[SFF Network] 12V Please, Simplify!

[confusis]

This is one of a series of mini-rants by your faithful correspondent, John Morrison. These are a regular series focusing on issues in the SFF niche. All content is entirely opinion of John, not of SmallFormFactor.net, and should not be taken as fact.

As the ATX specification evolved, and as component manufacturers changed designs and other components, we have slowly edged towards a 12v dominant world. The two heavy hitters in your system, the CPU and GPU, both utilise 12V inputs. There's a good reason for this - they require higher amperage than the rest of your system.

Read more here.

 

[|||]

Would you prefer this over the current standard?





A single connector, replacing both the ATX connector and the 4/8 pin CPU connector. Yes, I know I have also omitted the power_ok signal wire – this, with some clever design, could be interpolated from a shared 12v sense pin.

Oh yeah! How do we get this pushed through to be adopted as an ATXv2 standard?

You know that overclocking boards will still have supplementary CPU power connectors, but that's not precluded by this.

 

[confusis]

No idea, but I'd love to see it. And don't call it ATXv2, call it SFFv2 ATX brings images of behemoth rigs!

 

[|||]

And don't call it ATXv2, call it SFFv2 ATX brings images of behemoth rigs!

Wouldn't it just be the first generation of "SFF" power connectors then? I haven't looked closely, but are there alternatives to the Mini-Fit Jr. connectors for consumer applications? If you're redoing the standard, you might as well look at everything involved that is changing.

 

[LocoMoto]

Let's call it the PUDDING standard, cause pudding is nice.

 

[jeshikat]

but are there alternatives to the Mini-Fit Jr. connectors for consumer applications?

I need to run to work, but there is a smaller version of Mini-Fit Jr. I'll look it up later.

 

[EdZ]

As a counterpoint (to a degree): why 12V?

All the 'high power' components that 'run' off of the 12V rail don't. Instead, CPUs, GPUs etc use 12VDC as merely an input for their own DC-DC stepdown conversion to the votlage they actually need, usually ~1V.
Now, distributing 1V inside the case is not a workable idea. You'd need bug chunky conductors, and modern CPUs/GPUs need rapid self-regulation due to the way they vary their operating voltage dynamically, so you would not even eliminate local converters.
Instead, raising the internal distribution voltage makes sense. 24V or even 48V are good choices as there are plenty of discrete components already available to deal with these voltages. You could use an arbitrary voltage like 50V, but then all your discretes need to be custom or wastefully overspecced. Higher voltages allow for more efficient DC-DC conversion, and also allow for finer gauge internal wiring for the same power capacity, due to the reduced current. Potentially they could allow for high power distribution via the same connectors that carry bus interfaces (e.g. increase the power cap for PCIe slot power, up to ~250W for 48V, without increasing the current limit).

The drawback is the lack of backward compatibility. Not a problem for motherboards, as you've already eliminated 3.3V, 5V, etc, anyway, but add-in cards would need to be distinct between 12V versions and 'high volt' versions, and/or replace their on-board power conversion hardware with a 'flexible' design that can handle a large DC range, or add an extra stepdown stage before the normal stepdown & regulation stage.

 

[jeshikat]

I need to run to work, but there is a smaller version of Mini-Fit Jr. I'll look it up later.

Okay, it's Micro-Fit: http://www.molex.com/molex/products...oducts&chanName=family&pageTitle=Introduction

Has a 3.0mm pitch compared to 4.2mm on Mini-Fit Jr

raising the internal distribution voltage makes sense. 24V or even 48V are good choices as there are plenty of discrete components already available to deal with these voltages.

Yup, datacenter operators like Google are considering 48V: http://www.eetimes.com/document.asp?doc_id=1328741

 

[EdZ]

Yup, datacenter operators like Google are considering 48V: http://www.eetimes.com/document.asp?doc_id=1328741

And they've contributed it to the Open Compute Project of which Nvidia, AMD and Intel are all members. Which raises the liklihood of reference designs incorporating direct 48V conversion becoming available in the next few years, which could trickle down from DC hardware variants to consumer devices.

 

[Kmpkt]

So if I understand correctly, 24V uses half the amperage of 12V and 48V uses half the amperage of 24V correct? Real world, what does that mean for wire gauge and connector types? I assume much lighter wire and smaller connectors is the answer, both of which would obviously be beneficial to the SFF community.

 

[jeshikat]

So if I understand correctly, 24V uses half the amperage of 12V and 48V uses half the amperage of 24V correct?

Correct. Wattage = Voltage x Amperage

So doubling the voltage would halve the amperage, allowing thinner and/or fewer wires.

 

[Kmpkt]

I guess what I was getting at was how large a difference would that make? Could we safely swap to say 20 AWG at 24V and maybe 24 AWG at 48V or something like that?

 

[Thehack]

I guess what I was getting at was how large a difference would that make? Could we safely swap to say 20 AWG at 24V and maybe 24 AWG at 48V or something like that?

For reliability and sturdiness purpose we would likely maintain a decent sized wire gauge. Smaller wires require more delicate crimps and labor.

It is best to just maintain the mini fit size and lower total number of wires in my opinion. Reducing wire gauge you end up with weak connectors.

 

[jeshikat]

I guess what I was getting at was how large a difference would that make? Could we safely swap to say 20 AWG at 24V and maybe 24 AWG at 48V or something like that?

Edit: This chart is for the regular Mini-Fit Jr series.

But I agree with @Thehack, going really small on the wire gauge has its own set of issues. So I think 18-20AWG but fewer of them is more practical.

 

[The Dude]

A single connector, replacing both the ATX connector and the 4/8 pin CPU connector. Yes, I know I have also omitted the power_ok signal wire – this, with some clever design, could be interpolated from a shared 12v sense pin.

Since we're at it, let's throw in some I2C wires. For RGB LEDs configuration. And also for a less useful stuff, like temp/RPM/load monitoring.

 

[Josh | NFC]

48v all the way.

 

[Kmpkt]

Perhaps this community could be an opportunity to start pushing 48V with appropriate step down DC-DC boards as the new standard?

 

[Thehack]

Perhaps this community could be an opportunity to start pushing 48V with appropriate step down DC-DC boards as the new standard?

If we are talking about a separate step down dc-dc board for 48V to 12V then you end up having more components, more heat, more complexity.

We should follow whatever primary voltage board maker produces. If anything we should aim for 12V, 24V, or 48V. Easy standardized voltages that are common throughout components. The thin itx using 19V throws things off a bit.

 

[EdZ]

Yep, this is something that would only make sense if adopted by component makers - specifically, motherboard manufacturers (and therefore Intel and AMD who set their required specs) and GPU manufacturers - or we'd just be adding an extra DC-DC step for no benefit.

 

[Phuncz]

Or just reroute all the the power through one or two cables through the motherboard, if we can eliminate 3/4th of the 12V wires with 48V. I expect that 300W for GPUs should be plenty if 75W in a slot is now possible. Storage could easily run off of the board too, like Thin-ITX.