Quite smart idea, not for throttling when the fans are heard, but SPEEDING THEM UP while they are NOT heard, based on the noise floor of the current environment.
This, combined with an assessment how much of the environment the user currently hears (i.e. maybe he is watching a movie), could provide a lot of headroom for additional cooling without bothering the user.
Another trick is not to use constant RPM because you can hit some mechanical resonance point, but constantly varying RPM around target RPM so that the system even if it hits resonance stays there only for a brief moment and won't start vibrating.
They do the same with clock signals in some systems. Similar idea to avoid high peaks at a single frequency:
> Spread spectrum clocking is a technique used in electronics design to intentionally modulate the
ideal position of the clock edge such that the resulting signal’s spectrum is “spread”, around the
ideal frequency of the clock. In timing circuits, this has the advantage of reducing Electromagnetic
Interference (EMI) associated with the fundamental frequency of the signal.
This is controversial, because the same amount of noise is emitted, just with a different shape that is still able to be picked up by radio receivers just as well, but is more likely to pass regulatory tests.
We have a standard for the maximum energy on given frequencies, not the total EM emitted. Spread spectrum clocking definitely reduces the peak energy on specific frequencies, which is usually what you care about more.
I guess it makes you more likely to experience interference, but the interference you'll experience, on average, will be much less severe.
Sure--- I'm familiar with the bandwidth theorem. In general, a single narrow carrier with lots of energy inside your channel is more harmful to demodulators than spreading the interference over 20MHz, most of which isn't even on top of your channel (or, if your channels are reaaaaaallly wide, it's still just a slightly increased noise floor instead of a strong single frequency swamping your signal).
I've designed a lot of radio systems, built a lot of demodulators, and calculated a lot of link budgets. It's best to assume that people know a fair bit, here, instead of tossing out the simplistic dismissal.
I'm also an amateur radio operator, and I know that strong carriers are often far worse for me down in HF than the same energy spread out over tens of thousands of cycles; though I can often notch carriers, I only have so much dynamic range.
For VR perf its probably a bad idea to throttle the performance. A better goal would be to balance user comfort. You can trade noise for thermal comfort.
I can't speak for Apple, but for my Prusa 3D-printer: Worn/cheap ball bearings will make a rattling sound until they heat up and the balls expand. If you care about the auditory experience, using audio in the feedback loop makes some sense.
The goal is probably not to absolutely manage cooling but to focus on the audio experience and try to keep noise below some threshold.
I think that's an interesting idea, even if not everyone might be happy with the tradeoff (comes down to how much you care about noise vs getting throttled), especially for a vr device.
It is a machine you literally have on your head. I have not tried it, but I can believe that it requires some different UX considerations than normal computers. I would definitely not like to hear fan noise vibrating through my head.
There's a reason why the Vision Pro has separate chip that handles the real-time passthrough from cameras (and bunch of other sensor-fusion stuff, AIUI).
Wouldn't it still thermal throttle as the whole device gets hot, separate chip or not ?
I understand the threshold for heating up should be lower than with integrated chips, but we're talking about fan speeds, about when that threshold is still reached and cooling is needed.
In the old days, the device settings would allow the user to tune the trade-off to their own preferences. But current user design orthodoxy is that "settings are bad" to an extreme (I actually agree with the weaker formulation "too many settings are bad"... for some value of "many")
That really depends on what the effects of the throttling are. Stronger foveation or other rendering quality drops? Probably OK. Frame rate dropping below 75? Not in VR, never...
But they're a trillion dollar company because of their relentless focus on the end-user experience which you happen to dislike, but which most people love.
UX isn't NVIDIA's business model. It is Apple's, and they've found there's shittons of money to be made giving a shit about UX, especially when virtually no one else will.
Because the RPM of a fan is not a 100% reliable indicator of its loudness. Sometimes a lower speed can even be louder than a higher one, because of certain resonances...
Yes, that could work (it would need to be done for each fan individually, because of manufacturing variation). But it's still not guaranteed that the fan will keep working the same way in the future, after some use/deterioration or in untested conditions.
Makes sense. Humans are really bad at perceiving the environmental noise floor. Our brains just tune it out, but it is has a huge impact on your perception of the loudness of fans that you've strapped to your face.
I can certainly see things like this causing a lot of unintended consequences. "It speeds up when I play loud music" is going to confuse many who don't know about this "feature".
Twitter doesn’t show threads to people who are not logged in (e.g. if you don’t have an account). Shaming people for not reading what they can’t see (or even know exists) is unfair. Using archival sites doesn’t work as a bypass like on newspaper sites.
Not to mention people on Firefox, which from what I have read on HN might not even have access to it at all.
Twitter links should probably be downranked on HN until (if/when) they return to being more accessible.
With a paywall you know what you’re missing or not. With Twitter that’s harder to determine, which I’d argue is worse. But I wouldn’t say either of our opinions is objectively correct.
> and HN is mostly fine about those.
It is explicitely fine with paywalls which have workarounds.
That is definitely done for bearing in ratting machinery. You can often even pinpoint which bearing is damaged and which part of that bearing is damaged, based on the frequency spectrum.
This, combined with an assessment how much of the environment the user currently hears (i.e. maybe he is watching a movie), could provide a lot of headroom for additional cooling without bothering the user.