KILL ATX!!! (Power interconnects; form a new standard)

[Phas3L0ck]

For nearly 20 years now, we've been using that 20/24-pin power plug with a thick bundle of wires like it's just part of life, but as density becomes more relevant, cable management strategies for the current generation dictate that the ATX plug is OBSOLETE!

 

It's bad enough that it lands in a 2-inch (50mm+) connector, but the thick, massive bundle of weak, low-power wires are too painful to configure, rearrange, or design for as it takes up increasingly precious space in all parts of the system. And all of that bulk for a mere EIGHT signals (or 9 depending on system design) to go from the PSU to the mainboard!

Four of those eight are primary power; +12v, +5v, +3.3v, and Ground. Two others are +5VSB (Standby) and an oddly specific -12v. The remaining 2 signals are PS-ON and PWR-Good.

 

ATX NEEDS TO DIE!!!

 

Just short of redesigning existing boards entirely to use smaller, more compact connectors, I recommend a vertical right-angle adapter (or continuous vertical for larger systems) to re-route the 4 main power signals as converged rails on their own, singular wire(s) of a larger gauge. For example, the five +5v wires can be combined to run on a single line, reducing the wire count by x4.

Accounting for national electrical safety standards, and the allotment of current to each voltage, here are the wire sizes that would be appropriate for each line:

+3.3v: 12-10AWG (15-25 Amps) Copper, Aluminum, or CCA (copper-clad aluminum)

+5v: 10AWG CCA <30 Amps / 12AWG Copper ~20 Amps

+12v: 12AWG (10-20 Amps) Copper or CCA--- **EPS 12v Spec 2.92 states that common 12v connections should not exceed 240VA per rail**

GND 10AWG (~30 Amps, current varies) Copper or CCA; 12AWG may be used at engineer's discretion.

 

The remaining signals specified below shall remain 18AWG size, but should be fitted with their own, alternate connection in a smaller package to save board space and simplify a new wire harness design.

-12v _ +5VSB _ PS/ON _ PS/GD

 

I'm still working on how to re-fit the low-power and signal-only wires, but in the meantime, this is a prototype of what my idea would look like:

 

As you can see, I've converted the old 24-pin ATX connection into a right-angle cable assembly/adapter and modified the layout by recombining the cluster of 19 power wires, grouped by color-coded voltage, into a converged set of rails which terminate inside a 4-prong Molex Multicat connector; 2018400040.

Spoiler

 

 - The thing I love about the multicat connectors the most is that, not only can they handle a ton of current, (up to 40A) but they're excellent in my case for experimenting with different connection forms because the terminal pins are gold-plated over the entire surface! And the gold plating on the terminal pins makes them very easy to work with-- I don't have, nor can I ever afford Molex's hopelessly expensive crimp tool, (you don't wanna know) so I had to find some other way to adhere the wires permanently inside the terminals. Luckily, I have a powerful handheld Butane torch that I managed to use to heat up the terminals, which melted the solder inside and held the wires in.

So basically, I had to burn the connections together!  Took me months to learn how to do that without accidentally melting the thin wires before I was ready for a project like this. And it was sometimes hard to avoid letting solder run everywhere when it found a space to go through...

 

Click to open the hidden comment above to know why I chose this connector and how I managed to use it.

 

Great... I have a concept demonstration... Now what?

Well... Several problems.

First off, the connection via ATX is still too big!

Second, with that added non-standard Multicat plug, I'll have to make another custom cable like this, and possibly an extension to go between the two, using the new connector (yes, Multicat is actually a new model by Molex ~2016-2018) all to showcase ATX pumping energy across 4x 12-10AWG wires for this grand idea I came up with!

And third, with that awful ATX form being so widely adopted, this would be highly impractical for most end users, who likely don't have the high density or cable management problem that this is meant to solve!

 

Okay, maybe end users could see some level of benefit if some conditions are met. For a regular consumer to need this, the situation would be sort of like this:

 - Add or replace components inside PC fairly often (more than once every 2 months)

 - Space limitations inside system for moving or accessing components

  -- Airflow problem due to bad cable design or bulky cabling

 - Motherboard shape, size, or layout will never change  (this is a big one; I'm looking at you, Linus!)

