More bad caps

Well, resonance is always an "everything" problem, right? You take a bandpass filter tuned for 50 ohms, try using it at 25 or 100 ohms and you'll be sorely disappointed -- it'll be too flat, too peaky, lossy, who knows.

Same is true of power supply networks. Suppose you have a wad of capacitance at either end of a relatively simple trace. The trace has inductance (we assume), so you get a C-L-C loop. Which looks like caps in series, so their values reduce in series, and ESRs add.

Conventional advice, dotting 0.1 caps around every single chip, can absolutely work against you -- suppose you have a two layer logic board (old tech, good enough example) with chips spaced every inch or so, cap per chip. The trace length is an inch, and maybe the spacing is the thickness of the board because it's been layed out responsibly with ground pour. The power traces then have a characteristic impedance around 50-100 ohms. Against any capacitor under 10 ohms, say, and at frequencies much lower than the 1/4 wavelength frequency (i.e., anything in the > 0.1uF and ESR caps on the bottom of the motherboard.

Both laptops I've taken apart (I don't go through laptops very often, it seems) did it this way; PC motherboards not so much, but I think they're picking up on it now, too. They also use a generous helping of polymer and/or tantalum though; if you mean the Macs use MLCCs almost exclusively, that's probably fairly impressive.

I thought the big loads were all sucking from 12 these days? What with the 2x2 supplementary connector, and graphics monsters. But geez, this is like 2001 stuff. Both have been around at least as long, and in wide use at least half as long. Maybe things have wound back around again?

I'd kind of like to see everything go 24V. Laptops kind of do; they typically range from ~14V (on battery) to 19V or whatever the charger is. All the internal supplies are derived from that rail, usually using synchronous bucks (which is interesting actually, in that any given supply could back-charge the battery if driven externally). I counted exactly five inductors (or sets thereof) in my 2006 laptop, which were most likely: battery charge, +5V, +3.3V, +1.8V (or whatever the memory runs at), and Vcore (1-1.2V?). Vcore of course is actually a multiphase sync buck, so it has more than one inductor, but you get the idea.

Anyway, point being, you get so much more bang for your buck at 24, and although it is a pretty huge ratio down to 1.2V, it's clearly not insurmountable. A 1kW 24V PSU is a lot easier to make, and requires fewer output wires. Everything already has onboard converters where it matters (CPU, RAM, PCI(e) often enough, GPU, etc.), and drives could be wired for the different voltage (more turns on the motor -- assuming they're wound and not printed), or converters thrown on for not much extra.

Kind of like comparing 120V to 240V, or 240V to 480V. Sure, you can suck down 50kW at 240V, given a good and beefy circuit (and hopefully, three phase besides!), but it's *heavy*. 50kW at 480 (3ph) is almost a walk in the park. You still need beefy wire and diodes and transistors and all, but the sheer amount of power you can muster is pretty awesome. Not to mention the arc flash hazard. :)

Not to say server farms should migrate to 480, or something like that... though I suppose if they're interested, and as long as they can manage the safety features, they're welcome to it. Aren't DC buses common in blade servers, or something like that? Between the arc stability of DC and the fireballing potential of 480, it's probably something of a wash, safety wise... so if they're already doing that...

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

Just about every desktop MB uses +12VDC for the CPU core voltage generation. Same thing for high-end graphics and 3.5" disk drives as well.

Oops, I was wrong and forgot about the change in power supplies. Thanks for the correction and my appologies for the bad info.

Back in days of PIII and Athelon motherboards (about 2000), the problem was that the 20 pin ATX power supply connector had only one

+12V wire, which was inadequate to run a high current drain CPU. These were easily recognized by the overheated and burned Molex nylon connectors.

At the time, there were a few motherboards that powered almost everything from the +5V lines. This was also when I did some reverse engineering on motherboards and came to the erronious conclusion that most everything ran on +5v. A seperate 4 pin connector was added to improve the +12v current capacity and the spec was changed from ATX to ATX12V. This eventually morphed into the current 24 pin connector arrangement, which added one wire each for +12V, +5V, +3.3V, and ground, but retained the ATX12V1 4 pin connector. "I never met a standard that I couldn't improve." (former employer).

