain is bottlenecking it" possibility early on — it was actually the first alternative explanation I tried to rule out, before even getting to the thermal/firmware investigation. At this point I've cycled through charger, cable, and outlet combinations enough times that I'm fairly confident the limiting factor isn't anywhere in the power delivery chain. The constraint shows up the same way regardless of what's supplying the power.

One more data point to add to my last reply, since I think it makes the comparison clearer. Notebookcheck's review of this exact model gives a specific number for the High Power mode stress test: it settles at 42W (~85°C). I've also seen other YouTube reviews of this model reporting sustained temperatures around 80°C under similar sustained load, which is roughly in the same range. My unit, with High Power selected, settles at ~27-30W and 65-73°C instead. That's a meaningfully lower power point at a meaningfully lower temperature than what's been independently reported elsewhere for the same model, same benchmark, same mode. So to put the question more precisely than I did before: I'm not asking why the machine doesn't run hotter than necessary. I'm asking why my unit's equilibrium point in High Power mode doesn't match the documented equilibrium point Notebookcheck reported for this exact model in the same mode.

I follow the thermal model you're describing, and I'm not arguing the machine should be allowed to run at a sustained 95-101°C — I agree that would be unreasonable, and I wasn't asking for that. The core issue is simpler than the thermal ceiling discussion: Notebookcheck measured ~44W sustained Combined Power on this exact model at launch, with reported temperatures settling in this kind of range (other reviews I've checked report sustained temps in the 78-82°C range for the 14" Max chassis under similar sustained load). My unit settles at 65-73°C and ~27-30W under the same benchmark. I'm not asking why the machine doesn't run at 95-101°C. I'm asking why it settles at 65-73°C instead of something like 80°C — a temperature point that's still well within the range other reviewers report as normal sustained operation for this chassis, and that would correspond to meaningfully higher sustained power. Your fan/airflow analogy explains why temperature tracks fan speed under changing ambient conditions, but ambient conditions aren't what's different here — what's different is that my unit converges to a lower temperature/power point than other units of the same model reportedly do, under the same workload, same chassis, same cooling design. If the thermal model is working as intended, I'd expect my unit to converge to roughly the same temperature/power point as the launch review units, not a meaningfully lower one. That gap (44W → 27-30W, ~80°C → 65-73°C) is what I don't have an explanation for yet.

To answer directly: this is the native macOS app , not the iOS app running in compatibility mode. I have never used the iOS version for any of this testing. The result files I've referenced throughout (.3dmark-result) are macOS-native output. So the entire "window benchmark / compatibility mode / iOS app can't access full hardware" line of reasoning doesn't apply to my testing — that's a different scenario than what I've been describing. On the thermal framework you laid out: I follow the logic, but it still doesn't explain my data. By your own model, High Power Mode raises the fan speed ceiling, which raises the thermal ceiling (#3), which only matters if the machine is approaching that ceiling. My system does approach it — GPU briefly hits 95-101°C early in the run. After that, it settles at 65-73°C with fans still clearly at high RPM (not idling down), for the rest of a 20-loop test. If "Automatic" mode's fan cap was insufficient to dissipate heat at a higher power level, I'd expect to see temps climb back toward the ceiling as the workload continues — not settle comfortably 25-35°C below where they peaked, with the fans still working hard. The fact that temps drop and stay down while fans stay up suggests the system has more thermal headroom than it's using, not that it's constrained by Automatic's fan-speed limit. On the benchmark variance point — I understand run-to-run variance is real and that NotebookCheck's rigorous methodology and mine won't produce identical numbers. But the gap I'm describing isn't "3600 vs 3550." It's 44W sustained power vs. 27-30W — a ~35-40% reduction in steady-state power draw on the same chip, same test, same adapter wattage. I'm not asking why my score isn't identical to theirs; I'm asking why my sustained power ceiling is that much lower when my thermals indicate there's room to spare. Two more pieces of data since my last post: The High Power Mode: No status isn't just something I'm reading off system_profiler interactively — it's also what's recorded in Apple's own sysdiagnose output (SystemProfiler/SPPowerDataType.spx), under both AC Power and Battery Power. So this isn't a one-off terminal query; it's the same value captured by Apple's own diagnostic collection tool. To rule out Migration Assistant history as a factor (since my original post mentioned the M2 Air → M4 Max → M5 Max migration chain), I did a full erase of my M4 Max via iBoot Recovery and a completely clean macOS install — no Migration Assistant, no data restore, Apple ID sign-in only. Selecting "High Power" on this fresh install still resulted in High Power Mode: No. This rules out migrated preference data as the cause entirely; the behavior is present on a machine with zero migration history.

Thanks for clarifying the "recommended = required" distinction — that's genuinely useful context I wasn't aware of, and I appreciate you flagging it. That said, I've already tested both scenarios and the behavior is the same: Apple's own 96W USB-C Power Adapter (system reports "Wattage (W): 98" via system_profiler SPPowerDataType — matching the rated wattage) A Thunderbolt dock rated at 140W (system reports the same 96W negotiated input) In both cases: fans at high RPM, GPU/CPU temps peaking at 95-101°C in the early loops of the Steel Nomad Stress Test before settling at a sustained 65-73°C, and sustained Combined Power (powermetrics) staying flat in the 27-30W range throughout the stable phase. If "96W is required," and my 96W Apple adapter is delivering exactly that, the adapter side of the equation should be satisfied. Yet system_profiler SPPowerDataType consistently reports the following regardless of adapter or load, even during active benchmarking with fans at full speed: AC Charger Information: Connected: Yes Wattage (W): 98 AC Power: High Power Mode: No Low Power Mode: Yes High Power Mode: No has persisted across multiple days, reboots, and both power sources tested. This is not a transient state — it appears to be a stable, persistent condition independent of adapter wattage. I'm also curious how this explanation interacts with the launch-era Notebookcheck review, which reported ~44W sustained Combined Power and a stabilized Stress Test score of ~3600+ on this exact model. Were those results achieved with the 96W adapter? If so, what changed between then and now that would explain the ~27-30W I'm seeing with an equivalent power source? I don't want to be dismissive of the adapter angle — it's a concrete and testable factor. But given that my adapter is already delivering 98W (per system_profiler), and the gap to launch-era results persists regardless of power source, I suspect there's still something else going on beyond just adapter wattage.

Thank you for taking the time to dig into this and for the detailed explanation — I appreciate the effort, especially running your own test rig with concurrent workloads. That said, I don't think the "insufficient heat" explanation accounts for what I'm actually observing, and I'd like to push back on a couple of points with data rather than supposition. 1. Sustained temps after thermal settling are well within headroom — not "too cool to trigger," and not "too hot either" This isn't something I'm inferring — I have screenshots and powermetrics/Stats.app readings throughout the Steel Nomad Stress Test. Early in the run, GPU clusters did spike to 100-101°C during the initial burst. But after the first thermal throttling event, the system settles into a stable state around 65-73°C (GPU and CPU performance cores alike) for the remainder of the 20-loop test — and it stays there, fans still clearly audible at a sustained high RPM, not idling down. That stable 65-73°C range is the part that doesn't fit the "not enough heat" explanation. It's well below thermal limits (nowhere near the 100°C+ seen briefly at the start), there's clearly still headroom, and the fans are still working hard to hold it there — yet powermetrics Combined Power stays flat in the 27-30W range for the entire stable period. If heat were the gating factor, I'd expect to see power increase to use the available thermal headroom, not plateau well below it while the fans keep working. In other words, the machine isn't "too cool to enter High Power Mode" (fans are maxed, it clearly already passed through a hot phase) — it's stabilizing itself at a power/temp point that's far short of both the historical 44W figure and the 100°C limit, despite available cooling capacity. 2. The "synthetic benchmarks don't generate enough heat/IO" theory doesn't hold up against third-party data Notebookcheck's launch review of this exact model used the same benchmark (3DMark Steel Nomad Stress Test) and reported a sustained Combined Power of ~44W and a stabilized score of ~3600+. That's the same synthetic, GPU/CPU-bound, low-storage-IO workload you're describing as insufficient to trigger High Power Mode. If your theory (heat from storage I/O is the real trigger) were correct, Notebookcheck's environment running the identical test should also have failed to reach High Power Mode — yet they measured a sustained power level I cannot reach even with fans maxed out and GPU temps above 95°C. 3. The actual question So the core issue isn't "the machine won't get hot enough to engage High Power Mode" — by the metrics available to me, it clearly passes through a hot phase and then settles into a stable, moderate temperature (65-73°C) with fans still running hard, well short of the thermal ceiling. The question is why a machine sitting comfortably below its thermal limit, with active cooling clearly still engaged, sustains only ~27–30W instead of the ~44W reported for the same model/test at launch. Separately, I also confirmed using a different title (NTE) that switching the Power Mode setting does change total system power (Automatic ~37–40W vs. High Power ~55–57W), so the setting clearly does something — it's just unclear whether what "High Power" maps to today is the same power ceiling that existed at launch. I'm not trying to argue this is necessarily a bug rather than an intentional change (e.g., a response to the battery drain issues some reviewers flagged at launch) — but "the workload isn't generating enough heat" doesn't seem to fit the thermal data I'm seeing on my end. Happy to share the raw powermetrics logs and screenshots if that would help narrow this down further.