 - Cable too long or too short  (especially a problem with "modular" cables)

 - Just wishing your PSU could have been better thought out, and put the bulky stuff somewhere else

 

With that in mind, this modification is really intended for servers, specifically 2U size like the one I have and am working on.

  **Side note, 1U is completely useless, 2U needs serious work in every aspect (most designs are HORRIBLY flawed), and 3U and larger are pretty much open-box assemblies that can be abused and re-worked like regular desktop cases.

The thing is, servers aren't technically that different from regular computers, except for one important thing; density.

Moreover, power cores in servers tend to be much more modular in a very literal sense-- so much to the point that any experienced technician (or serious power geek) could re-purpose a bar module like the DPS-450 or the 1K28P for almost anything quickly and easily, and make use of all the potential current specified by the nameplate capacity through a single massive blade surface.  Something that you can't do with most ATX PSU's. That's why there are so many adapters for CRPS-based modules to use them for bitcoin mining and GPU mods.

 

So what's the point? Modularity and density.

You can't change the connector on the motherboard without an expensive redesign, but that's where this connector prototype comes in...

 - "but you said you need another one like this to connect to the power supply. where's the point in that?"

This is where it gets interesting... Server power cores are more modular than you think.

Spoiler

And I refer to them as power 'cores' because they're not like a traditional power supply unit, which by definition implies there is only one, whereas common archetypes use two or more modules for power.

Based on pinouts of the ends of server power modules from several company designs and references from hobbyists who modded server modules, they aren't really proprietary, except in shape, and still follow enough of the ATX/EPS standard that they can be re-used in newer system designs, or accommodate other purposes. Technologically and financially, this makes sense for the developers and engineers since all they have to do to continue using a module from 10+ years ago (if it even physically fits, still*) for example, is modify parts of the design of the power distributor. Leave the important control circuity for monitoring and redundancy, but change the shape of the output!

That's all a power distributor is; a redundancy component with a set of connectors to match the shape of those on the target board!  After all, symmetrical cable assemblies are faster, cheaper, and easier to build than proprietary ones.

Why do you think they stopped making as many systems as there used to be with proprietary power wiring?  2 reasons; ATX standards/compatibility, and cost savings. That's why.

 

Therefore, it would be plenty reasonable to imagine a slight modification to power distributors, a relatively cheap part, to use fewer wires of a larger gauge for power delivery!

The remaining signal wires can be small and cheap still, using a smaller connection for itself as well...  Although I'm not sure I'd professionally use Multicat as my connector type for the PCB side or for singular individual wires across 12AWG for everything-- I'm honestly leaning towards Molex's MegaFit connector type for the final cable assembly design.

MegaFit has a slightly more compact size and familiar square shape, but Multicat happens to have the advantage of being great for high-current lines and prototype cables like this one as the terminals are big enough to handle that hideous bundle of BLACK from the ground line; seriously, 8 wires for one rail??? WHY?????  Even though these are all 18AWG wires, it was a pain in the A$$ squishing 8 of them together at once and getting them to fit properly in a terminal that was designed for 10-8AWG wires!!!  I couldn't imagine making this prototype cable with MegaFit terminals, designed for a maximum of 12AWG...  So, for the sake of this project, THANK YOU MULTICAT!  **Don't worry MegaFit, you'll shine soon enough**

 

 

Epilogue: I'm still trying to decide on a new shape for those last 4 signal wires, but I have 2 candidates so far for what I could use...

Here's what I think would be best for the signals with very little energy:

Connector option A: Molex CP 3.3 Wire-to-wire  (useful for a cable harness, but not for board connection as no PCB header is available)

 - Parts 5046930402 & 5046940402

Connector option B: Molex SL Wire/Crimp connectors (prime choice for wire-to-board connection; parts available for wire harnesses and PCB headers)

 - Parts 50579404, 701070003, & 705430038

 

Leave a comment and vote for which of these parts would work best for the signal-only wires and why. Also let me know if you have other ideas for what part should be used to connect the last 4 signal wires. For reference, the signals can be connected with any wire size ranging from 22-16AWG, so if you think you can find something better than what I mentioned, go crazy!

