Need efficient PSU for 24/7/365 on home server

RavinDJ

Supreme [H]ardness
Joined
Apr 9, 2002
Messages
4,440
I'm building a home server and the server will be on 24/7/365. I'd like to spend a little more initially and then save on electricity and potentially avoid explosions from a crappy PSU. I'm looking into a 750Watts. I'll have 6-8 3.5 SATA drives in it.

Any recommendations will be appreciated... and let me know if I should get less or more than 750W. Also, I'm not an expert on the Bronze/Gold/Platinum (I'm assuming Platinum is best) and what 80+ means or what I should look for.

Thanks, guys!!
 
The Wikipedia article covers the various 80-Plus efficiency ratings pretty well. A gold-rated PSU would probably be fine, anything higher may not be worth the cost relative to the actual power saved.

Assuming all this system is going to be used for is file serving (not running virtual host, transcoding video, etc.), a 750 W PSU is way overkill. I'd be looking more in the 4-500 W range. Also, PSUs are generally the most efficient in ~50-80% of their rated range, so you want to get as close to that as possible. And most of the time the system will be sitting idle and using relatively little power.
 
Gold should be more than enough, going higher than that nets you small percentages for large increases of cost.
 
The Wikipedia article covers the various 80-Plus efficiency ratings pretty well. A gold-rated PSU would probably be fine, anything higher may not be worth the cost relative to the actual power saved.

80Plus ratings should not be a major factor in consideration of a power supply. The testing conditions are unrealistic, being in open air at ~73F at all times. The certification system is rife with abuse (OEMs cheating ratings with cherry picked units, applying ratings to rebranded units, fake ratings, etc.) Plug Load Solutions (the standard creator) does nothing to police their own standard in the market, but they'll happily accept your $2000+ checks all day though to test your power supplies.

Jonny Guru does proper testing of power supplies, with complete teardowns so you know exactly what you're getting.

I wish people would stop recommending power supplies based on 80Plus ratings, its a trash standard that helps nobody but PLS get rich.
 
80Plus ratings should not be a major factor in consideration of a power supply. The testing conditions are unrealistic, being in open air at ~73F at all times. The certification system is rife with abuse (OEMs cheating ratings with cherry picked units, applying ratings to rebranded units, fake ratings, etc.) Plug Load Solutions (the standard creator) does nothing to police their own standard in the market, but they'll happily accept your $2000+ checks all day though to test your power supplies.

Jonny Guru does proper testing of power supplies, with complete teardowns so you know exactly what you're getting.

I wish people would stop recommending power supplies based on 80Plus ratings, its a trash standard that helps nobody but PLS get rich.

Got any citations? First I've ever heard anyone claim the cert is broken.
 
Got any citations? First I've ever heard anyone claim the cert is broken.

I remember reading more than a few [H]OCP and Jonnyguru reviews where the PSUs were unable to meet their efficiency rating in real world testing. They've also noted incidents where a sample sent for review performed better than one bought retail. Can't recall off the top of my head, and primarily happened with budget brands (Coolermaster, Logisys, etc).

@OP, if your computer is primarily running as a file server, a 4-500 watt PSU is what you should get. SATA drives idle in the 1-3 watt range, and spun up they use about 10 watts each. A low power CPU with iGPU wouldn't run you more than 50 watts. Motherboard and the minimal cooling necessary is another 40 watts. If it's transcoding videos, that's a different story. You'll be using a more powerful CPU, but even then a 3950x wouldn't be using more than 170 watts.
 
Spun up, not seeking, sata drives consume about 3.5 W a piece as measured by me. That is unless you count 2.5 inch disks

A desktop motherboard, with ram, will idle around 25w unless it is a power saving motherboard.

Expect that a 10 drive system will idle with spun up disks at about 65-75w including secondary controller, and when it seeks it will likely hit 85-95W .

Obviously, if you run the processor at full pelt it is going to consume more, but you’re not doing that

If you want to save power, go with few bigger disks and use the motherboard’s drive controllers.

Also, the bigger the PSU in terms of watts the worse it is running at low output. 400-450w psus are probably what you want

For the record, I have a cooler master V450s (gold) in my nas/esxi box that has been running 24/7 for about 7 years, idle 8 drives spinning is 71w at the wall. Seek is about 85w

(Previously had a sea sonic bronze which idled around 90W)

if you break it down:
32w drives
20-25w motherboard, 5 fans, 32 gig memory, ssd
10w or so for the controller card
 
Last edited:
Got any citations? First I've ever heard anyone claim the cert is broken.

