Proper Power Supply Sizing Guidance

[jgreco]

I'd like to add a detail to the excellent OP:

Approve of the updates to the OP?

 

[Ericloewe]

Approve of the updates to the OP?

Looks good, thanks for the writeup.

 

[Harsesis]

You found a non-rackmount chassis that holds 20 drives? That just strikes me as a bad idea (especially for cooling) but then again I work mostly with rackmount gear, so you can disregard my bias. :)

For a large system, you probably don't want to follow my sizing math outlined above, because it rapidly gets untenable above 12 drives (arguably even at 12 drives). I don't have particular recommendations beyond that point, because even the Seasonic recommendation isn't truly mine, but rather that of many happy forum users plus my eyeballing of a bunch of reviews.

My best guess is that you could very well still be fine with a ~900W quality power supply. We used to sell 950W PSU's that powered 24-drive systems which peaked at 6.6A (792W) but that was back in the days of 250GB SATA disks. More recent arrays are still generally in that ballpark. The difference is that we're actively measuring this stuff as we build systems, and have the tools to do so. I am ... very nervous ... about making recommendations that are not provably reasonable on a mathematical basis. However, I do encourage you to read every word I've written, and then to look at the 45Drives data above (bearing in mind that's drives ONLY), to make some educated guesses as to power requirements. A lot of this business is just doing research, then doing your homework, and understanding the realities.

Its a 19" inch case from Inter-Tech (they come without any power supply). Currecntly I'm using a G-Series power supply in it but now would like to update it to a bigger model properly sized for the full populated system. So specs of the final system would be:
- X10SL7-F (25W + 10W onboard LSI)
- Xeon E3-1230 v3 (80W)
- 4 memory modules (24W)
- 1 additional 8 port LSI controller (10W)
- 6 fans (100W - 175W)
- 20 sata drives (700W)

This would result in 1190 - 1280 Watt. As you said the scaling changes when using more then 12 drives, but what would that mean for a proper size of a PSU? You guessed a ~900W PSU would properly be enough, would you also guess that with the addional information given above? What would be the best choice for that kind of PSU. When looking at the models available from Seasonic (which seems to be the most recommended brand in this forum) there are only 850/860 W models and the next step would be 1050 or 1250 W from the Seasonic X-series.

As a sidenote: The system draws only arround 100-120W in idle at the moment, beeing populated with 13 drives. I would guess the system will run only at 10-20% off the power supplied by a PSU offering over 1000W even fully populated.

 

[jgreco]

Well, the real problem is that you need to start worrying about the possibility that the PSU may actually malfunction if it isn't driving a sufficiently large load. I'd definitely start to be concerned about this as the idle load for a PSU gets below 10% its rating; it would be prudent in such a case to contact the PSU manufacturer and actually ask whether driving a smallish load would represent an out-of-range configuration for the PSU.

Because the numbers all get insane as the increased numbers of drives cause spinup current to become a massively dominant factor in my suggested sizing equation, the only good answer I have is observation and experimentation - both of which imply that you need instrumentation not readily available to the average guy-at-home.

The realistic answer is probably more along the lines of getting a little more daring as you get up past maybe eight drives. The numbers I've generated are extremely conservative and are demonstrably capable of sustaining the system load because of "maths."

I've seen some rather scary things done, such as the AsRock 1U12LW-C2750 which supports 12 drives but comes with a 250W PSU. I think that could be made to work as long as you were maybe carefully configuring your system to do a staggered spinup, but quite frankly I'd want to throw something bigger in there AND STILL do staggered spinup, because my gut feeling is that the thing is probably frying its silicon trying to start the drives. SuperMicro sells a similar system with a dual 400W supply. If we look at just the spinup current, 2 * 12 * 12 = 288W, so I can comprehend how that basically works out fine when you add in drive 5V current and the mainboard and all that.

You can probably scale back and if you wanted a pulling-it-out-of-my-arse guess as to your specific situation, I'd *GUESS* that you are likely to be fine with a nice ~900-1000W PSU. That's based on the observation that the 45Drives guys are pulling 600 watts on the 12V to spin their drives up and 80 watts on the 5; with half that many drives, you're probably not exceeding 400 watts total between the two. Can you run that system on a 650W PSU? I don't know. If I was selling it to a customer, I'd measure the heck out of it and then move on with my life. I'd feel that it is more likely you can safely run that on a 900W supply because I see that kind of stuff being sold by reputable companies like Supermicro. So please feel free to understand that I'm absolutely erring on the side of caution and that I do encourage you to "use your head."

