Power Supply A Guide to 12V PSU

[Valantar]

Do you experts think the Meanwell EPS-65S-12 would be suitable for an APU build (e.g. Athlon 200GE-based)? This thing is even smaller than the EPP/RPS 100 and 200 Watt models!

https://www.meanwell-web.com/en-gb/ac-dc-single-output-open-frame-power-supply-output-eps--65s--12

How would the PC be used? I'd say it's a stretch, but it should work as long as you're conscious of the risk involved. The ripple/noise spec of that unit is right on the ATX spec (120mV p-p), but according to the test report real-world ripple is below 50mV p-p, which is perfectly fine. With a 65W max output you'd need to be very careful of the parts you put into this build - a 35W APU can spike up higher during short loads, and at this level every watt from other components will count. If you go for 1.2V DDR4, a low-power SSD (not a high-end NVMe drive) and nothing else you ought to be fine. 35W from the APU + 5-ish from an SSD and around 10 from the RAM and motherboard places you at 50W, but that number assumes that all these will be consuming their full power at all times, which isn't really the case. But again, the APU might spike noticeably higher. BIOS power limits would definitely be worth looking into.

I've got an Intel J1900 board (10W, mind you, not 35) that I've run off a 12V3A (36W) power brick from an external 3.5" HDD - so it's definitely possible with the right hardware. If I was you, I'd give it a try while constantly monitoring system power draw with a kill-a-watt-style power meter, and declare it okay if total system power under a heavy load stays below 60. Remember that whole-system power includes PSU efficiency losses, which with the EPS-65S-12 should be about 12% - so a 60W reading for the whole system places system power draw at around 54W or slightly below. If the system pulls that much under a steady load, I don't think I would quite trust it, but if total system power is at 50 or so, it'll probably be fine for a few years.

 

[Choidebu]

While ripple wise that EPS unit is fine, I'd just like to add that:

1. It's not a power factor corrected unit, and it shows on the efficiency. It's not very efficient. A good model would have >90%. These have around 85%.
2. It doesn't have ground. Only Live and Neutral goes in.

However for 65W application both these points should not be a hindrance to you whatsoever.

But if you only need 65W, why bother with meanwells? Can't you just get a brick? In my area you can get one for like 8$ used. 65W is THE common size for mainstream laptop bricks.



Edit: @Thehack , I spent some time a while ago looking through most of their lineups, and I believe the models suitable for 12V PC application is RPS, UHP and EPP. If you think this is correct then maybe it is time to update the first post?

 

[Thehack]

While ripple wise that EPS unit is fine, I'd just like to add that:

1. It's not a power factor corrected unit, and it shows on the efficiency. A good model would have >90%. These have around 85%.
2. It doesn't have ground. Only Live and Neutral goes in.

However for 65W application both these points should not be a hindrance to you whatsoever.

But if you only need 65W, why bother with meanwells? Can't you just get a brick? In my area you can get one for like 8$ used. 65W is THE common size for mainstream laptop bricks.



Edit: @Thehack , I spent some time a while ago looking through most of their lineups, and I believe the models suitable for 12V PC application is RPS, UHP and EPP. If you think this is correct then maybe it is time to update the first post?

PFC is not necessary for such a low power unit. Efficiency is not correlated with PFC, in case others may infer from what you said.

But yeah, internal is the current meta. 65w external is cheap and straightforward though.

 

[Choidebu]

Yeah that's what I'm getting at. Just mount the small brick in the case et voila it's internal.

 

[grbh]

Thanks for the replies, everyone! I wasn't aware that the RPS-120S-12 existed. I might as well go for that one with my APU build just in case I want to move to a Ryzen APU later.

 

[Valantar]

Edit: @Thehack , I spent some time a while ago looking through most of their lineups, and I believe the models suitable for 12V PC application is RPS, UHP and EPP. If you think this is correct then maybe it is time to update the first post?

Related question: I've been looking for an AC-DC unit as small as possible (I need it to fit inside my Chieftec IX-01B!) to run an ultra-low-power J1900 build I have lying around. I've previously run it off a 12V3A wall wart from an external HDD, which worked fine even running Prime95+Furmark, so this thing sips power. I've found the MW APV-35-12, which admittedly is an LED driver, but also seems to fit the bill - size, power output, and so on. Efficiency is not exactly great at 83% (but at this wattage I doubt I'll find higher), and it's rated at 150mV p-p - but only tested at 39mV p-p in MW's own report. Is there any reason why this wouldn't work? My plan is to use the PC as a firewall or something similar. There's no way I could fit an off-the-shelf power brick into that case, btw


A secondary question: I'm looking to replace the PSU in my modded Optiplex 990, and the only real option seems to be the MW UHP-350-12 due to the weird dimensions of the original PSU. This has a ripple/noise rating of 200mV, and tested at 129mV in the report. In other words, it's too high. I've been talking to my brother (who is an electrical engineer), and it seems like putting a reasonably sized filter cap across the output might help, but he's warned me that a too big capacitive load on the output might cause the PSU to start oscillating its output and essentially fry everything. Which would be bad. My question is: I obviously plan to do something to lower the ripple, but how low do I need to get it? Would, say, a 20% reduction be enough, down to about 100mV, or is that still bad enough that components might take damage?