Compare PLS listed 80PLUS units with ones on the market. Under their standard, rebranded units are not allowed to use the PLS 80PLUS logo unless re-tested, but often do anyway. There are also many supplies that list "80PLUS compliant" on the market that have never been tested, as well as similar fake certification ratings like 85PLUS, 90PLUS, etc. The average consumer isn't going to dig into hundreds of pages of PLS testing methodology and large lists of PSUs to validate their unit is listed, they're just going to trust what the sticker says.

Hardware Secrets has reviewed several units in the past that fall on their face when tested in a real world scenario, but pass some 80PLUS rating, making the standard useless. Unless you have an open air bench in a cold room, 80PLUS is worthless.
 
Got any citations? First I've ever heard anyone claim the cert is broken.

It's common knowledge at this point. The the testing is inadequate and they don't go after companies that slap the cert on their boxes regardless of passing testing. You can bet any no name raidmax, coolmax, logisys, etc with an 80+ cert is lying to you.

https://www.tomshardware.com/reviews/80-plus-psu-efficiency,4848.html
https://www.hardwaresecrets.com/power-supplies-with-fake-80-plus-badges/
https://en.wikipedia.org/wiki/80_Plus#Misleading_power_supply_advertising
 
OP, I think you are over thinking this. Your PSU efficiency will probably only change your electric bill a dollar or two a month, unless you are running you server under full load all day. Realistically your server is probably using less power than a traditional light bulb when at idle or light load.

Knowing the specs of your server and intended use would probably be helpful. As Keljian pointed out you probably don't need anything near 750 watts.
 
Let's say in a worst case scenario the computer uses 80 watts on average. A 2% difference in efficiency is just under 2 watts. In more expensive states, electricity is 20 cents per kwh. 2 watts accumulates to 17.5 kwh a year. At 20 cents per kwh, that is $3.5 a year, needing 5.7 years to recoup a $20 cost difference. 2-3% is the approximate efficiency difference between each 80+ level.
 
Let's say in a worst case scenario the computer uses 80 watts on average. A 2% difference in efficiency is just under 2 watts. In more expensive states, electricity is 20 cents per kwh. 2 watts accumulates to 17.5 kwh a year. At 20 cents per kwh, that is $3.5 a year, needing 5.7 years to recoup a $20 cost difference. 2-3% is the approximate efficiency difference between each 80+ level.

yah but certain PSUs are better at lower power, some are particularly bad. This difference can be as big as 30% or more. As stated, I had a seasonic bronze (sII2/450w) that was idling at 90w with the above hardware, changed it to the cooler master and it dropped 19w at idle.
 
Compare PLS listed 80PLUS units with ones on the market. Under their standard, rebranded units are not allowed to use the PLS 80PLUS logo unless re-tested, but often do anyway. There are also many supplies that list "80PLUS compliant" on the market that have never been tested, as well as similar fake certification ratings like 85PLUS, 90PLUS, etc. The average consumer isn't going to dig into hundreds of pages of PLS testing methodology and large lists of PSUs to validate their unit is listed, they're just going to trust what the sticker says.

Hardware Secrets has reviewed several units in the past that fall on their face when tested in a real world scenario, but pass some 80PLUS rating, making the standard useless. Unless you have an open air bench in a cold room, 80PLUS is worthless.

It's common knowledge at this point. The the testing is inadequate and they don't go after companies that slap the cert on their boxes regardless of passing testing. You can bet any no name raidmax, coolmax, logisys, etc with an 80+ cert is lying to you.

https://www.tomshardware.com/reviews/80-plus-psu-efficiency,4848.html
https://www.hardwaresecrets.com/power-supplies-with-fake-80-plus-badges/
https://en.wikipedia.org/wiki/80_Plus#Misleading_power_supply_advertising


There's not much the org can legally do to go after companies that use weasel language like "80-Plus compliant", make up their own fake certs, or improperly use the OEM's cert instead of getting their own (kinda a silly requirement, but whatever). They could maybe try suing for trademark violation or the like. It's not a sign the cert is broken. Bad companies are going to do sketchy shit. No one is going after the USB consortium because of all the cheap non-compliant USB-C cables that are out there for sale. All the more reason to buy from trusted companies and not the cheapest crap or stuff off of AliExpress.

There's certainly areas where the cert could be improved, but I still feel that it's a good guideline when shopping for a new PSU. But like anything else, if one doesn't do their own research they stand to get burned.
 
yah but certain PSUs are better at lower power, some are particularly bad. This difference can be as big as 30% or more. As stated, I had a seasonic bronze (sII2/450w) that was idling at 90w with the above hardware, changed it to the cooler master and it dropped 19w at idle.