[Crunchy Dragon]

Quite honestly, I'd be happy if Molex died. That connector is such a pain...

[Phas3L0ck]

Just now, Crunchy Dragon said:

Quite honestly, I'd be happy if Molex died. That connector is such a pain...

To be clear, Molex is a vast company that makes lots of different connectors. You mean that you'd like to see the 20/24-pin MiniFit Jr. plug die as much as I do

[Crunchy Dragon]

2 minutes ago, Phas3L0ck said:

To be clear, Molex is a vast company that makes lots of different connectors. You mean that you'd like to see the 20/24-pin MiniFit Jr. plug die as much as I do

No, I'd like to see the 4-pin Molex connector die off.

[Phas3L0ck]

4 minutes ago, Crunchy Dragon said:

No, I'd like to see the 4-pin Molex connector die off.

Oh, the old drive connector?

Yeah, I've wondered about that since day-1 of building and rebuilding computers.

No idea what to replace that with.

Look around for a small 4-prong connector that can handle more than 4 Amps and let me know if you have any ideas on that one.

[LukeSavenije]

you can move something else out with an upcoming atx specification, being the minor rails

 

unfortunately that specification doesn't include any changes to the connectors that i know of

[porina]

9 minutes ago, Phas3L0ck said:

Oh, the old drive connector?

Yeah, I've wondered about that since day-1 of building and rebuilding computers.

No idea what to replace that with.

Look around for a small 4-prong connector that can handle more than 4 Amps and let me know if you have any ideas on that one.

The old 4 pin molex is practically dead, as pretty much everything has moved to the SATA power connector now. Of course they're not equal, we're replacing 12v and 5v with 5v and 3.3v, and I have no idea on the ultimate current carrying capacity. Maybe it doesn't matter at a practical level as things have got more efficient.

 

Back on the thread topic, is the goal to retrofit an alternative to the 24-pin mobo connector but keeping all its functionality? If a fresh start were possible, I'd like to go the server style route. One power rail only, 12V, or even higher like 24V to lower current per power, although 12V would still be required for PCIe things like GPUs. This does make me wonder, how much power still goes through the 24-pin as oppose to the EPS connector. If you really need other voltages, generate it on mobo. Maybe not so good for SFF builds... SATA remains a separate niche and requires both 5v and 3.3v to a lesser extent, so those would still have to be generated by PSU, but just not necessarily fed to the mobo.

[Phas3L0ck]

16 minutes ago, LukeSavenije said:

you can move something else out with an upcoming atx specification, being the minor rails

 

unfortunately that specification doesn't include any changes to the connectors that i know of

Actually, what connector gets used is ultimately the decision of the developers. The only reason motherboards follow the Spec first and foremost is to get them certified, and be more appealing for commercial use in existing infrastructure. PSU makers aren't always bound by such strict requirements, and can usually do what they want. (except in the case of companies that stick to ATX for their customer base) Server designs go their own route in terms of power delivery, and makers like Supermicro have gone as far as doing away with connectors all together and soldering wires directly to the distributor board-- they still have the standard plugs to match with the motherboard, but still, they do whatever they feel like.

 

11 minutes ago, porina said:

Back on the thread topic, is the goal to retrofit an alternative to the 24-pin mobo connector but keeping all its functionality? If a fresh start were possible, I'd like to go the server style route. One power rail only, 12V, or even higher like 24V to lower current per power, although 12V would still be required for PCIe things like GPUs. This does make me wonder, how much power still goes through the 24-pin as oppose to the EPS connector. If you really need other voltages, generate it on mobo. Maybe not so good for SFF builds... SATA remains a separate niche and requires both 5v and 3.3v to a lesser extent, so those would still have to be generated by PSU, but just not necessarily fed to the mobo.

First thing, yes.

Second, amen to that.

