Low-Power ATX PSU's - Where to get

[FreeNASBob]

Searching 80+ ATX PSUs here, I find quite a few PSUs that look good for low-power applications but they all appear to be OEM parts for HP, Dell, and Lenovo. The worst part is searching for the part numbers on the OEM Web sites turns up nothing. For example, Lenovo has a 180W PSU rated at over 91% efficiency at 20% load (36W), but I can't find the part for sale anywhere in the US. Dell has a 200W PSU rated for 84% efficiency at 10% load, but same story. It seems like owners of machines with these PSUs would have to be able to purchase replacements, but maybe I don't know the proper channels to locate one.

If anybody has any knowledge about where to secure OEM PSUs it would be appreciated. I already know about picoPSUs and the grossly over-powered PSUs that are the typical fare at NewEgg and other retailers. Even a platinum rated PSU at 450W is going to be wasting almost 50% of the power in a system that will idle around 15 - 20W.

 

[Ericloewe]

The smallest unit I recommend is the Seasonic G-360. It'll be better than almost any OEM PSU, by a mile.

You need to plan for spin up current anyway. How many disks are you going to use?

 

[FreeNASBob]

6 laptop drives. Total spinup current will be under 6A on the 5V rail.

I'm leaning towards the picoPSU solution for now. I was hoping to be able to secure something in the ATX form factor with 80+ in the 15W range and capable of 12A on the 12V rail, but nobody makes such a thing it appears. That way if the laptop drives give me too much trouble I can migrate the WD EFRX drives without buying a new PSU.

 

[Ericloewe]

6 laptop drives. Total spinup current will be under 6A on the 5V rail.

I'm leaning towards the picoPSU solution for now. I was hoping to be able to secure something in the ATX form factor with 80+ in the 15W range and capable of 12A on the 12V rail, but nobody makes such a thing it appears. That way if the laptop drives give me too much trouble I can migrate the WD EFRX drives without buying a new PSU.

150W is certainly not enough for 6 desktop drives at spin up, though. You'd want something closer to 220+W

 

[FreeNASBob]

I guess it depends on the peak power of the PSU. That information isn't always easy to find, but for the ones I do find it seems like it ranges from 1.25-1.5 times the maximum constant load. The 180W Lenovo PSU looks as though it can take over 12A on the 12V rail, which is all I would need.

That being said, a 220-250W titanium PSU would be just what the doctor ordered for NAS-type systems. Just seems that nobody is interested in supplying that market niche.

 

[Ericloewe]

I guess it depends on the peak power of the PSU. That information isn't always easy to find, but for the ones I do find it seems like it ranges from 1.25-1.5 times the maximum constant load. The 180W Lenovo PSU looks as though it can take over 12A on the 12V rail, which is all I would need.

12A is not enough, much less if it's peak power (do not ever trust the peak power rating. PSUs that need it are typically garbage). Each drive (WD Red for instance) requires some 2A, with a 10% margin. After you account for system use (CPU and motherboard, essentially), you arrive at the 220ish Watt figure (assuming a modern PSU design that can output nearly all of its power on 12V and with secondary rails strong enough for it not to be a problem).

 

[FreeNASBob]

12A is not enough, much less if it's peak power (do not ever trust the peak power rating. PSUs that need it are typically garbage). Each drive (WD Red for instance) requires some 2A, with a 10% margin. After you account for system use (CPU and motherboard, essentially), you arrive at the 220ish Watt figure (assuming a modern PSU design that can output nearly all of its power on 12V and with secondary rails strong enough for it not to be a problem).

I suppose it's true that you shouldn't generally trust manufacturers as they will advertise their products in the most advantageous manner possible. A PSU with a 250W rating is probably telling you its peak rating, and not continuous, which would make it prudent to treat the advertised rating as the peak rating.

I still keep my fingers crossed for a Titanium 80+ PSU in the 200 - 250W range. I imagine that it's not likely given that it would take higher quality components and engineering without additional return. It's just a shame that we can't trust PSUs to competently supply their advertised peak power since we only need that ~25W per disk for a few seconds and the rest of the time we're operating at about 10% of that. I guess motherboard manufacturers could do a lot to ameliorate this particular predicament by providing wider support for staggered spin up.

 

[panz]

Please, take into consideration capacitors aging...

 

[jgreco]

6 laptop drives. Total spinup current will be under 6A on the 5V rail.

Well, hooray, someone who did some basic math

But yeah, you're going to have a bit of a time. As noted they just don't make the supplies we'd like for some of these applications.

