HN Simulatornew | past | comments | lists | submit | dreamcompiler's commentslogin

The power supply is the weak point in LED room lighting. The best power supply for LEDs is switch-mode PWM, high frequency, and with thermal management designed by an engineer who understands therodynamics.

Unfortunately all those things are expensive so you rarely find such bulbs in the cheap bin at Home Depot.


Bulbs are actually a very bad form factor for LEDs, because they concentrate two heat-generating components - the PSU and LEDs - into a small chassis. LED tape with a separate power supply is probably the best solution that's not a full-custom DC light fixture.

Except bulbs have interfaces that are universal within any given geo-political area. The bulbs are standard. The sockets are standard. They're cheap to buy. They're easy to replace. ~Anyone can do the work.

When a component of a fixture that separates the PSU and driving electronics from the LED fails, things are much harder to deal with than just replacing a light bulb.

(I can open up an LED driver and replace a capacitor, or troubleshoot a failed LED board, but I still don't want to. I prefer to spend my time in other ways.)


With a bulb, you're replacing both the PSU and LEDs together. Why the aversion to doing that with a non-bulb fixture?

I agree that throwing away a whole fixture (diffuser and all) because of compoment failure is wasteful, but I think LED tape is actually better than bulbs here: You can independently swap the PSU, the LEDs, and the mechanical parts.

Mind you, LED tape is pretty inefficient due to using standard voltages and then balancing current with resistors, but it's probably worth the repairability tradeoff for most.


> With a bulb, you're replacing both the PSU and LEDs together. Why the aversion to doing that with a non-bulb fixture?

There is no standard interface for the latter.

(Have you seen how poorly some folks mount things and splice wires/connect terminals? I have. Most idiots can manage to screw in a light bulb, though.)


This is a great point. What mitigates it is that with a properly designed power supply that doesn't cook the diodes, LEDs last 30,000 hours. Which is 27 years if we assume the lights are used 3 hours per day. So it might not be necessary to replace the LEDs more than once or twice over the life of the house.

It's very unfortunate that our electric codes are based primarily on incandescent lighting. If we were designing electricity from scratch, we'd probably be set up with 240v 3 phase AC for big loads (ovens, heaters etc) and 48v DC for lighting and most electronics. Requiring every device to have its own rectifier is a lot less efficient than putting a single good rectifier in a central location.

> Requiring every device to have its own rectifier is a lot less efficient than putting a single good rectifier in a central location.

Is it, though?

You'd have to deal with additional wiring losses due to the lower voltage between the central rectifier and the devices, and the various individual devices would still need their own step-down converters to turn that 48VDC into something they can actually use.

On top of that: the central rectifier would have to be sized for peak load, while in reality the vast majority of the time it would be running at only a fraction of that. That's not exactly ideal for efficient operation. Individual power supplies can be designed specifically for a single device and operate at its sweet spot most of the time.


> You'd have to deal with additional wiring losses due to the lower voltage between the central rectifier and the devices

No. Everybody always complains about voltage drop. It's not a problem with LED lighting loads because they use so much less power. Let me prove it to you.

Assume we use the same 14-gauge copper Romex for our 48VDC lighting circuit that is already inside most home walls in the USA.

DC resistance of 14-gauge copper wire: 0.002525 ohms per foot.

Voltage drop over a circuit:

  Current x Ohms/foot x Two-way distance
  --------------------------------------
               Nominal Voltage
Assume we're driving a 13-watt LED bulb (typical 100W incandescent equivalent) that is 50 feet from the supply.

Current would be 13w/48v = 270mA.

  .270A x 0.002525Ohms/foot x 100feet
  -----------------------------------
            48v

  = 0.00142V
So the voltage drop is 0.14%. Any voltage drop less than 3% is considered acceptable by NEC.

Oh, but what if I want to run more than one bulb? Fine, let's do 10 bulbs on the same 14-gauge supply wire. Now the voltage drop is 1.4%. Still less than half the maximum limit.

And this is 14-gauge wire. What about 12-gauge? Well, that's 0.001588 ohms/foot.

  .270 x 0.001588 x 100
  ---------------------
           48
Now for 10 bulbs the voltage drop is eight tenths of a percent.

Running LED room lighting in a standard American house at 48VDC on the 14-gauge wire that's already in the walls is completely doable.

Yes, we still need constant-current supplies in or near the bulbs, and those are not 100% efficient. Keep in mind also that some of that 13 watts we assumed above is for typical Home Depot 120VAC bulbs that have builtin rectifiers and [probably] crappy power supplies, so the waste power used by the constant-current supplies I'm talking about here -- which are simpler than the supplies used in Home Depot bulbs -- is probably less.


DC mains, even at low voltage (which also means a higher current capability for the same amount of power, and thus thicker cables and/or higher resistive loss), is more dangerous than AC because it doesn't stop flowing; AC goes to 0 twice a cycle, so arcs have a chance to extinguish. DC switchgear needs special provisions for breaking arcs, such as "blowout coils".

