How Intel's changing the future of power supplies with its ATX12VO spec

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
More information about this cool power supply spec! We've seen this previously documented!

"Your ATX12VO Main Power Connector would be far smaller, and the cable would be more flexible. If the board has enough power from the single connector, the board maker may not even require you to plug in the auxiliary 8-pin power connector. The spec does allow for auxiliary 12-volt power through the EPS12V connector.

One tricky part could be hooking up any SATA-powered drives, such as hard drives or 2.5-inch SSDs. Today, you’d plug those directly into the PSU. On an ATX12VO build, you’d first plug a power cable into the motherboard and then into the drive. The spec allows for up to six power connectors, but it’s up to the motherboard vendor to determine how many power connectors there are. These same SATA power connectors would be used to power up your drives as well as your AIO/CLC cooler or RGB LEDs.

If you want to plug in an old-school Molex connector, the new spec allows PSU vendors to offer that directly from the PSU, but only with 12 volts, of course. If you’re hooking a really old 5-volt Molex device, you’ll need to get that from the motherboard power using a SATA-to-Molex connector.

For a DIYer, it really won’t be that different. The real question is how it’ll work with motherboards and PSUs."



1583771165385.png


https://www.pcworld.com/article/351...-of-power-supplies-with-its-atx12vo-spec.html
 

Attachments

  • 1583771168232.png
    1583771168232.png
    5.9 MB · Views: 0
12V is dumb. Give us 48V already...
Four years ago, this was supposed to be the next thing. Sadly, little seems to have come of it.
higher-density-graphic.jpg

http://powerblog.vicorpower.com/2016/10/vicor-transforming-datacenters-12v-48v/


Way past this old demo in both GAN and SIC today.
I test a lot of GAN, can't speak for SIC except paper specs.

With zero voltage switching, even plain old silicon will do.
VICOR above wasn't using wide bandgap transistors.
Least, I don't think it was...
 
Last edited:
Yea lets move the cost from a 1 time power supply purchase to be spread out and added to every plug in device purchase for the life of the format. No thanks.

This means the power handling needs to be done in the motherboard or the plug in device that needs to draw power from the motherboard now. I don't think I like this... We already have power capacitor issues with some vendors.
 
I said it before and I'll say it again. This "solution" by Intel is a solution looking for a problem. All the while creating problems of its own.

The only real "solution" proposed here is moving what already works perfectly fine on PSUs to the motherboard. It's highly unlikely we'd get any cost savings with the PSUs in the process. However, motherboard complexity will go up as obviously will motherboard cost. Not to mention that there are enough issues with complexity on motherboards as it is. Adding in additional complexity, additional trace runs, additional electrical noise and who knows what other problems to motherboards is not a solution. It's a problem looking to be created.
 
Not sure of the benefit of moving to 12V only PS when it seems most devices want 3.3V or lower. What devices actually use 12V any more? Spinning HD maybe? I would much rather have the heat of turning higher voltages into lower voltages contained in the PS where it is trivial to have a large cooling fan and matching heat sink.
 
I said it before and I'll say it again. This "solution" by Intel is a solution looking for a problem. All the while creating problems of its own.

The only real "solution" proposed here is moving what already works perfectly fine on PSUs to the motherboard. It's highly unlikely we'd get any cost savings with the PSUs in the process. However, motherboard complexity will go up as obviously will motherboard cost. Not to mention that there are enough issues with complexity on motherboards as it is. Adding in additional complexity, additional trace runs, additional electrical noise and who knows what other problems to motherboards is not a solution. It's a problem looking to be created.

There is a major problem with ATX and the insane amount of power rails that are required. Good design would move power supplies up to 48v and move voltage regulation closer to the load, minimising voltage drop under load. You could also supply the same effective power with 1/4 the conductors / copper.

Intels spec is garbage though.
 
What devices actually use 12V any more? Spinning HD maybe?

Video cards, certainly--the PCIe connectors are +12V and ground. IIRC SATA drives take 12, 5, and 3.3.

The CPU uses lower voltage, of course, but there's already parts onboard to handle that.