 

[Thehack]

Related question: I've been looking for an AC-DC unit as small as possible (I need it to fit inside my Chieftec IX-01B!) to run an ultra-low-power J1900 build I have lying around. I've previously run it off a 12V3A wall wart from an external HDD, which worked fine even running Prime95+Furmark, so this thing sips power. I've found the MW APV-35-12, which admittedly is an LED driver, but also seems to fit the bill - size, power output, and so on. Efficiency is not exactly great at 83% (but at this wattage I doubt I'll find higher), and it's rated at 150mV p-p - but only tested at 39mV p-p in MW's own report. Is there any reason why this wouldn't work? My plan is to use the PC as a firewall or something similar. There's no way I could fit an off-the-shelf power brick into that case, btw


A secondary question: I'm looking to replace the PSU in my modded Optiplex 990, and the only real option seems to be the MW UHP-350-12 due to the weird dimensions of the original PSU. This has a ripple/noise rating of 200mV, and tested at 129mV in the report. In other words, it's too high. I've been talking to my brother (who is an electrical engineer), and it seems like putting a reasonably sized filter cap across the output might help, but he's warned me that a too big capacitive load on the output might cause the PSU to start oscillating its output and essentially fry everything. Which would be bad. My question is: I obviously plan to do something to lower the ripple, but how low do I need to get it? Would, say, a 20% reduction be enough, down to about 100mV, or is that still bad enough that components might take damage?

1. As long as the 12v is a constant voltage. While power is power, I think LED psu have less tolerance and can be more failure prone and likely lack safety features, since they normally only drive dumb leds.

2. I think 129mv is okay since the max is only 9mV lower. I see two gpu so I'm assuming you need something capable of 250w? Why not a flex atx or tfx unit? If you can fit a low profile gpu you can usually fit a flex atx unit. It does look like the stock one is a tfx one.

 

[Valantar]

1. As long as the 12v is a constant voltage. While power is power, I think LED psu have less tolerance and can be more failure prone and likely lack safety features, since they normally only drive dumb leds.

2. I think 129mv is okay since the max is only 9mV lower. I see two gpu so I'm assuming you need something capable of 250w? Why not a flex atx or tfx unit? If you can fit a low profile gpu you can usually fit a flex atx unit. It does look like the stock one is a tfx one.

Let's do 2 first: The Optilex doesn't have an LP GPU, I've stuffed a full-height ITX RX 570 in there It pulls about 260W at the wall when gaming which going by the 65% mean efficiency of the stock PSU translates to something like 170W internally (the 12V rail of the stock PSU maxes our at 17A or 200W). Meaning I want 250+, but 350 would be very nice for future upgrades. The case can't fit FlexATX or TFX unless I replace the motherboard with an ITX board (and modify the case even more), and I don't plan on doing that just yet. It would kind of defeat the purpose of a dumpster-dived build if it got all-new components My initial reaction was the same as you say, that 129 isn't that far above 120 - but then that's the max tolerance in the spec, and good PSUs have 1/5th of that or less, not to mention the reputation poorly filtered PSUs have for killing components over time. The time consuming solution would be to ship this directly to my brother and have him do a ripple test with an added filter cap on an oscilloscope, but I don't know if I have the patience for that right now. Maybe I'll come around to the thought after sleeping on it.

1: I'll give it a try then, I guess. The worst that can happen is that I brick a PC that has minimal value anyhow. And I might gain a very nicely compact firewall

 

[SilverJS]

Has there been more progress, or even a guide, on how to build a load switch for dual PSU solutions?

 

[k0n]

Meanwell sent me the test report of the EPP-500-12.

Ripple is pretty high at 187 mVp-p.

Why am I reading this only hours after ordering it... :')

Fell in love with the form factor and ordered parts before looking through this thread.

My intention is to power the CPU through the picoPSU 160XT (4Pin) and the graphics card (5700 XT red dragon/pulse) directly from the EPP-500-12.

Case could be as small as 350*260*55mm ~5L.... so how bad is it?

 

[smitty2k1]

Why am I reading this only hours after ordering it... :')

Fell in love with the form factor and ordered parts before looking through this thread.

My intention is to power the CPU through the picoPSU 160XT (4Pin) and the graphics card (5700 XT red dragon/pulse) directly from the EPP-500-12.