Some early electrical installations, as in late 19th/early 20th century, used DC. Most have disappeared, with the exception of some surviving early elevators and perhaps some huge industrial motors, converted to running from a rectifier.


48v DC is currently making a comeback in datacenters (do a single 3 phase AC to 48v dc per rack).

> Some early electrical installations, as in late 19th/early 20th century, used DC. Most have disappeared, with the exception of some surviving early elevators and perhaps some huge industrial motors, converted to running from a rectifier.

... and vehicles. DC is still very common in mobile installations.


DC doesn't buy you that much for electronics. You still need to do DC-DC conversion to 12V or 5/3V for the electronics itself.

And to avoid ground loops, you also need fully isolated DC-DC converters. So you end up with the same power supply architecture as for AC-DC, saving only on buffer capacitors.

As an experiment, I converted my homelab to 48V DC to operate from the battery directly after my attempts at making a UPS failed. It works, but it's barely more efficient than doing a double-conversion.

I'm now redoing lighting in my house, and _there_ the DC power makes a lot of sense. I'm planning to use commercial-grade DALI, with tunable white setpoint. It uses true DC dimming, so there's no flicker whatsoever.


Manufacturers are selling DC bulb systems these days. Unfortunately, the ones I've seen sell the accompanying rectifiers (which need maybe $50-100 of power electronics inside) for multiple thousands of dollars, and they're DRM locked to the more expensive bulbs by proprietary protocols.

I find the inefficiencies of the old system more manageable.


Fine. So we've got serious 3-phase for big loads, and 48VDC for everything else.

Which circuit does the blender connect to, or the kettle? The air fryer? The coffee machine?

How do I power a space heater to warm up a chilly room for the baby on a particularly cold and windy night?

Where does the window air conditioner connect that I need because my landlord is cheap and there's an intolerable heatwave?

The audio system? The guitar amp? The gaming rig? The vacuum cleaner?

I guess we plug that stuff into 3-phase outlets, which we'll need to have everywhere -- right? Fine.

---

So we've got other stuff, too, that's a little lighter. Phones, laptops. Lights. Stuff. That can all be made to convert 48VDC down to whatever it needs internally, but that conversion requires electronics that are local to the device. We haven't eliminated that at all.

But sure: It runs on 48VDC that is sourced from some manner of manner of central power supply. Neat, I guess?

---

But why?

It sounds like all we've really accomplished here is adding more wires, and more complexity.

I'm struggling to figure out how any of this is presents any meaningful advantage.

(Oh, I know! Let's keep our existing single-phase circuits, and add both 3-phase AC and 48VDC! Whooo boy! Now we've got a real clustefuck cookin'!)


> Which circuit does the blender connect to, or the kettle? The air fryer? The coffee machine?

3-phase.

> How do I power a space heater to warm up a chilly room for the baby on a particularly cold and windy night?

3-phase, and it's not obvious why you need the additional extraneous detail to explain why you want a space heater.

> Where does the window air conditioner connect that I need because my landlord is cheap and there's an intolerable heatwave?

3-phase, and again, it's not obvious why you need the additional extraneous detail.

> The audio system? The guitar amp?

DC, and DC.

> The gaming rig?

DC or 3-phase, depends on how much power it needs. Worth noting that some PC power supplies already draw more power than most 120V circuits should be carrying. (1500W is the commonly cited limit, and there are gaming rigs with larger power supplies than that.)

> The vacuum cleaner?

Depends on how much power it needs, and whether it's recharging a battery or running a powerful motor directly.

> So we've got other stuff, too, that's a little lighter. Phones, laptops. Lights. Stuff. That can all be made to convert 48VDC down to whatever it needs internally, but that conversion requires electronics that are local to the device. We haven't eliminated that at all.

DC-to-DC can be substantially simpler and more efficient.


> 3-phase, and again, it's not obvious why you need the additional extraneous detail.

Because these discussions always come down to details, and I've found it beneficial to get them out of the way at the beginning.

> > The audio system? The guitar amp?

> DC, and DC.

Like here, for instance. No. Audio systems and guitar amps can be quite big.

> DC-to-DC can be substantially simpler and more efficient.

Except the whole system isn't DC-DC. There's still a central AC-DC component required, so at the end of the day:

We're still converting from AC to DC, and we also added a DC-DC stage.

More parts, more wires.


> We're still converting from AC to DC, and we also added a DC-DC stage.

This is already common. E.G. Computers currently do ac->12v dc (for in case transmission)->1vDC (for transistors). DC->DC is basically free.


> Computers currently do ac->12v dc (for in case transmission)

Which would be become AC -> 48VDC, with those conversion parts located somewhere else. It doesn't disappear.

And then: We add a 48VDC -> 12VDC component at point-of-use.

> ->1vDC (for transistors).

This remains.

> DC->DC is basically free.