Not surprising as that is a much older unit that didn't have DC/DC conversion for the 3.3 and 5 volt rails. All of the more efficient units use DC/DC conversion.
 
In the interest of saving money... which one of these would be better?

The server will have the following specs:
ASUS AM4 TUF Gaming X570-Plus ATX Motherboard
AMD Ryzen 7 3700X Desktop Processor
nVidia GeForce GTX 960 Video Card
64GB HyperX Fury 3200MHz DDR4 CL16 DIMM (4 x 16GB) Memory
1TB Samsung 970 EVO SSD M.2 NVMe SSD
10TB WD Desktop SATA Hard Drive
10TB WD Desktop SATA Hard Drive
8TB WD SATA Hard Drive
8TB WD SATA Hard Drive
Windows Server 2016 Standard

I will use it as a file server and host a few VMs. VMs will be for learning and not mission critical. They will never be under full load.

Also, if you think a different PSU is recommended, I will spend the required $$$.

Thanks again, guys!!
powersupply.jpg
 
Last edited:
Out of the two, the XFX is the better choice. It's a Seasonic PSU that is around 82% efficient at 100 watts. Purchasing a Gold rated PSU will save you about $10-$15 a year, depending on electricity rates where you live. It will take you about 6-10 years to recoup the cost of a new ~650 watt gold PSU, depending on what deal you're able to find.
 
Out of the two, the XFX is the better choice. It's a Seasonic PSU that is around 82% efficient at 100 watts. Purchasing a Gold rated PSU will save you about $10-$15 a year, depending on electricity rates where you live. It will take you about 6-10 years to recoup the cost of a new ~650 watt gold PSU, depending on what deal you're able to find.

Awesome... thanks for the input! I'm going to put the XFX in tonight. I'll post a pic of the finished product :)
 
I remember reading more than a few [H]OCP and Jonnyguru reviews where the PSUs were unable to meet their efficiency rating in real world testing. They've also noted incidents where a sample sent for review performed better than one bought retail. Can't recall off the top of my head, and primarily happened with budget brands (Coolermaster, Logisys, etc).

@OP, if your computer is primarily running as a file server, a 4-500 watt PSU is what you should get. SATA drives idle in the 1-3 watt range, and spun up they use about 10 watts each. A low power CPU with iGPU wouldn't run you more than 50 watts. Motherboard and the minimal cooling necessary is another 40 watts. If it's transcoding videos, that's a different story. You'll be using a more powerful CPU, but even then a 3950x wouldn't be using more than 170 watts.
N xt your going to tell me my file server doesn't need 2 870w power supplies :). J/k of course (not about the power supplies, they are redundant and my server does a bit more than file serving). Lower wattage higher quality tend to give better average power consumptions. I'm sure my 1250w power supply uses a bit more power to idle than my 1k or 850w. Ok, I just realized I may have a problem with overkill PSUs. Don't do what I did, lol.
 
OP, I think you are over thinking this. Your PSU efficiency will probably only change your electric bill a dollar or two a month, unless you are running you server under full load all day. Realistically your server is probably using less power than a traditional light bulb when at idle or light load.

Knowing the specs of your server and intended use would probably be helpful. As Keljian pointed out you probably don't need anything near 750 watts.
Kind of late here, but if you talk about server, then you need to talk about UPS. Forget electric bill, any server would benefit from graceful shutdown or prevent a 1h outage.
I'm on my 3rd mini-itx server, which uses 1x 14TB drive and 2x M.2 NvMs. My current i3-9100 (w/ECC memory) consumes 20W in idle (and spinning), up to 60W in high load. My total (idle w/ switch&router) UPS consumption is 36-40W. And this gives me 2h30m estimate. Raising by 20W (60W load) brings me down to 1h30m. 74W (CPU full load) goes to 1h9m. So yeah....I'm looking at every watt, and low-power supplies.

This is with a Jou-jye FlexATX 250W PSU.
A 120W pico-psu+75W brick brings it down 3W overall (15-40W load).
A Silverstone TFX 300W raises it by 5W.

My conclusions, buy an 80+PSU with lowest rated power (but above your components sum). That way, your actual consumption should be above (or even closer to) the 20% rated minimum. But if no GPU is involved, 200-250W should be enough even for 10 3.5" drives.
PS: my previous server had 4W in just the management engine, the main reason I replaced it with one without it.
 
Back
Top