And third, according to my calculations, and using the current Spec as a reference, the ATX connector block should be capable of anywhere from 250-300+ Watts, just for the 20/24 pin link alone.

Lastly, 24V wouldn't work since boards would need a completely redesigned VRM and years of testing on components and new designs...

[LukeSavenije]

3 minutes ago, Phas3L0ck said:

ctually, what connector gets used is ultimately the decision of the developers. The only reason motherboards follow the Spec first and foremost is to get them certified, and be more appealing for commercial use in existing infrastructure.

what alternatives do you have to atx and eps specifications then? there's your main problem

 

then again, intel can easily make it 18 pin without really removing much but the minor rails, but haven't done so yet

 

and for the notice, this does mean you have two rails technically, the 12v and 12vsb

[Phas3L0ck]

On 1/5/2020 at 1:42 PM, LukeSavenije said:

what alternatives do you have to atx and eps specifications then? there's your main problem

 

then again, intel can easily make it 18 pin without really removing much but the minor rails, but haven't done so yet

 

and for the notice, this does mean you have two rails technically, the 12v and 12vsb

Seriously? Did you even read the full post? Or just the title?

For ATX, Multicat and MegaFit.

For EPS... custom. requires a special patch... more on that in a later post.

 

18-pin? HAH! HP did it with 6 or 7.

 

WHAT? Where did you get that from? Only the bloated ATX consumer bricks have more than one 12v rail. I'm talking about what runs in servers!

[porina]

12 minutes ago, Phas3L0ck said:

Lastly, 24V wouldn't work since boards would need a completely redesigned VRM and years of testing on components and new designs...

I'm thinking about it from a power transfer perspective. We've been on 12V for a long time, and the currents are getting rather high. I know there is still a lot of stuff still designed for 12V like PCIe power, but for things like CPU VRM it can be redesigned relatively easily.

 

Then again... for consumer level stuff, we seem to have reached a kinda power plateau so the push to higher powers isn't necessarily there any more. 

[Phas3L0ck]

1 minute ago, porina said:

I'm thinking about it from a power transfer perspective. We've been on 12V for a long time, and the currents are getting rather high. I know there is still a lot of stuff still designed for 12V like PCIe power, but for things like CPU VRM it can be redesigned relatively easily.

 

Then again... for consumer level stuff, we seem to have reached a kinda power plateau so the push to higher powers isn't necessarily there any more. 

That makes sense I guess.

Now that I think of it, we could retrofit existing 12v lines with buck-boost converters on both sides to alleviate some of the heat involved in high current/power. But yeah, there's not really a point for most stuff.

[Phas3L0ck]

UPDATE: Finally made the counterpart for the prototype I showed earlier, and while it may be ugly as h3ll, it works.

I assembled a MultiCat cable to act as the bus bar for main power, and terminated the signals with Molex CP3.3 just because I was curious about what it would look like and how it would feel.

CP3.3 is perfect for the signal lines on this prototype because the terminal pins are fairly large and easy to crimp onto 20, 18, and 16AWG wire sizes. The only thing I don't like about CP3.3 is that, not only do you have to press the release button to insert the connectors together as well as disconnecting them, but on the sides of that press button there are pegs that make it difficult to keep your finger there and apply the necessary pressure.

This is likely the last time I'll use obscure connectors on a power cable-- even a prototype.

Don't mind that extra black wire hanging out in the open-- it wouldn't fit (surprisingly) in the largest available MultiCat terminal pin, but that's okay since it's a common signal and isn't always necessary.

 

Enough screwing around; now that I know how to arrange the wires, my next prototype will be a completely custom direct-wire only version of this. In the 2nd prototype, a direct run of 12AWG copper (pure, stranded) across each power rail like a real bus bar will be used.  The only problem I have to solve is insulating each rail on bare wire, given the close proximity of each connection.

 

For those of you wondering, there will be a 3rd protoype; a pre-final test assembly that combines the ideas of the first 2 layouts. In version 3, the plan is to assemble the whole thing with a hand-made interposer-- the entire ATX plug will be short soldered against a strip of PCB prototyping board, with attachments made to a new connector intended to be used on the final design.