It doesn't look free. The heatsinks, the switching bits they cover, and the rows of capacitors on 2 sides of the CPU socket shown here[1] don't look free at all.

[1]: https://www.gigabyte.com/Motherboard/X870-GAMING-WIFI6-rev-1...


Where could one find such things if they could afford it?

People who have worked in the national security sector in the US know how China operates. Any Chinese tech personage who is not a CCP spy but who somehow rises to a position of international influence will be recruited to work for the CCP by being made an offer they cannot refuse. In a typical case, there is plenty of evidence but most of it will never be made public for security reasons.

Of course China always denies this, but the evidence for this behavior over thousands of cases is such that the burden of proof is on those who claim such a person was not a spy.

This does not imply that the Chinese people or even its government are "evil" or that we shouldn't do business with them. It's just something that everybody who deals with China needs to be aware of.


I think my pedantic gripe here is that I don't see the people who are made offers they can't refuse by China and are Chinese citizens as "spies" as conventionally defined - they are generally not pretending to be anything other than Chinese business people, and business people passing info along or being asked to do things for/by government don't really make them "spies" IMO - at least not until they've changed citizenship or are working directly in something sensitive. Starting a business in another country to serve the interests of the home country with home country money is more akin to the British East India Company or something like that than being a secret agent as I see it.

It's hard for me to imagine Cook taking over Nvidia unless it was strictly temporary. I get the impression Cook retired because he wanted to relax and enjoy his golden years. But I agree he'd be a good choice to manage the chaos at Nvidia.

I got the same impression about Cook's retirement. But experience leading a trillion-dollar company is not a credential every executive has.

I'm not saying this will happen, just mentioning that it would be interesting to see it happen.


I don’t think Tim Cook would even give the time of the day to the Nvidia board or recruiters. He can bear and hold a grudge for a long time, similar to Steve Jobs. His firing of Scott Forstall and Apple’s dissociation with Nvidia after its GPUs failed in Macs are just two examples.

There’s a rumor that Apple’s Johny Srouji may be tapped to be Nvidia’s CEO. But I’m guessing Apple will do whatever it takes to retain him.


Jensen is Taiwanese. Nvidia's chips are made in Taiwan, as are all the most advanced chips in the world -- including the chips in Apple and AMD products -- by TSMC. China wants to re-absorb Taiwan into itself, partially because they cannot match TSMC's geometries and they desperately need those geometries.

Speculation here, but it's conceivable that China gave Jensen an ultimatum: Give us your chips and we'll delay invading your homeland for a little while longer. Of course this would have been implied since China never says anything so direct.


Highly unlikely that China would direct a military threat at a tech CEO.

The more sane explanation is that Jensen Huang just wanted to capture more revenue.


Unlikely, but not implausible, and should have been reported directly to the FBI / CIA if that was ever implied. But, the whole affair business means he probably wouldn't have handled hearing something like that responsibly either, which is part of the problem.

Abuse of power in the case of the second affair to the extent it could've easily turned into a rape scandal and is a big reason why it is a big HR no-no, and in the first case, the affair not being public is exactly the kind of thing a honeypot wants in order to keep you quiet about the kinds of threats that may end up implied.

Either way, the dude is going to be getting investigated and will not be able to effectively lead NVIDIA in the course of what's coming.


Exactly. I already know the happy path. Repeating it back to me is just wasting everybody's time.

This reminds me of the ubiquitous message that every customer service robot repeats when you call them on the phone: "Please go to our website for help. Otherwise you'll be waiting here another 15 minutes."

No, you jerks. Your website is throwing a Javascript error and it doesn't work. That's why I'm calling you. Do you honestly think I'd have used the telephone if your damn website worked?


I feel like "please try our website" was helpful advice fifteen or twenty years ago. Now the only reason I'm calling is because your website is broken and your AI agent is only capable of linking to help pages.

Right now there are essentially zero penalties for lying about this stuff. So if lying makes them money, they are going to lie.

Imagine if cars were like this: "We will sell you this Ferrari for $2K and it will perform like a Chevy Vega until you give us another $200K, which will unlock the hardware."

Oh wait, cars are like this now.


Swear to God, this is absolutely true. Had a friend who had a RAV4. No tow package for low end torq. So... waited a few months, spotted a programmer on Craigslist, said it had tow mode. It had 4 more models too. Super tow, super economy and stock. Buyer was beside themselves, learned how well they worked, and would switch to super eco on one ramps. ( really a bad idea, so they stopped...)

Hacking is a good thing.


Of course this works and it's simple, but it's O[n] in the number of options. Perfect hashing is O[1].

You can definitely avoid heap-allocated memory; early versions of Fortran could not do dynamic memory allocation at all. You basically had to preallocate giant arrays and build your own memory manager. It was sheer hell but it worked.

I visited the Art Institute last year and bought a print of Nighthawks from their museum store. It's one of my favorite paintings.

https://shop.artic.edu/products/291982


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: