Power Supply A Guide to 12V PSU

[Thehack]

Thank you @Thehack for the wonderful guide and everyone who contributed to the discussion. It made for a worthwhile read.

I've been trying to figure out the power solution for my next build that runs under 400W. Run one EPP-400-12 through a PDCB or run one EPP-200-12 through a PDCB and one EPP-200-12 to the video card? Handling power to the power supplies won't be a problem from the AC side, it's more a space savings measure.

I prefer not to run 2 AC-DC PSUs because you get weird issues like some motherboards/GPU wiring the 12V rail together and the power following from one PSU to the other.

Though most power is "consumed" at closer to the load device, but I'd avoid as much weird multi power supply unless there is a working load sharing circuit.

 

[aquelito]

This eGPU is using a Meanwell EPP-400-12 as well as an interesting PCB on the AC side of the PSU.

The main component on the small PCB is a miniature 5V PSU.
The small black cable may to the PS_ON connector on the Meanwell PSU (located on the right bottom corner) ?

Part of the solution ?

https://egpu.io/blackmagic-egpu-review-apples-ultrafine-curse/

 

[Thehack]

It is probably for the controller logic. Most controllers use 5V for power.

 

[Soul_Est]

I prefer not to run 2 AC-DC PSUs because you get weird issues like some motherboards/GPU wiring the 12V rail together and the power following from one PSU to the other.

Though most power is "consumed" at closer to the load device, but I'd avoid as much weird multi power supply unless there is a working load sharing circuit.

Points taken. I'll look into using the EPP-400-12 + Pico PSU then. Thank you for the excellent advice.

 

[aquelito]

Ok so the circuit controlling the Meanwell PSU I mentioned earlier on is much more than I thougth it was.

This is in fact a kind of PicoPSU that the person behind my risers designed a while back.
It controls the PS_ON signal of the Meanwell : no need for a load switch anymore !

He made a prototype for his computer and will try to make an other one.
And maybe a full product out of it.

 

[smitty2k1]

Ok so the circuit controlling the Meanwell PSU I mentioned earlier on is much more than I thougth it was.

This is in fact a kind of PicoPSU that the person behind my risers designed a while back.
It controls the PS_ON signal of the Meanwell : no need for a load switch anymore !

He made a prototype for his computer and will try to make an other one.
And maybe a full product out of it.

I have no idea what this means but it makes me excited none the less.

I'm about to tear apart and rebuild my MeanWell 200W + Realan/MiniBox DC-ATX board build this weekend and rebuild it with a slightly more powerful DC-ATX board and a shiny new 2400G APU + B450 build to replace an aging 3245 series i3 and H77 motherboard.

 

[Thehack]

I have no idea what this means but it makes me excited none the less.

I'm about to tear apart and rebuild my MeanWell 200W + Realan/MiniBox DC-ATX board build this weekend and rebuild it with a slightly more powerful DC-ATX board and a shiny new 2400G APU + B450 build to replace an aging 3245 series i3 and H77 motherboard.

Are you using the minibox version?

 

[smitty2k1]

Are you using the minibox version?

No, I've got the 180W and 250W as listed here: http://www.minicase.net/power-supply.html

They are not the plug in DC boards, which in hind sight I should have gotten but my Realan case has mounts built in for these boards

 

[Thehack]

No, I've got the 180W and 250W as listed here: http://www.minicase.net/power-supply.html

They are not the plug in DC boards, which in hind sight I should have gotten but my Realan case has mounts built in for these boards

SPEC wise the "250W" version says it does 120W continuous and 144W peak. Which is fine for 2400G but it's a bit misleading to call it "250W."

 

[smitty2k1]

SPEC wise the "250W" version says it does 120W continuous and 144W peak. Which is fine for 2400G but it's a bit misleading to call it "250W."

144W on the 12V rail, my math says around 237W between all the rails, right? The "180W" version has been powering my 55W i3 CPU for many years now, but I figured I'd bump up to the larger one for my new 65W APU. Will definitely be running some Prime95 and Furmark to stress test it over the weekend when I get it all put together.

Regardless, they are certainly not the greatest, but it was only $22 delivered and the case already has mounts for it. It also has a mysterious 4pin connector next to the 20 pin ATX connector that I think is supposed to power a graphics card, but I haven't tested it out as a I don't have a multimeter. The 20-24pin ATX cable has the typical breakouts for a SATA, Molex4 pin, and 4pin motherboard header as expected.

The PicoPSUs are much more expensive albeit smaller.

 

[NateDawg72]

144W on the 12V rail, my math says around 237W between all the rails, right?

Yeah that may be the case. The reason TheHack calls it out is because modern PCs make little use of the other rails, so the 12v rail is by far the most important. Since they add up all the rails to get close to the rated wattage it suggests either an old design or an attempt to pad the numbers - or both if the combined rails is only 237.

 

[Supercluster]

For my use case the 400W ones are just too low. How bad would it be if i used something like this Tracopower. It is rated for 450W ventilated + OCP kicks in only over 115%, which if I am interpreting it correctly means that properly cooled it can push 450W+14% safely which is over 500W. There is a downside though... The ripple is cited as 250mV Pk-Pk. There is no test report to check for actual value... Basically what I am asking is for an educated guess if the ripple would normally be under 100 mV?

Any thoughts appreciated. (as well as alternatives to push over 450W in the 3 x 5 inch form factor)

 

[Thehack]

For my use case the 400W ones are just too low. How bad would it be if i used something like this Tracopower. It is rated for 450W ventilated + OCP kicks in only over 115%, which if I am interpreting it correctly means that properly cooled it can push 450W+14% safely which is over 500W. There is a downside though... The ripple is cited as 250mV Pk-Pk. There is no test report to check for actual value... Basically what I am asking is for an educated guess if the ripple would normally be under 100 mV?

Any thoughts appreciated. (as well as alternatives to push over 450W in the 3 x 5 inch form factor)

For above 400W you're looking at active cooling for most of the components. It may more worth while to use SFX because their foot print is similar to 3x5 + DC board.

But if you want stupid amount of power density in exchange for worse ripple control then check out fidus power.

http://www.fiduspower.com/ac-dc-power-supply/open-frame/nevo-600s

There are other products nearing 600W with active cooling.

 

[Supercluster]

For above 400W you're looking at active cooling for most of the components. It may more worth while to use SFX because their foot print is similar to 3x5 + DC board.

But if you want stupid amount of power density in exchange for worse ripple control then check out fidus power.

http://www.fiduspower.com/ac-dc-power-supply/open-frame/nevo-600s

There are other products nearing 600W with active cooling.

SFX is roughly double the volume (and triple the weight probably, both as important). However, the FSP500-50FSPT could be stripped and would probably be about 1cm longer but otherwise almost identical as the 3 x 5 form factor. My guess is about 100-150g heavier (stripped). The question is whether to potentially waste 150 USD just to see the size inside.

 

[Supercluster]

Hmm, I hope that this is possible (I just read the about the -12v deficiency of the FSP). Could two 2 x 4 inch be used in "faux" parallel configuration. The idea is that one be delegated to power the plug-in DC-DC with a 4-pin EPS and an 6pin(or an 8 pin) GPU, and the other with the other 4 pins of the EPS + 8pin GPU, thus splitting the load (in theory). Is that technically parallel or not... am I missing something?

 

[Thehack]

Hmm, I hope that this is possible (I just read the about the -12v deficiency of the FSP). Could two 2 x 4 inch be used in "faux" parallel configuration. The idea is that one be delegated to power the plug-in DC-DC with a 4-pin EPS and an 6pin(or an 8 pin) GPU, and the other with the other 4 pins of the EPS + 8pin GPU, thus splitting the load (in theory). Is that technically parallel or not... am I missing something?

For our hobby stuff... it's okay. Sketchy but it's okay.

For a hardy system, you can't do that unless the AC-DC units are designed to work in parallel. The slight difference in voltages causes power to flow from one PSU to the other. For example, PSU-A is 12.01 V and PSU-B is 12.1V. Therefore, PSU-A flows to PSU-B.

Above 400W you can implement a load share circuit: http://www.ti.com/lit/an/slyt100/slyt100.pdf

two 2x4 is about the size of 1 3x5 anyways, but I know it offers certain advantages due it's modular size.

 

[Supercluster]

For our hobby stuff... it's okay. Sketchy but it's okay.

For a hardy system, you can't do that unless the AC-DC units are designed to work in parallel. The slight difference in voltages causes power to flow from one PSU to the other. For example, PSU-A is 12.01 V and PSU-B is 12.1V. Therefore, PSU-A flows to PSU-B.

Above 400W you can implement a load share circuit: http://www.ti.com/lit/an/slyt100/slyt100.pdf

two 2x4 is about the size of 1 3x5 anyways, but I know it offers certain advantages due it's modular size.

Not only the modular size but also great power density with tested ripple control and a complete certification: check it out. Also has a beautiful STEP file.
The load sharing circuit sounds awesome, even better than one high powered unit. however, it is beyond my skill at the moment. Although the circuit does not seem too scary.

 

[Thehack]

Not only the modular size but also great power density with tested ripple control and a complete certification: check it out. Also has a beautiful STEP file.
The load sharing circuit sounds awesome, even better than one high powered unit. however, it is beyond my skill at the moment. Although the circuit does not seem too scary.

If you want 400-500W, you can, use the HD-PLEX 400. Which it is a bigger unit due to being a DC-DC regulator (12V DC-DC boards don't need to regulate the main rail, just switch it on/off), you can run the PSU in series, 12+12=24V which is within the specs range of the HDPLEX 400.

In the end it ends up being a wash, because if you use two 12V in parallel, you need a load share circuit. If you use two 12V in series, you need a fat VRM circuit to pull it back down to 12V again. Though, component cost wise, the 12V parallel edges out in being lower cost.

 

[Supercluster]

If you want 400-500W, you can, use the HD-PLEX 400. Which it is a bigger unit due to being a DC-DC regulator (12V DC-DC boards don't need to regulate the main rail, just switch it on/off), you can run the PSU in series, 12+12=24V which is within the specs range of the HDPLEX 400.

In the end it ends up being a wash, because if you use two 12V in parallel, you need a load share circuit. If you use two 12V in series, you need a fat VRM circuit to pull it back down to 12V again. Though, component cost wise, the 12V parallel edges out in being lower cost.

Meh... I guess it's been too long since i've slept. Thanks, i've learned things.

Edit:
The cheapest variant is most definitely the EPP-400-12. Then I finalize the design and wait for technology to bring the 500W.

 

[Choidebu]

you can run the PSU in series, 12+12=24V which is within the specs range of the HDPLEX 400.

May I ask @Thehack , how do you wire 2 ac dc converters in series? I've never heard of it. Do you just wire the second's negative to the first's positive? And maybe insulate the second's case just in case it's grounded?

The amateur EE in me thinks it's not possible...