New observation: the "High Power" power state itself may have been changed/removed Since my original post, I ran further tests and want to share a new data point that may reframe this issue. Testing with another title (NTE), I observed a clear difference in total system power depending on the selected Power Mode: Automatic: Total ~37–40W High Power (selected via System Settings): Total ~55–57W So the Power Mode selector does change something — switching to "High Power" measurably increases total system power draw in that workload. This means the setting itself is not entirely inert. However, this is where it gets interesting. Comparing against third-party reviews (Notebookcheck) of this same model at launch: Reviewed sustained power under Steel Nomad Stress Test: ~44W (powermetrics-equivalent Combined Power) My sustained power under the same test, same conditions (fans pre-spun, GPU >95°C): ~27–30W And on score: Reviewed Steel Nomad Stress Test stabilized score: ~3600+ My stabilized score across 4 separate runs on the same day: 3142–3432 My current working theory is that the "High Power" flag/state itself is functioning correctly as a switch, but the power ceiling it switches to may have been lowered in a firmware/OS update since launch — possibly in response to the battery drain concerns raised in early reviews (Notebookcheck specifically noted High Power Mode could draw enough that the 97W input limit couldn't keep up, causing the battery to discharge under load). One more data point that points in this direction: after the initial thermal throttling event (brief GPU spike to 100-101°C early in the run), the system settles into a stable state around 65-73°C for the rest of the 20-loop test, with fans still clearly running at high RPM. That's well short of the thermal ceiling, and there's visibly still cooling headroom being used — yet sustained power stays flat at 27-30W rather than climbing back up to use the available thermal margin. This makes me suspect the thermal/power policy itself became more conservative at some point after launch — i.e., the system is now choosing to settle at a lower power/temp point even though it has more thermal room than that to work with, rather than the previous behavior of pushing closer to the actual thermal limit. If that's the case, what is currently labeled "High Power" may no longer correspond to the same power ceiling that reviewers measured at launch. The flag isn't broken — the table/policy it points to may have changed, and possibly become more conservative across the board (not just for High Power Mode specifically). I don't have a way to verify this from the OS side, but the magnitude of the gap (44W → 27–30W, ~3600 → ~3150-3400 stabilized score) seems too large to attribute to environmental variance alone, especially given that fans were confirmed to be at full speed and GPU temps exceeded 95°C in my tests. Curious if anyone else who reviewed/benchmarked this model at launch vs. now has noticed a similar drift.