As for floppies, CD's, and HD's, some run on +5V only, while others require +12V. For example, a 3.5" SATA drive needs both +5v and +12V power, while a 2.5" laptop drive needs only +5V. The easy way to tell is that if there's a USB powered external peripheral for the device, it probably only needs +5V.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

I get the impression that the 2 +12 V, 2 ground connector is mostly for the CPU. Graphics cards started with, I think, a 3 +12 V and 3 ground connector to get extra juice, and now there are several different kinds of extra power connectors for graphics cards.

There's a lot of inertia working against you, though. The Molex connector used (these days) on 3.5" hard drives, and on optical drives, is the very same one that IBM specified back in 1981.

I also wonder if the inductors required to step the +24 V down to +5 V or less at each peripheral would make some components, like SSDs, "too thick". I suspect you might end up with "bricks" that lived between the main power supply and some of the peripherals... sort of like the IDE-to-SATA adapters that you used to use when SATA was brand new.

As a customer, I'd be kind of mad if I suddenly couldn't use any of my old power supplies anymore. Of course, if I ran the power supply factory, I'd be ecstatic.

When I lived in Oklahoma, Google built a data center there. There was a state law that every business that used more than a certain amount of power had to file an annual report with the state, saying how much they used. (I don't know what the original motivation for this law was.) Google lobbied the state legislature to change this law. They claimed that the power and cooling requirements for servers were standard enough that if you knew the total power used by a data center, you could figure out about how many servers it contained, and they considered that propietary information.

I'm not sure about blades. I know that a few years ago, there was a little bit of noise about server farms having "regular" rackmount servers run from 48 V DC, with a telecom-type DC power plant installed in the building. The claim was that doing the AC-DC conversion in one "big" power supply was a few percent more efficient than doing it in lots of "little" power supplies. I'm not sure if it went anywhere, though.

It's not hard to buy regular ATX power supplies that take 48 V DC; last time I looked, Newegg had a few. If you dig around a little, you can also find 1U PC power supplies that take it.

I worked on a project a couple of years ago that had to be installed in "telecom" locations, and originally the customer thought they would have access to the existing DC bus. They specified 48 V DC input on everything, and all the hardware that I worked on was designed for that. Then the customer found out that they would only have commercial AC power, so they added in their own 48 V power supply and backup battery system.

I went to site for a couple of the installations and I noticed that they included a couple of small rackmount 48 V DC to 120 V AC inverters. I asked the install tech about that and he laughed, pointed to two routers, and said that they were the only things in the rack that didn't run on 48 V. These routers were from an extremely well known company, and I am pretty sure they could have been fitted with 48 V power supplies.

Matt Roberds

Yeah, very same idea with the automotive industry; main difference being, big OEMs can bully their way around anyway. So you get weird oddballs like 28V being thrown around, or whatever they're doing.

It's probably a lot harder for PC OEMs to do that kind of thing; the more oddball ones like Apple could probably get away with it, but I guess usually don't go for it anyway.

The size isn't much bigger. The reactive current isn't any worse for a given design (CCM or whatever % ripple), and the inductance or losses are only a little higher (RMS voltage is higher). If wide, flat inductors aren't available, they can always put two or more in series or parallel.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

and:

"The nice thing about standards is that there's so many to choose from!"

Dave

Cars are 12 V battery, ~14 V charge. Heavy-duty trucks and big buses (and light aircraft) are 24 V battery, ~28 V charge. The automotive industry was kicking around the idea of 36 V battery, ~42 V charge, with the usual ideas of thinner wires, possibly lighter battery, possibly easier to beef up the starter motor for things like idle stop. Germany also thought there'd be a DC-DC converter to support a 12/14 V system as well. I don't think this went very far, though... I think Toyota tried it on a Japan-only model, but they never tried to export it to the US - this is probably a clue.

IMHO 36/42 V is a dumb place to stop, anyway. If 12/14 isn't doing it for you anymore, either go up to 24/28 to use the bus and truck parts bin, or go up to 48/54 V to use the telecom parts bin.

