Proper Power Supply Sizing Guidance

[Craigk19]

I am using a U-NAS 810A for my setup, I would recommend you to use something else.

It is a flimsy case, allowing drives to vibrate. The way the mainboard is mounted on 4 protruding lips, requiring a plastic sheet on the back to prevent shorting the back is very dubious. And the PCIe riser card that comes with it should NOT be used because it doesn't use impedance controlled transmission lines for its lanes.

I'm still hoping, that one of these days, Ablecom will grant me the privilege of purchasing a CS-T80 8-bay NAS case from them. No luck so far, but I keep on hoping.

Thank you very much for the heads up because that case and PSU was going to be a $$ purchase. I reached out to silver stone about the CSI381 and this is what I got. "Unfortunately the project is in a delay but it is still open and we are very close to bringing it to production, we should see it in the market this year but Unfortunately I don’t have a firm release date to share at this moment. We believe it will be available around Q3."

I want a small form factor case hot swap if possible which leaves the node 804 which is out of stock and overpriced right now. I currently have a fractal r6 but it is massive.

 

[Constantin]

The IXsystems Mini XL case is nicely built and allows up to 8 hot-swap drives. The C2750D4I sips power yet offers 13 SATA drive connectors.

My experience re: AsRock board failures with iXsystems was absolutely top-notch, a great team to work with. I just got my latest 1.03 revision board from them as a RMA and with that hope that all Intel, watchdog, power supply rail, etc. failures are behind me.

 

[ornias]

@jgreco Slight addition that might be nice to add:

Cables, Cables, Cables.
Even if you have a nice modulair overpowered single rail PSU, people might think it a good idea to use one of those nice breakout cables with multiple PATA connectors to their backplane.
That would work just fine with 6 disks, with 24 disks on 1 breakout cable... that might not work out that nicely.

 

[Constantin]

Agreed. One of the nice features of the Lian Li cases is that they have optional backplanes for the hard drives to slide into. By default, those backplanes are not populated with capacitors but if you’re persistent, you can create a local power reservoir (by clearing extant but unpopulated holes and pads of solder and soldering in capacitors). Some power breakout cables also feature built-in capacitors to help reduce dips in voltage as drives are added or subtracted from an array.

The disks in my pool can be hot swapped without issues even though I prefer to shut the unit down for swaps. That’s likely a function of the power supply (a platinum rated unit nominally operating at 1/5 it’s rated output), the low power consumption of the drives, as well as the local capacitors (shared between 2-3 drives per backplane board). It's all a far world away from some of the VDC power wiring I've had to deal with (snaking 4/0 AWG bus wires through the bowels of a boat, for example).

 

[ornias]

Agreed. One of the nice features of the Lian Li cases is that they have optional backplanes for the hard drives to slide into.

I'm always a fan of case manufacturers that often that option :)
Sadly enough, most cases now seem to move their addon budget to more expansive glass and watercooling hubs

By default, those backplanes are not populated with capacitors but if you’re persistent, you can create a local power reservoir (by clearing extant but unpopulated holes and pads of solder and soldering in capacitors).

Hmm, I think this deserves it's own article almost... differentiating between different feature sets on different backplanes (power delivery, pin 11 support, integrated extender etc)

Some power breakout cables also feature built-in capacitors to help reduce dips in voltage as drives are added or subtracted from an array.

Cool never knew! any examples?

The disks in my pool can be hot swapped without issues even though I prefer to shut the unit down for swaps. That’s likely a function of the power supply (a platinum rated unit nominally operating at 1/5 it’s rated output), the low power consumption of the drives, as well as the local capacitors (shared between 2-3 drives per backplane board).

Quality of the PSU also maters ofc, with some PSU's the voltage stability really suffers when peak-loads hit. But Thats no issue with the average platinum or titanium rated PSU ofc.

It's all a far world away from some of the VDC power wiring I've had to deal with (snaking 4/0 AWG bus wires through the bowels of a boat, for example).

Sounds a lot like pulling wires through pre-war european houses... which are often expanded by amateurs 4 times already... :')

 

[Constantin]

Hmm, I think this deserves it's own article almost... differentiating between different feature sets on different backplanes (power delivery, pin 11 support, integrated extender etc)

The Lian Li Backplanes I have here are very simple. Just SATA and 5 & 12VDC. There are pads for two SMD caps (I put in small ceramic units for high-frequency noise) as well as through-holes for Aluminum electrolytic capacitors. For a description of the pads, see here.

Cool never knew! any examples?

The Silverstone CP06-E4 is one, I am sure there are others? They also offer a two-outlet model.

Quality of the PSU also maters ofc, with some PSU's the voltage stability really suffers when peak-loads hit. But Thats no issue with the average platinum or titanium rated PSU ofc.

Usually, it depends in part on the type of load, how big the load is relative to the ampacity of the cable, and so on. In computer cases the runs tend to be short and the loads quite small vs. the ampacity of the cable, so that helps. It's not like the VAX 750 / PDP 11 in high school that required a dedicated transformer on a pad outside because of the hard drive!

Sounds a lot like pulling wires through pre-war European houses... which are often expanded by amateurs 4 times already... :')

Joy joy, especially if you have solid walls. Thankfully, those aren't typical in pre-war homes!

Our former apartment in Cologne featured alleged concrete walls, and I say alleged because the builder of that 5-story mixed use building literally kept adding sand to the concrete mix while the workers were trying to mix it. 30 years later they found holes in the facade that extended into the floor systems between floors over 1m deep. Thankfully the place didn't collapse but one "benefit" was that a rotary hammer drill was not necessary to make way for holes.

 

[Sumppi]

I'm planning to extend my existing NAS, and I'm looking for some conformation on my new PSU size.
My current system:
MB: Supermicro X10SL7-F
CPU: Intel® Xeon® Processor E3-1246 v3
RAM: Micron 32GB KIT 4X8GB PC3L-12800E DDR3-1600Mhz
HDD: 8 * 10TB Western Digital white labeled RED-drive (shucked from WD My Book) + 1TB NVME via PCI-E adapted
PSU: Seasonic Focus GX 650W
Fans: number of ARCTIC P14 PWM PST and ARCTIC P12 PWM PST

Planned extension (existing drives will remain in the system):
HDD: 8 * 14TB Western Digital White label RED-drive (Shucked from WD elements)
HBA: LSI 6Gbps SAS HBA LSI 9211-8
I will also be adding some new fans (Artic P12/P14 PWM PST's).

I can get a very nice deal on a Seasonic PRIME Gold 1300W PSU.
The PSU maybe a tad overrated for the system, but is it too large?

 

[Constantin]

I’d suggest you use a cheap power meter to see how the system actually pulls. Some ups’ also report the draw. With 8 drives, my system pulls about 125W max. It would not surprise me if your total system draw is less than 200w even with all the additional drives.

PSU’s usually operate at their most efficient from 40% on up re load capacity. The biggest deltas re: energy efficiency between titanium vs. platinum, etc occur typically in the 0-20% load range. So I’d measure your current draw during boot and typical operation and to be conservative, double whatever number is bigger. Then see if your current 650W unit can handle it. I’ll bet an ice cream sandwich that it will be fine.

 

[jgreco]

I’d suggest you use a cheap power meter to see how the system actually pulls. Some ups’ also report the draw. With 8 drives, my system pulls about 125W max. It would not surprise me if your total system draw is less than 200w even with all the additional drives.

PSU’s usually operate at their most efficient from 40% on up re load capacity. The biggest deltas re: energy efficiency between titanium vs. platinum, etc occur typically in the 0-20% load range. So I’d measure your current draw during boot and typical operation and to be conservative, double whatever number is bigger. Then see if your current 650W unit can handle it. I’ll bet an ice cream sandwich that it will be fine.

This is definitely not the thing to do, and I would thank you not to post this kind of thing in my thread on how to properly size power supplies. The only reason I'm not raining flaming doom on you is you did say "double whatever number is bigger", which at least puts us into what is likely to be a not-actively-dangerous sizing realm.

And while I do appreciate the "double the observed" strategy, the genesis for this thread was someone who thought very dangerous things about sizing power supplies, and actively argued for undersized supplies based on ridiculous theories that PSU's could be relied on to handle momentary overloads. "Double the observed" too easily becomes "just use the observed" to someone quickly skimming your post.

Since PSU's tend to be a bit flaky when pushed to their spec, and get flakier as they age, I am real big on the "do the homework" strategy.

 

[Constantin]

Jgreco, so am I, hence my “measure the actual draw” suggestion. If it turns out that the maximum observed power draw even with 16 drives never exceeds 40% of the rated PSU capacity as I expect will be the case here, then the need for a 1000+ Watt PSU is likely obviated.

But for this method to work, one has to have an accurate-enough power measurement device that captures the maximum power draw under a variety of stress tests (such as drive spin-up, running a CPU-intensive workload, etc).

So, for the benefit of the skimmers out there: @jgreco’s method for sizing power supplies is the preferable/safer one. Deviate at your risk and only with ample, accurate-enough data on the maximum power draw of your system.

Plus, if/when your use case changes, that also has to be considered. Switching from helium-filled to air-filled drives will increase power draw, as will transcoding, folding proteins, adding VMs or new toys inside the case, etc. For example, I need to change my system sig since the addition of BlueIris in a VM, four SSDs, etc. increased the average plug load from ~90W to 120W+. The maximum plug load of this system has yet to crack 25% of its PSU capacity, however.

 

[rogerh]

ISTR actually measuring the start-up surge on some hard disks, for this or a similar thread. It is good to avoid voltage droop, or even PSU protection being triggered, during start-up.

 

[Constantin]

Agreed. Per Farmerpling2’s resource, official maximum power draws for 10TB+ helium-filled CMR, 7,200 RPM hard drives are hard to find.

I only see two entries suggesting that 26W is a good starting point for such drives, assuming the published data is accurate and that all vendors have similar power draws for similar drives.

Thankfully, my system staggers drive startups (I can hear that) and the plug power draw never reaches 200W, which is what it should exceed easily if all HDDs spun up at once. Droop can be a real problem, especially in low voltage systems with skinny wires.

 

[jgreco]

Thankfully, my system staggers drive startups (I can hear that) and the plug power draw never reaches 200W, which is what it should exceed easily if all HDDs spun up at once.

I know that *you* know what you're doing, but I feel compelled to mention that relying on staggered spinup will kill the guy who decides to turn on idle spindown, and the array gets accessed and spins up all the drives rapidly.

 

[Ericloewe]

Or if/when the setting gets forgotten in an SAS controller firmware upgrade.

ISTR actually measuring the start-up surge on some hard disks, for this or a similar thread. It is good to avoid voltage droop, or even PSU protection being triggered, during start-up.

Yeah, there was, with nice scope shots for several disks. I think the record was close to 3A.

It's actually a fairly tricky measurement unless you have a proper current probe (rather expensive).

 

[danb35]

Yeah, there was, with nice scope shots for several disks.

I know I posted a few, but I don't remember in which thread.

 

[Sumppi]

A little more background information into my PSU selection.
Initially, I calculated that a 850W to 1000W PSU should work well for me. (Mostly I used the values in the guide, except for fans, which according to Arctic, will draw 0.12A).

Unfortunately, the availability for 850/1000W Seasonics is very poor at the moment (´In stock´ estimates are in February).
I could buy off Amazon, but then the price is close to the 1300W Seasonic.

The lack of availability seems to be affecting other quality brands as well.
I do have a couple of good choises at 850W range, mainly Asus ROG Strix 850W which is built by Seasonic.

 

[danb35]

I know I posted a few, but I don't remember in which thread.

Ah, here:

How to measure the drive spin-up peak current

Other method with an oscilloscope (or a DMM with fast peak hold): buy a clamp-on current probe like this one for about $50 US. It runs on a 9V battery, clamps over the wire you want to measure current on, and converts the current into a voltage. The model I ordered has two switch-selectable...

www.truenas.com

and here:

How to measure the drive spin-up peak current

Ok, well, I don't know what's wrong then. The scope input wasn't 50 Ohms terminated?

www.truenas.com

 

[jgreco]

Or if/when the setting gets forgotten in an SAS controller firmware upgrade.

Thank you, that too.

It's actually a fairly tricky measurement unless you have a proper current probe (rather expensive).

It's the oscilloscope that's expensive, not the probe. And some of these new DSO's aren't really that expensive at that.

 

[Constantin]

Since I don’t have access to my lab equipment at the moment, I substituted a Fluke 87 DMM with a i140 current clamp. Measured the max current on 12VDC as well as 5VDC for two drives, a old Hitachi 2TB air-based unit as well as a new Wd120 12TB helium-filled unit. Both are 7,200 RPM.

The fluke allows capturing of max and mins, showing both drives pulling a maximum of about 1.5-1.6A on the 12V bus and 0.5-0.6A on the 5V bus. So that comes out to 22W, which is in pretty close agreement with the two helium 10+TB, 7,200 RPM entries in farmerplings database. Those two were likely started at a lower temperature, increasing the resistance of whatever oil / grease the OEM is using.

Pretty nifty that both drives I tested have about the same power draw and yet one has 6x the capacity of the other. Once the pandemic is over, I’ll try to hook up two yokogawa 210’s in tandem to do precision measurements showing the power curves for each bus superimposed.

that’s another thing worth considering... power supplies are sold on the basis of total output yet they feed multiple voltage busses across multiple outlets. The actual capacity for each plug outlet on the PSU will vary based on the design. Presumably that data is buried somewhere deep in the data sheets.

 

[ornias]

that’s another thing worth considering... power supplies are sold on the basis of total output yet they feed multiple voltage busses across multiple outlets. The actual capacity for each plug outlet on the PSU will vary based on the design.

I think this one needs more emphasis indeed.
It's not just the power rails that are important, its also the plugs (you shouldn't try putting 24 drives on 6xmolex plugfor example) and the sata/molex plugs are not always balanced accross powerrails.

Presumably that data is buried somewhere deep in the data sheets.

Sadly enough, the actual distribution of the 12V rails over the Sata and Molex connectors is often not supplied in any way-shape or form. I would almost always recommend single rail psu's just to be sure. But also be sure you have a mini-arc-welder in your case when using single rail psu's