As the article points out, one place this could be useful is ITX-format boards, because shrinking the power connector would free up a bit of space. However, two SATA power connectors would eat all the savings.
 
Video cards, certainly--the PCIe connectors are +12V and ground. IIRC SATA drives take 12, 5, and 3.3.

The CPU uses lower voltage, of course, but there's already parts onboard to handle that.

As the article points out, one place this could be useful is ITX-format boards, because shrinking the power connector would free up a bit of space. However, two SATA power connectors would eat all the savings.

They don’t have to be 12v, that is just how the mosfets are configured. It would be a trivial task to change them to 24v or 48v and simplify the PCB layout a fair bit.
 
They don’t have to be 12v, that is just how the mosfets are configured.

That's not what I was talking about. I was talking about the other voltages. Take +5 and +3.3 off the PSU connector and you need to put voltage converters on the motherboard, like the article says. (I suppose you could change hard drives to not use +5 or +3, but even if that makes sense, now you have different connectors and a compatibility issue. "Is this hard drive SATA or alt-SATA?"
 
Video cards, certainly--the PCIe connectors are +12V and ground. IIRC SATA drives take 12, 5, and 3.3.

The CPU uses lower voltage, of course, but there's already parts onboard to handle that.

As the article points out, one place this could be useful is ITX-format boards, because shrinking the power connector would free up a bit of space. However, two SATA power connectors would eat all the savings.
They accept it because that is what current PS supply. But how many devices USE 12V? Memory is often 1.5V or less. CPU voltages are getting lower as transistor gaps shrink. Do SSDs use 12v or mostly ignore those wires in favor of the 5v lines? Do GPUs use 12V as is or spend card space for voltage regulators to drop it to the voltage needed? At some point, the space needed for motherboard voltage converters and the requisite heat sinks will get close to the space needed for the actual computing parts.

I guess my question is does it save MB space and heat to feed 12 (or 24/48V) to the MB and let the MB convert to 3.3 and lower voltages or is it better to let the PS do the conversion to 3.3 or 1.5 and use larger wires to feed the power to the MB? If we are redesigning the power standard, nothing says we can't have several two or four pin low voltage plugs that connect to the MB adjacent to the parts that use that voltage. Might be some space and trace savings over the current 20/24 pin standard near the edge of the MB.
 
Why would you want to route power to all devices trough the motherboard? What a terrible idea. So you can force people to buy hiend MBs if they wish to connect more than 4 drives? Beh.
 
The majority of the PSU is dedicated to AC-DC conversion. Converting 12v to the other voltages doesn't take much space at all. PicoPSUs do exactly that, and are hardly any larger than the 24 pin ATX connector.
It's not about space, psus don't get obsolete, so why would you want to move the DC to DC conversion to the mB that does get obsolete and often gets replaced. Besides if you need more power you just replace the PSU, now you'd need to replace the MB if you want to power 10 SATA HDDs.

This is just a completely uneccessary reason to make MBs more expensive and complicated while power supplies less of a self contained unit in the PC that you could upgrade independently.
 
It's not about space, psus don't get obsolete, so why would you want to move the DC to DC conversion to the mB that does get obsolete and often gets replaced. Besides if you need more power you just replace the PSU, now you'd need to replace the MB if you want to power 10 SATA HDDs.

This is just a completely uneccessary reason to make MBs more expensive and complicated while power supplies less of a self contained unit in the PC that you could upgrade independently.

It is already on the mainboard.... it is just different voltages.
 
It is already on the mainboard.... it is just different voltages.
Just because vcore is generated on the MB doesn't mean adding sata power connectors and supplying power to drives is not more unneccessary clutter increasing the cost of the MB. Not to mention if you add more than 6 drives what then? How many Y cables can the MB handle before overheating or shutting down? This is a way to segmentate hiend and loend boards by one more metric: How many drives can they supply power to. Who needs this? Who asked for this? This is innovating for the sake of innovating, not solving an actual problem.
 