 

[FreeNASBob]

But yeah, you're going to have a bit of a time. As noted they just don't make the supplies we'd like for some of these applications.

What gets me bummed is that they do make supplies that would be pretty good in these applications, but I can't find them to purchase for the life of me. Dell, HP, and Lenovo all have 80+ PSUs that are in the 250W range, but I can't find them anywhere except for Ebay and I don't trust a PSU from Ebay.

 

[panz]

You're worried about power... My system has an Intel Xeon 1230 V2 ~ Supermicro X9SCM-F ~ 32 GB of Kingston ECC RAM ~ Gooxi RM4024-660-BX 4U Rackmount Chassis ~ Corsair HX650 80 Plus Gold ~ IBM M1015 SAS/SATA Controller card (flashed to IT mode) ~ Intel RES2SV240 SAS/SATA Expander.

Without hard disks it draws about 35W... :) and the PSU is an overkill 650Watts...

 

[FreeNASBob]

Without hard disks it draws about 35W... :) and the PSU is an overkill 650Watts...

So about 53W with 6 WD Reds. That PSU has a good efficiency in your power range. I'm aiming much lower (like less than 20W idle). A 24x7 53W appliance would increase my electricity bill by almost 13%. That's saying something considering that includes charging an electric vehicle every day.

 

[panz]

Please, consider the "capacitors aging" problem: if you choose a PSU that fits "tight" to your needs, the aging of the capacitors is going to lower dramatically the efficiency of the power supply in a few years.

My system could run with a 450W PSU, but if you calculate the aging, you'll see that you are going to need to replace it soon. A 650W PSU hasn't this problem (in my situation).

 

[mjws00]

The efficiency curves on most of these power supplies is pretty flat. Especially on the good ones. There isn't going to be much difference between a 400w rated supply at your ~20w load and a 20ow rated supply at your ~20w load. If both power supplies are high efficiency units, they are going to do as well as commodity hardware allows. The larger unit may even be better. You can consider that you may not actually be putting enough load on the psu for it to be happy... but most quality designs will do fine.

You are right that they aren't going to do many titanium 80+ psu in small sizes. There is no demand as the benefits when sipping this tiny bit of power will never pay for the additional costs of the unit.

The setup shown by panz shows this pretty clearly. But it is simple to test for yourself with spare gear. Borrow something or just take your wattmeter to a friends house. An idling system ~20w will draw very very similar input current from different sized power supplies of high quality. Possibly small enough to be difficult to test on consumer gear. If you are intent on going absolutely nuts over .1 watts look into a custom solution. There are some uber high efficiency dc-dc rigs that could be run with a high efficiency ac-dc supply that might gain you a few % at the cost of a lot of messing around and expense. You end up with closer to a laptop or nuc style power supply. You may also be forced to use laptop drives etc, as is your intent.

A different OS with solid WOL and aggressive power management functions might be a lot easier and more efficient overall.

To put this in perspective. If you watch 1 or 2 extra movies in a year... you probably undo all the gains you'll see from all this hassle and expense.

 

[FreeNASBob]

The efficiency curves on most of these power supplies is pretty flat. Especially on the good ones. There isn't going to be much difference between a 400w rated supply at your ~20w load and a 20ow rated supply at your ~20w load. If both power supplies are high efficiency units, they are going to do as well as commodity hardware allows. The larger unit may even be better. You can consider that you may not actually be putting enough load on the psu for it to be happy... but most quality designs will do fine.

They're relatively flat, except that 10% for a 600W PSU is way different than 10% for a 200W PSU.

Look at the difference between the high-power Seasonics and the HP 200W in the area around 20W. If you follow the curves downward the Seasonics are going to come in somewhere below 80% efficient at 20W while the HP is 89%. The absolute difference in wattage is small, but those number add up over time. Just a 5W difference is 44 kWh/year. That's enough power for me to drive about 330 miles.

You are right that they aren't going to do many titanium 80+ psu in small sizes. There is no demand as the benefits when sipping this tiny bit of power will never pay for the additional costs of the unit.

Unfortunately this is true right now since a lot of the costs of electrical generation are not paid by the consumers of that electricity.

To put this in perspective. If you watch 1 or 2 extra movies in a year... you probably undo all the gains you'll see from all this hassle and expense.

That's not possible. If I didn't go through this hassle and expense and watched those movies I'd still have used that much more power. Shaving off a few percentage points here and there seems like an exercise in futility when you look at each individual case, but the incremental effects are large. That's how I'm able to have central air conditioning, charge an electric vehicle, run routers, big screen televisions, and computers while still using less than 1/3 of the electricity most households use. When it comes to water the difference is even more. My household uses less than 8% of our baseline allocation, using in one year what most households use in a month.

