Register for the iXsystems Community to get an ad-free experience
Resource icon

Proper Power Supply Sizing Guidance

jgreco

Resident Grinch
Moderator
Joined
May 29, 2011
Messages
16,481
Startup current of the drives 1,8 A (1,8 * 12 is 21,6 per driver)

Ah, there's a mistake. Startup current is additional power pulled on the 12V rail for spinup. Your drives also take power for normal operations, on the 5V rail, sometimes on the 12V rail too. Some drive manufacturers do actually report peak startup current on the 12V rail as the "startup current", while others expect you to add together the nominal and startup amps. These days, it's hard to find detailed information. Consider a typical hard drive to eat, oh, I dunno, maybe 4-6 watts for idle operation. The problem is that when you multiply that by your ten drives, you could be missing 40-60 watts in your calculation.

processor AMD EPYC 3301 TDP - 35 watt

And there's another, that's a 65 watt TDP part. That doesn't mean it'll actually USE 65 watts, but it's a ballpark approximation.

Anyways, one of the big reasons not to try to "right-size" your PSU to the exact number of watts you think you need is that it adds risk, but oversizing it somewhat is quite forgiving and offers some room for error.
 

Niels80

Cadet
Joined
Feb 6, 2020
Messages
4
Aha! After correcting I have a consumption of 586 watt. A 750 watt psu should be the right choice.
 

Achelon

Cadet
Joined
Jun 7, 2012
Messages
5
This post is playing devils advocate, but it's not intended to be purposely antagonistic. Although this guide is a great way to ensure that users are playing it safe with PSU sizing, I really think that it leads to PSUs that are over-specified and risks wheel-barrowing money to (the admittedly reputable) Seasonic.

The following lines in particular caught my eye:

jgreco said:
"I've seen about 1,000 threads like this one where people decide that they can power a dozen hard drives off a 360 watt supply. DO NOT DO THIS."
...
"For an E3-1230v3 (32-98W board+CPU, 12W memory):
...
9-10 Drives: 510W peak, 174W idle -> SeaSonic G-650 or X-650"


I have two servers: each with 32GB of DDR3 RAM, a 10gig NIC, a 16 port SAS HBA and 10 HDDs - a mixture of genuine 'Enterprise' drives (Seagate Exos SAS) and some consumer ones (Ironwolf). Both servers have a 300W PSU (Seasonic SSP-300-SUG). The aim here isn't to brag about "what I managed to get away with", but to thoughtfully show with a real-life example that - contrary to what you might think - a dozen drives on a 360W supply might genuinely be an appropriate design choice and people probably don't need as large a PSU as they think.

Breaking down the power consumption of one server:

Screenshot 2021-06-17 at 05.44.45.png


Given this server gets relatively light usage, true power demand likely fluctuates between the total in the 'Idle' column and the total in the 'Typical' column, though under rare conditions may approach 'Max'. This would place the upper and lower limits of load on the PSU during normal operation within the optimum range of the PSU's efficiency curve. The shaded area green shows the area between 'Idle' and 'Max'.

It is true that startup power demand would appear to be at or just beyond the PSU's limit, but...


1) These are worst case figures - even without staggered spin-up it would probably be okay. If it's a concern, why not use a 350W or 400W PSU? Why recommend a 650W one?
2) I'm slightly aghast at the idea that:
jgreco said:
The rule of thumb in the shop here is that a power supply should never be pushed beyond 80% of its rated capacity
This would mean we have to derate every advertised PSU wattage by 20%. In the case of the Seasonic SSP-300-SUG, the specification permits continuous operation at 100% load with permissible peak loads of up to 110% (330W) of maximum for a duration of 1 second. Overpower protection doesn't even trip until 135% (405W) of maximum load.

The guide talks about derating, but according to specifications for this PSU, for a 20% derating to occur due to temperature, the maximum permissible operating temperature (40C) would have to be exceeded by 25% which seems unlikely given the PSU's active cooling and a 'normal' environment.

I realise theres an element of so what? Who cares if it costs a bit too much, don't you prefer certainty and safety? But if the aim by over-specifying is to ensure design safety margin or for peace of mind - be mindful you can spend as much as you want on those sort of things!
 

jgreco

Resident Grinch
Moderator
Joined
May 29, 2011
Messages
16,481
This is all explained in great detail upstream.

So if you're going to spend thousands of dollars on a system and the hard drives to go into it, basically a PSU failure that cooks your system is a very expensive mistake. It is not a "wheelbarrow" of money to move from a 400W PSU to a 650W PSU, it looks to be about in the range of $25. Seasonic 650's are on sale at NewEgg for $105.

The tested ratings for a PSU are for a brand-new PSU, yet a NAS unit is likely to run for many years, generally much longer than the normal expected 3-5 year lifespan of a desktop PC, ten years or more not being unusual. Derating simply adds a more conservative factor that allows for aging of components, fan stalls, and a variety of other issues. This is not the same thing as the temperature based derating you refer to, by the way.

Relying on staggered spinup or the numbers from today's hard drives to justify smaller sizing is a bit foolish, because a simple firmware update can bork staggered spinup (was that LSI P13?), and the temptation to upgrade hard drives in a few years might result in drives with much hungrier current requirements being added to the machine.

Engineering isn't just about making the numbers work out according to spec sheets. It is about understanding the nuances to a design, and where the risks are, and seeing if you can mitigate those risks without breaking the bank. You don't have to do it. It's like driving around without a seat belt. It will be perfectly fine, until maybe one day it isn't, and you find yourself propelled through the windshield. Even if you are wearing the Grinch's "proper sizing" seat belt, it might not save you, but we know that in many cases it does. To my mind, adding $25 cost to a $2500 machine in order to reduce the likelihood of HDD-damaging brownout is sensible.

I am not answering your post in more depth because everything that you have said, everything that I have said in this answer, and everything else that I might want to say have already been discussed upstream. Answering these tedious "are you serious" messages is ... tedious. Am I serious? Yes, I am serious. No, you don't have to follow the advice. I show the work and the reasoning, and it follows good engineering practices. Do as you wish.
 

Achelon

Cadet
Joined
Jun 7, 2012
Messages
5
This is not the same thing as the temperature based derating you refer to, by the way.

I was well aware of this, but sadly, Seasonic do not publish (publicly at least) derating curves for PSU ageing (only those for temperature), so I refrained from commenting on it in the absence of data to confirm or disprove what I was saying.

I respect your view but I just wanted to put an evidence-based alternate view point down for those who perhaps are more budget conscious - tedious though it evidently is!
 

jgreco

Resident Grinch
Moderator
Joined
May 29, 2011
Messages
16,481
Correct, PSU aging is not generally taken seriously by consumer gear manufacturers who are putting out products targeted at the gaming community, where typical lifecycles are 3-5 years. To figure this out, you have to look more generally at electronics design, where issues such as thermal stresses on capacitors, and other component degradation issues, are already well understood. It is possible to generate these numbers for specific devices, but requires more work than a PC manufacturer is likely to do. They mainly want their gear to reach the end of warranty without becoming an RMA. The problem is that a NAS could (and should) have a much longer lifespan than your typical several-year warranty.

One of the plusses to Seasonic are a history of good build quality and an unusually long warranty, which hints that they've done a better job at component selection and design than many others. This doesn't mean that it's smart to stress the components.

I just wanted to put an evidence-based alternate view point down for those who perhaps are more budget conscious

Well, feel free to post your evidence. Finding credible evidence for this kind of thing is really difficult, as manufacturers generally do not release data on failure rates, and even if they did, it's really hard to back-analyze that to see what conditions the PSU was subjected to that might have led to the failure, so I don't have any idea what sort of evidence you might have. If your evidence is just "I built my system and look it worked", well, yeah, I'm sure it did, but that's not really evidence of much. The goal in this thread is to reduce the chances of damaging an expensive computing and storage system through undersized PSU's. We already know that undersizing PSU's isn't a guaranteed system killer, which is all that I'm really getting out of your messages to this point. The point is that properly sizing a PSU through good engineering principles can reduce the risk of catastrophic failure at some point from, who knows what the actual numbers are, let's say a 2% chance to a 1% chance, and the incremental cost isn't that large.
 

Achelon

Cadet
Joined
Jun 7, 2012
Messages
5
I guess the point I'm making is that despite intuition telling you that a 300W PSU is unsuitable for a NAS with 10 SAS HDD, a 10gig NIC and a 16 port SAS HBA, it arguably isn't undersized for that sort of application at all. The evidence here being not just the anecdotal experience but the specification sheets that show that in the example i gave, under the conditions that the equipment will sit at for >99% of its life, the load on the 300W PSU will be at or close to the peak of the efficiency curve in this application - i.e. the conditions the PSU is optimised for.

While it is true that there may be mission critical enterprise servers that require added robustness through (for example) redundant PSUs, in most cases (and certainly the "home NAS" type setups that a lot people on these forums are doing) it's hard for me to see what is gained by over-speccing the PSU. At best it saves you time from doing detailed right-sizing calculations, a worst it seems to be to simply offer some vague sense of peace of mind.
 

jgreco

Resident Grinch
Moderator
Joined
May 29, 2011
Messages
16,481
Mission critical enterprise servers are an entirely different ballgame, and the manufacturers of such have already taken care of these issues in the design of the overall platform, which they have a lot more control over, ranging from firmware guaranteed to do things like staggered spinup, to the selection of hard drives that behave predictably, to oversizing of PSU's through the use of redundant power supplies. End users are not placed in the position of having to do the engineering work for this, it's guys like me with an education in electrical engineering who end up doing it during the design of the server.

This thread is showing end users who do not do this stuff professionally how and why this stuff is calculated, and why the numbers are larger than you might expect if you wish to have a robust design.

You do not need to believe this, I can't force you to believe this, but what you posted isn't any sort of evidence that what you are advocating is a good idea. You haven't analyzed what causes failures or what the effects of a failure are. The measure for this kind of thing isn't that it's alright "for >99% of its life"; you can take an automotive engine that should be good for a quarter million miles, run it at redline for a few minutes now and then, and still cause damage to it in fairly short order. Would you really do this to an engine?

In any case, it is clear that you are just rehashing issues that are thoroughly discussed upthread, and this is basically pointless. Feel free to do as you wish. At least you've actually researched some numbers, which puts you ahead of the numskulls who have taken a reading of what their server consumes with a Kill-a-Watt and then sized their PSU to that.
 

sniper29a

Cadet
Joined
Jan 13, 2022
Messages
5
35 HDD, E3-1275 v3, 32GB RAM Supermicro MB. Start/spin-up 550W peak, reboot 400W, normal usage 250-280W.

Mix of 14-16TB modern HDDs + 8x old HDDs 8-12TB various crap like survey lance, archive drives.
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,515
One of the key issues is testing.... with a small sample size, comes greater risk. One can likely use a smaller power supply than recommended by jgreco and be potentially fine, but the devil is in the details. For example, the Mini XL is fitted with a 250W power supply (Model # FSP250-50LC) that is apparently good for eight 3.5" drives, a low power mini-ITX board, a PCIe card, and a few SSDs and fans sprinkled throughout the case. I run about the same number of drives in my tower at about 126W with a slightly more power hungry processor.

I have no doubt that iXsystems tested the daylights out of this rig, confirming many years of faithful service before putting it on the market. But that's the thing - they qualified that PSU for use with that case, a specific mini-ITX board, and that loadout of drives and cards. Your use case may be different. Interestingly, the PSU has a sticker with the allowed amp loads by voltage bus, with 18A being allowed on the +12VDC bus and up to 80W getting shared between 3.3V and 5.5V. That's a uncommon level of detail. Most PSUs are scant on details re: individual voltage bus capabilities and that may bite the average user in the bum.

So that's why oversizing is not as bad as it seems. I'll also mention that getting a sub-500W platinum ATX PSU was pretty much impossible when I was shopping for one. The smallest ones are around 550W unless you want to get into non-ATX form factors. That said, some of the OEM PSUs for servers can be really good deals ($ spent vs. capacity, efficiency, quality, etc.) if you can make them fit. A prime example were the SuperMicro PSUs for their servers which were really inexpensive compared to the gaming rig stuff out there the last time I checked.
 

jgreco

Resident Grinch
Moderator
Joined
May 29, 2011
Messages
16,481
For example, the Mini XL is fitted with a 250W power supply (Model # FSP250-50LC) that is apparently good for eight 3.5" drives, a low power mini-ITX board, a PCIe card, and a few SSDs and fans sprinkled throughout the case.

Never seen a Mini XL, but I'm guessing that this also involves some configuration such as staggered spin-up to support it. Some years ago, I had AsRock Rack send me one of their 1U12LW-C2750's for evaluation, and I believe it also came with a 250W PSU.

This is probably going to be fine at least for awhile, but just like the Microserver N36L/N40L/N54L's, there will be a healthy demand for replacement PSU's after maybe five years.

There's also the risk of someone doing something stupid like spinning down the drives, and when ZFS wakes them up, ... all at once ..., the smoke comes billowing out.
 

Constantin

Vampire Pig
Joined
May 19, 2017
Messages
1,515
That's thing... quality power supplies aren't going to have issues with running at part load. If designed well, they will even run quite efficiently at loads as low as 10%. That's the bulk of the difference between lower-efficiency and higher-efficiency / Platinum / Titanium / etc. power supplies. But the higher efficiency associated with a Titanium Power supply usually also comes bundled with better component selection, more expensive designs, and better overall performance across all PSU categories.... except for the hit on your wallet.

Usually, when it comes to ATX power supplies, the marginal cost of a Titanium PSU over silver or gold is hard to make up with energy efficiency savings alone. The are only a few scenarios I can think of, such as off-grid operation, remote island power prices ($0.5/kW and up) and like use cases. Over at Newegg right now, there is not a single Titanium PSU below 600W capacity. Over at Amazon, the situation is no different.

Switching to non-ATX PSUs for server use, at wiredzone, the smallest Titanium PSU I was able to find was a 1000W unit. For Platinum Plus, the choices at WiredZone expand significantly, with units starting around 340W.

FWIW, the 250W PSU in the Mini XL is a bronze 80, so it's "value-engineered." It's also absolutely tiny and has a small screaming fan to let you know it's on. :smile:
 
Last edited:

Ericloewe

Not-very-passive-but-aggressive
Moderator
Joined
Feb 15, 2014
Messages
18,088
Titanium is pretty standard on servers these days. Between the lower cost sensitivity, the significant desire to get power down to meet rack/DC power limits and just plain upselling, the not-bottom-of-the-barrel stuff is Platinum at a minimum and Titanium very often.
 

rvassar

Guru
Joined
May 2, 2018
Messages
862
Those whole "Metal" efficiency rating thing has gotten absurd. Titanium is rated higher than gold? How is that intuitive?

Is the CIA tricking the Russians and exporting paint pigment via shell companies to build spy planes again? :smile:
 

danb35

Hall of Famer
Joined
Aug 16, 2011
Messages
13,538

Etorix

Guru
Joined
Dec 30, 2020
Messages
1,010
…but marketing departments are run by Black Elves, aren't they?
 

Chris Moore

Hall of Famer
Joined
May 2, 2015
Messages
10,085
35 HDD, E3-1275 v3, 32GB RAM Supermicro MB. Start/spin-up 550W peak, reboot 400W, normal usage 250-280W.

Mix of 14-16TB modern HDDs + 8x old HDDs 8-12TB various crap like survey lance, archive drives.
Is this meant to be a question? You have told us what the power consumption of the system is, I can only guess that is based on readings from some test equipment, but you have not told us what power supply you are using, or anything about why you made the post.
 
Top