I guess my question is does it save MB space and heat to feed 12 (or 24/48V) to the MB and let the MB convert to 3.3 and lower voltages or is it better to let the PS do the conversion to 3.3 or 1.5 and use larger wires to feed the power to the MB? If we are redesigning the power standard, nothing says we can't have several two or four pin low voltage plugs that connect to the MB adjacent to the parts that use that voltage. Might be some space and trace savings over the current 20/24 pin standard near the edge of the MB.


The wires to power a CPU from such a supply would be too large to be practical. Remember CPUs can operate at several hundred amps.
 
The wires to power a CPU from such a supply would be too large to be practical.

In all cases? I suspect not for budget systems. Besides, if Intel's talking about 12V-only connectors, surely they must feel that it's feasible. Unless you were talking about 24/48V.


I went and opened up my current work box, a Dell Optiplex 5060. It has a 6-pin motherboard connector: green, gray, and two each white and black. There's also a 4-pin EPS. That's all that's coming off the motherbaord. Just like on my older Dell, there's a 6-pin socket on the motherboard, feeding two SATA power cables, each with 2 sockets.
 
In all cases? I suspect not for budget systems. Besides, if Intel's talking about 12V-only connectors, surely they must feel that it's feasible. Unless you were talking about 24/48V.


I went and opened up my current work box, a Dell Optiplex 5060. It has a 6-pin motherboard connector: green, gray, and two each white and black. There's also a 4-pin EPS. That's all that's coming off the motherbaord. Just like on my older Dell, there's a 6-pin socket on the motherboard, feeding two SATA power cables, each with 2 sockets.

Feeding 1.5v directly for a CPU is entirely not feasible unless it's a laptop level CPU. You would need almost 100 times the thickness or number of cables to carry the same amount of power as compared to 12 volts.
 
The wires to power a CPU from such a supply would be too large to be practical. Remember CPUs can operate at several hundred amps.
Around 100A. My 9900K will use around 145W at 1.32V which is 110A. Your point is taken, though, as the primary issue with electrical transmission in a PC is the distance from the source which is why we use +12V for devices needing higher wattage.
 
Feeding 1.5v directly for a CPU is entirely not feasible

Ah--I had to go back and re-read several posts. You were replying to the person who was apparently asking about adding MORE voltages from the PSU, and I missed that nuance. Carry on.
 
Not sure of the benefit of moving to 12V only PS when it seems most devices want 3.3V or lower. What devices actually use 12V any more? Spinning HD maybe? I would much rather have the heat of turning higher voltages into lower voltages contained in the PS where it is trivial to have a large cooling fan and matching heat sink.

Almost nothing in the computer uses 12V. I mean, other than the CPU, GPU, and fan, but those are not really important components.

You're correct that the CPU and GPU cores do not operate at 12V. The problem is that they pull a lot of watts. Where P=IV, we see that a 300W GPU requires 25 amps of current if delivered via a 12V supply. At 1.1V core voltage, that's 272 amps. But why does this matter?

Any time you move power down a wire, you lose some of it to heat. P=IV, and with P and I fixed, we get a voltage drop due to that loss. The amount of the loss depends on the resistance of the wire and the number of amps you're pushing through it - just like pressure loss in the straw when you're sucking down a Frosty. This loss can be quite substantial. Let's take a look for a GPU.

If we assume the wires are 18AWG with 65/36 stranding, we get a wire with a resistance of 0.00678 ohms per foot. We have a round trip distance of 3ft, our load is split evenly across 6 round trip pairs, and our contact resistance is .04 ohms per pair (.01 x 4 connections). This works out to a total round trip resistance of .06 ohms on each pair or 0.01 ohms for the set.

We also know that V=IR. Thus, if a 300W GPU is fed via 12V using this 6-pair set, we see that the V(oltage drop) = 25 * .01 = .25V. That's almost 25% of the core voltage. That's a substantial drop. That works out to 6.5W of heat that gets lost simply to the cabling.

But what happens if we use your idea of feeding the core voltage down the line? Well, V=IR again, so 272 * .01 = 2.72V of drop. Ergo, it would literally be impossible to supply the core voltage over the 18awg wire. BTW, this would be 740 watts of heat lost just in the cable. Are you SURE you want this in your case? You'd have to run a massive custom water loop simply to cool a single GPU cable.