Part of the enjoyment of engineering things is engineering them to be as efficient as possible ... and succeeding. Anybody can make a car that can drive 100 miles on a tank of gas, but how many can make a car that can drive 100 miles on one gallon? Achieving the former isn't nearly as satisfactory as achieving the latter ... at least to me.

 

[cyberjock]

Unfortunately this is true right now since a lot of the costs of electrical generation are not paid by the consumers of that electricity.

Actually, this isn't true. The real problem is the cost of the additional efficiency (because of the cost of higher quality components, designs, electricity involved with creation of the PSU, etc.) are NOT offset by the economy of the electrical cost. If electricity was suddenly $2/kWh tomorrow you'd see returns very rapidly. Of course, the increase in electricity costs associated with the actual production of the PSU will probably raise the prices of the PSU themselves.

For comparison, if you could go with a 100% efficient PSU using superconductors and such you'd definitely see gains due to the higher efficiency. But you'll find the gains will never outweigh the losses.

 

[FreeNASBob]

Actually, this isn't true. The real problem is the cost of the additional efficiency (because of the cost of higher quality components, designs, electricity involved with creation of the PSU, etc.) are NOT offset by the economy of the electrical cost.

That's what I mean. I mean generating a kWh of electricity costs society far more than consumers pay for a kWh of electricity. The taxpayers pick up some of the extra cost in the form of subsidies, property owners pick up some of the costs in the form of additional insurance premiums, and consumers pick up some of the extra costs in increased health insurance premiums no matter whether they use a lot of power or not. The reason it doesn't make economic sense to worry about a few percent is because you're not paying for the costs of those few percent. If consumers had to pay the full costs to society of their power, the economics would change quickly.

 

[mjws00]

I've cut weight racing, and juggled efficiency numbers in high powered electrical systems for a long time. I know how to hunt for every gram and watt. Spent silly amounts of time and money for incremental gains.

I admire your intent, but shy of custom hardware your toolset is limited. A current market can not exist for what you want. However it does address the largest part of the puzzle which is what happens when the load rises. It isn't perfect, but gives you a chance to cut overall kWh. So you choose the best option.

If minimizing watts used is super crazy important. Shut your system down and unplug it. If you can concede that something can idle at all. Why use a freight train like zfs/freenas? Why aren't you considering a lighter platform? Current laptops are crazy good at managing power. What do you REALLY need? Can a tablet do it? What about a modern cell? We can do 128gb of storage, sip power and will stream HD until the cows come home. Lots of off the shelf NAS's are aggressive at managing power, and WOL nicely. You are using a server OS designed to crank 24/7 never sleep, never rest, just burn watts. Seems like a poor fit?

It's a fun discussion. I hope you find a good answer. I think I could hit 95%+ pretty easily buying an atx dc-dc board and combining with a high end ac-dc unit.

 

[FreeNASBob]

If minimizing watts used is super crazy important. Shut your system down and unplug it. If you can concede that something can idle at all. Why use a freight train like zfs/freenas? Why aren't you considering a lighter platform? Current laptops are crazy good at managing power. What do you REALLY need? Can a tablet do it? What about a modern cell? We can do 128gb of storage, sip power and will stream HD until the cows come home. Lots of off the shelf NAS's are aggressive at managing power, and WOL nicely. You are using a server OS designed to crank 24/7 never sleep, never rest, just burn watts. Seems like a poor fit?

It's a fun discussion. I hope you find a good answer. I think I could hit 95%+ pretty easily buying an atx dc-dc board and combining with a high end ac-dc unit.

Most of this is simply because apparently FreeNAS removed the ability to spin down idle drives this year. For all the prior versions it seems FreeNAS was able to offer its feature set without requiring the late shift to stay on at the local coal plant. Since one of the largest opportunities for power saving in an NAS system is no longer an option with FreeNAS, the only alternative is to look towards hyper-efficient PSUs, motherboards, and disk drives to make up for the political decisions of the FreeNAS team. Don't get me wrong, it's completely within their right to make that decision. I'm just explaining why the focus is on eking out a watt here and there when we used to be able to save tens of watts just by having the drives spin down when nobody was using them. If you want the data security features of ZFS and the features of FreeNAS, you have to take the good with the bad. You used to be able to have both, but now you can't.

 

[cyberjock]

We didn't "remove" the ability. FreeNAS really never had it. People hacked it in.