Case could be as small as 350*260*55mm ~5L.... so how bad is it?

I think you will be fine, but I'm pretty sure you want a load switch like the J-Hack Distro.

 

[Valantar]

I think you will be fine, but I'm pretty sure you want a load switch like the J-Hack Distro.

Ripple 50% above the maximum ATX spec is hardly "fine". That's high enough that it is quite likely to cause instability even at stock speeds for modern hardware (with GPU boost 3 and similar clock-boosting tech), and might very well cause premature hardware failures over time. Overclocking or undervolting will in all likelihood be entirely out of the question. It might work perfectly, but that is a big chance to take. At the very least I'd pick all hardware based on having VRMs with plenty of filtering caps, but some sort of filtering circuit (or even just an added filter cap across the output) would be a very welcome addition. Remember, even middling quality ATX/SFX PSUs usually have 12V ripple numbers below 50mVp-p, and good units reduce that by half or more.

I'm looking into the effect of adding a filter cap across the output of my UHP-350-12, but I'm having trouble finding a suitable dummy load for testing (tried 12V car light bulbs, but apparently a resistive load like that means there's barely any ripple to begin with).

 

[Thehack]

Ripple 50% above the maximum ATX spec is hardly "fine". That's high enough that it is quite likely to cause instability even at stock speeds for modern hardware (with GPU boost 3 and similar clock-boosting tech), and might very well cause premature hardware failures over time. Overclocking or undervolting will in all likelihood be entirely out of the question. It might work perfectly, but that is a big chance to take. At the very least I'd pick all hardware based on having VRMs with plenty of filtering caps, but some sort of filtering circuit (or even just an added filter cap across the output) would be a very welcome addition. Remember, even middling quality ATX/SFX PSUs usually have 12V ripple numbers below 50mVp-p, and good units reduce that by half or more.

I'm looking into the effect of adding a filter cap across the output of my UHP-350-12, but I'm having trouble finding a suitable dummy load for testing (tried 12V car light bulbs, but apparently a resistive load like that means there's barely any ripple to begin with).

It's debatable the effect of ripple noise on modern VRM circuits. Modern VRM, especially the Nvidia rtx series have a very high frequency and fast feedback loop. All power deliveries are further regulated and filtered, primary rails are 1-1.3V for cpu/gpu. With how precise modern vrm are, I would say it is likely a non-issue as long as the transient response is good.

Middling and high end atx power ripple are very overspecc for what you need. This is because the market is highly competitive and third party reviews are comparing your outputs. Remember, cpu and gpu feeds won't see that ripple because they are regulated down. It's good that the competition drives high quality products but it is very much unnecessary.

I think 120mV-200mV is likely fine, but of course I prefer to stick to the specs to reduce any worries and liability.

 

[Choidebu]

@Valantar


In short, one does not just slap a capacitor and be done with it. At the very least you'd need an LPF (Low Pass Filter). Otherwise people can just slap a zener or two and be done with it, ha!

 

[Kepha]

Hello,

New to the forum and in hope I can have my question answered in this thread and avoid creating a new one for what will be a quick answer for the experienced ones!

I am finishing a small build and using the Silverstone FX350-G. The 24 Pin for the motherboard comes with a 2-Pin +5Vsb connector and I am lost on where and if I need to plug this? I googled for a while and nothing obvious came up.

FX350-G

As a leading brand in small form factor (SFF) case design and power supply miniaturization, SilverStone launched its first Flex ATX power supply, the FX350-G, to further widen the scope of SFF design possibilities and to serve upgrade needs of embedded systems or 1U rack-mount server...

www.silverstonetek.com

The motherboard is a B450 GB Aorus I Pro Wifi. It has a m.2 drive, 1TB HDD, APU with stock CPU cooler and an ElGato HD 60 Pro. nothing too demanding.

Regards
Kepha

 

[Choidebu]

That's way too off topic - make your own thread.

Also, I have no idea what you're talking about. A PSU with 2 pin 5v vsb? Never heard of it. What for? Power good Led? If you can't find anywhere to plug it in then maybe you don't need it, no? Just turn it on and see if it POSTs.

 

[Kepha]

I said that the 24 Pin connector for the motherboard, has a 2-Pin +5Vsb connector, like the image above shows.

 

[Thehack]

I said that the 24 Pin connector for the motherboard, has a 2-Pin +5Vsb connector, like the image above shows.

No motherboard requires it. You can ignore it.

 

[Kepha]

Big thank you @Thehack .

 

[BaK]

Is there any advantage of choosing a long board version like below over a PDCB?

Also, for a general good practice, is it better to choose both the PDCB and the power brick of a same output wattage? Or is one supposed to be of higher wattage than the other for some safety reasons?