What if we supply 1.5V in order to keep voltage low but still allow a little bit of drop? How big of a wire would we need? We need our 272 amps to have only a 0.4V drop. V=IR and thus R = V/I. From this, we get a target resistance of 0.001 ohm. Hmmm. If each wire pair has 0.04 ohms of resistance just from the connectors, we're going to need more circuits - at least 41 of them (with 2 wires each) if we stick to regular ATX pins. With this 82 cable harness, we're left with 0.000025 ohms of resistance for the wires. How big of a wire are we talking for this? Well, 0.000025 ohm split across 41 pairs allows a 3ft wire resistance of .000341 ohms per foot. To hit that number, we're looking at 4AWG 7/12 wire. This wire is .242" diameter and weighs 0.426lbs per 3ft.

With all of that math out of the way, what are the final specs on your 1.5V GPU cable harness? Each of these cable harnesses (one per GPU plus one more for the CPU) weighs 17.5lbs and comes in a 1.25" diameter bundle. Good luck trying to route this.

So, yeah. That's the start of why PSUs don't supply core voltage directly.
 
Last edited:
Reduced wire loss is why power should be distributed at the highest voltage (and lowest current) that folks aren't likely to Darwin themselves on it.
48V might be pushing the envelope, yet not any worse than POE. A few tards will still find a way, no doubt...
 
Last edited:
This is not progress unless Intel have something up their sleeve that depends on this being implemented.
 
Almost nothing in the computer uses 12V. I mean, other than the CPU, GPU, and fan, but those are not really important components.

You're correct that the CPU and GPU cores do not operate at 12V. The problem is that they pull a lot of watts. Where P=IV, we see that a 300W GPU requires 25 amps of current if delivered via a 12V supply. At 1.1V core voltage, that's 272 amps. But why does this matter?

Any time you move power down a wire, you lose some of it to heat. P=IV, and with P and I fixed, we get a voltage drop due to that loss. The amount of the loss depends on the resistance of the wire and the number of amps you're pushing through it - just like pressure loss in the straw when you're sucking down a Frosty. This loss can be quite substantial. Let's take a look for a GPU.

If we assume the wires are 18WG with 65/36 stranding, we get a wire with a resistance of 0.00678 ohms per foot. We have a round trip distance of 3ft, our load is split evenly across 6 round trip pairs, and our contact resistance is .04 ohms per pair (.01 x 4 connections). This works out to a total round trip resistance of .06 ohms on each pair or 0.01 ohms for the set.

We also know that V=IR. Thus, if a 300W GPU is fed via 12V using this 6-pair set, we see that the V(oltage drop) = 25 * .01 = .25V. That's almost 25% of the core voltage. That's a substantial drop. That works out to 6.5W of heat that gets lost simply to the cabling.

But what happens if we use your idea of feeding the core voltage down the line? Well, V=IR again, so 272 * .01 = 2.72V of drop. Ergo, it would literally be impossible to supply the core voltage over the 18awg wire. BTW, this would be 740 watts of heat lost just in the cable. Are you SURE you want this in your case? You'd have to run a massive custom water loop simply to cool a single GPU cable.

What if we supply 1.5V in order to keep voltage low but still allow a little bit of drop? How big of a wire would we need? We need our 272 amps to have only a 0.4V drop. V=IR and thus R = V/I. From this, we get a target resistance of 0.001 ohm. Hmmm. If each wire pair has 0.04 ohms of resistance just from the connectors, we're going to need more circuits - at least 41 of them (with 2 wires each) if we stick to regular ATX pins. With this 82 cable harness, we're left with 0.000025 ohms of resistance for the wires. How big of a wire are we talking for this? Well, 0.000025 ohm split across 41 pairs allows a 3ft wire resistance of .000341 ohms per foot. To hit that number, we're looking at 4WG 7/12 wire. This wire is .242" diameter and weighs 0.426lbs per 3ft.

With all of that math out of the way, what are the final specs on your 1.5V GPU cable harness? Each of these cable harnesses (one per GPU plus one more for the CPU) weighs 17.5lbs and comes in a 1.25" diameter bundle. Good luck trying to route this.