In the late 90s, Dell had the brilliant idea to equip motherboards with the ATX connector, *but not the standard ATX pinout*. Their power supplies had the wires scrambled to match. Many exploded motherboards and power supplies later, they stopped doing this. Reference

near the bottom.

Matt Roberds

-48 volt power is still common for server stuff in telco buildings in sites. I have no doubt that it's more efficient, and a "few percent" can be thousands or tens of thousands of watts of saved power.

All the -48 volts PC power supplies I saw appeared to be higher quality and less no-name than the typical AC to DC server power supply. They were likely to be more efficient no matter how you measured it.

I've seen lots of this as well. My favorite setup like this was a series of racks setup, each with an -48 to 120VAC inverter for the equipment in each rack, all of which could have been ordered with DC power supplies. Local electical codes required circuit breakers be added for each inverter, so conduit was run across the room and back to circuit breaker panels on the wall, one box per inverter, each with one breaker and then back to the racks with a normal outlet. The amount of money burned on this must have been impressive, but it's how the the phone companes do stuff. The colocation provider also ran double the amount of fiber network connections that we needed as their order system believes everything is a SONET ring and there has to be a Protect pair of fiber. So each piece of fiber was carefully labelled and tested, but they lost records of where everything was connected on their side. Whoops.

So as silly as this all sounds, the phone companies know this too and have the last laugh when they bill you for their services.

Higher voltage (than 12/14V) drives up the cost of the electronics in a car significantly. Since the cost of electronics is already rising faster than any other part of the car, it's a losing idea.

Telecoms don't have to worry about load dumps.

Dell did a lot of stupid things to try to lock in their customers.

That's dubious; semiconductors to handle power can be made for high current by adding area, and for high voltage by modulating the doping. Adding area raises the production cost quickest. The conductor part of insulated wire is more expensive per foot than the insulation part, too. Forget about low logic voltage; that only takes fleapower. Electric motors and actuators work just fine off higher voltage. No cost premium there!

Probably a big drawback would be the tendancy for high voltages to arc, and continue arcing, in the solenoid or other contactors and fuses. So they have to be made differently, and usually more expensive.

You could switch the starter with a transistor of course, but then you need to go make a whole module for it. Which isn't really too bad, 30V (plus load dump) and maybe 200A. That said... it ain't never going to wear out, if it's rated properly, and it would probably save on weight too.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com

I don't know... I'm sure at some point somebody put a backhoe through a

400-pair cable that supported lots of subscribers chattering away on their 500 sets. On the other hand, there was probably a lot of load on that exchange from other cables that *didn't* go away right then.

Telecoms *do* have to worry about power crosses, but those wouldn't usually go directly into the DC plant.

Matt Roberds

"Properly rated" in this application means that it dies the day after the powertrain warranty expires. :) It's up to you whether you under-spec the transistor and heat sink, or put it on the bus so it can read the odometer and die at the appropriate time.

The module does let you replace the $5 BOM/$20 aftermarket retail part with a $3 BOM/$200 dealer-only retail part, so that's good.

Matt Roberds

It's not dubious at all (BTW, I said "electronics", not "motors"). High voltage processes are expensive. Assuming the load dump is a function of the voltage, it makes the mess *much* more expensive.

The load dump in a car is caused by the alternator becoming unloaded from the battery, not the load. In your example, there isn't an alternator that suddenly has no battery.

Completely different issue.

But, there only needs to be a 'high voltage' rating on the little DC/DC converter at the corner of the board, not on all the little electronic bits!

Bullshit. You obviously know nothing about vehicle electronics. It's not "little" and it is expensive.

There's always the 3 wire option: 24v battery, starter, alternator, headlights, motors, with 12v electronic goods. So even more fuses.

NT

Now you're offsetting the electronics to one side of the battery. The simplest way to do this is to use two batteries but you'd be discharging them at different rates. It took a while but the manufacturers figured out that it was a dumb idea to raise the battery voltage. BTW, electric vehicles have this problem in spades but no one is buying them to save money.