So, yeah. That's the start of why PSUs don't supply core voltage directly.
I seriously doubt the traces on the MB could begin to carry the 272 amps in your example. If a wire in free air can't do it, a trace buried in layer three of a multi layer MB can't either. So either the components on the MB are NOT using 272 amps of current @ 1.1 V or there is some serious magic juju going on as 272 amps in a MB trace should produce an expanding plasma ball and high speed MB bits.

Your concerns on lower voltage = higher current for same power are legit
 
I seriously doubt the traces on the MB could begin to carry the 272 amps in your example. If a wire in free air can't do it, a trace buried in layer three of a multi layer MB can't either. So either the components on the MB are NOT using 272 amps of current @ 1.1 V or there is some serious magic juju going on as 272 amps in a MB trace should produce an expanding plasma ball and high speed MB bits.

Your concerns on lower voltage = higher current for same power are legit

Eager to hear your New Math thoughts on how many amps it takes for 300W at 1.1V.

I'm also curious about which CPU you're running that has only one transistor in it.
 
Honestly I'm not sure what to think about it. On one hand it makes sense but on the other it's going to make PSUs cheaper and motherboard more expensive and I'm likely to swap out motherboard sooner than psu. Also this would make using old LSA or sata cards impossible since motherboard would be likely to have just enough power connectors for its use.
 
I seriously doubt the traces on the MB could begin to carry the 272 amps in your example. If a wire in free air can't do it, a trace buried in layer three of a multi layer MB can't either. So either the components on the MB are NOT using 272 amps of current @ 1.1 V or there is some serious magic juju going on as 272 amps in a MB trace should produce an expanding plasma ball and high speed MB bits.

Your concerns on lower voltage = higher current for same power are legit

When the distance between power source and power user is measured in centimeters instead of meters, the amount of power loss due to power transmission is significantly lower. 1 cm has 100 times lower resistance than 1 m. In addition, the entire motherboard acts as a heatsink for the trace. An insulated wire has limited heat transfer properties.
 
have no idea where your '1 transistor' thought came from.

It's a follow on from your 1-trace assertion. If some new generation of CPU came out that had more than one transistor in it, then it would be reasonable to assume that there would also be more than one trace to share the load. Imagine a future where there are CPUs with, say, 1.7 billion transistors. Then each one would only be taking on 0.00000016 amp.

I bet in that future, the main supply would be delivered to the motherboard at 12V through an EPS plug in order to keep the amps at a level which the motherboard traces could handle before going through a multistep conversion down to the core voltage wherein each step also increases the number of traces and items sharing the load.

What a crazy future that would be, right?
 
12V is dumb. Give us 48V already...
Four years ago, this was supposed to be the next thing. Sadly, little seems to have come of it.


No thanks. 48v is about the shock theshold for humans. I work with a lot of live power supplies. lol.

Eager to hear your New Math thoughts on how many amps it takes for 300W at 1.1V.

I'm also curious about which CPU you're running that has only one transistor in it.

Agreed man. There's a reason why modern power transmission lines are all around 2-3v. And, as a side note, no dead parrots when sitting on the power lines.
 
Will GPUs be able to be powered directly from the motherboard with this change or do we still need extra connectors? I would be game for a standard upgrade where we didn't need any external connectors to anything. It would sure clean up the cases.
 
Will GPUs be able to be powered directly from the motherboard with this change or do we still need extra connectors? I would be game for a standard upgrade where we didn't need any external connectors to anything. It would sure clean up the cases.

[glances over at Mac Pro]
 
Will GPUs be able to be powered directly from the motherboard with this change or do we still need extra connectors?

You will most likely still need extra connectors.
 
Shrug. Most of the server power supplies I work with already have only 12V and one standby voltage (either 12VSB or 5VSB).

Intel should put pmbus into the standard too. Would be nice to get some dynamic PSU info.

Like others mentioned, it would be interesting to see a move to 48V. I doubt it would be a big shock hazard except in really strange situations (getting stabbed with 48v pins for example). I believe some nvidia enterprise stuff already requires direct 48V. Would definitely make it easier with the really high current stuff.
 
Back
Top