I regret the unhappy answer to your question; it is an unhappy question to which there isn't a truly good answer (besides "get out the meter and measure it.")

 

[Harsesis]

Thank you jgreco for your opinion on that topic, I really appreaciate it! I will overthink my choices now. I never got my handy on staggered spin up and after a quick google search it seems like that can get quite coplicated to get to work properly.

One open question is still: Is the Seasonic X, and Platinum series similary recommended compared to the G series? Should I prefer one of them?

 

[Bidule0hm]

both of which imply that you need instrumentation not readily available to the average guy-at-home.

Hum... in fact if you just want to know the current peak on the 12 V + 5 V rails (because of the very simple circuit we can't see one current at a time but I think it's ok for what we want to know) then all you need is a molex/SATA adapter, 3 resistors, 2 diodes, an Op-Amp, a capacitor and a DMM (even a crappy one at $5 would be OK if you don't already have one) so the total budget should be under $15 which is very reasonable I think :)

I'll draw a schematic if someone is interested ;)

One open question is still: Is the Seasonic X, and Platinum series similary recommended compared to the G series? Should I prefer one of them?

They are better than the G series, so yes, you should prefer them ;)

 

[Harsesis]

Hum... in fact if you just want to know the current peak on the 12 V + 5 V rails (because of the very simple circuit we can't see one current at a time but I think it's ok for what we want to know) then all you need is a molex/SATA adapter, a resistor, a diode, an Op-Amp, a capacitor and a DMM (even a crappy one at $5 would be OK if you don't already have one) so the total budget should be under $15 which is very reasonable I think :)

I'll draw a schematic if someone is interested ;)

For me I'm not to much into hardware (till now) so that is definetly nothing for me :D.

Yhey are better than the G series, so yes, you should prefer them ;)

That answered the first question if they are both recommanded, but I wanted to know if I should prefer eg. Platinum over X-series. But as you said they are both better than G-series I think I will just go for the better price tag.

 

[jgreco]

Hum... in fact if you just want to know the current peak on the 12 V + 5 V rails (because of the very simple circuit we can't see one current at a time but I think it's ok for what we want to know) then all you need is a molex/SATA adapter, 3 resistors, 2 diodes, an Op-Amp, a capacitor and a DMM (even a crappy one at $5 would be OK if you don't already have one) so the total budget should be under $15 which is very reasonable I think :)

I'll draw a schematic if someone is interested ;)

Feel free. Useful information and I'm relative crap with the analog/power electronics, more of a digital guy.

 

[Bidule0hm]

Ok, I'll even test the circuit to be sure it works ;)

I wanted to know if I should prefer eg. Platinum over X-series.

Well, there's only a 2 % difference of efficiency between Gold and Platinum PSUs so if the price of the Platinum one is far higher than the Gold one then you'll never recover the difference in electricity costs.

 

[Ericloewe]

Seasonic Platinum is basically an X-Series with better components. Same design, same features, but slightly improved efficiency.

 

[Bidule0hm]

Ok, I'll even test the circuit to be sure it works ;)

So, circuit done and tested OK. I'm too tired to do the text and video now but in short: I've used a few more components than planned but it's to be able to measure the current on the +12 V rail only and also to avoid a nasty grounding issue of the drive (because a fried drive and/or HBA is not the goal here...), so you'll need an op-amp, 10 resistors, 3 capacitors, one diode, two 9 V batteries, a foot of CAT5(e) cable and of course a DMM. The precision is quite good (about 5 %) for such a simple circuit.

I give one like for who can find why you need the CAT5 cable... :D

 

[danb35]

...or a single resistor and a 'scope...

 

[Bidule0hm]

Yeah but where do you find a scope that fit in a 15 dollars budget? plus there's a big problem with that: if you put the shunt resistor on the +12 V rail you'll not be able to put the probe on it because one side is grounded. If you put the resistor on the ground side then you'll be able to do the measurement but you also need to float the drive (including unplug the SATA cable) because its ground isn't at the same potential than the PSU ground anymore.

And you can't use the CH1 - CH2 trick because the signal you want to see is way too small comparatively to the 12 V you'll have (I tested). So then there's two options: an isolation transformer (bad idea with a PSU like that) or a differential probe (but the price...).

 

[jgreco]

I give one like for who can find why you need the CAT5 cable... :D

To tie up the mad scientist in the room that is hungrily looking at our NAS boxes, duh.

 

[Ericloewe]

I give one like for who can find why you need the CAT5 cable... :D

You mean in addition to wires connecting stuff on the breadboard?

 

[Bidule0hm]

To tie up the mad scientist in the room that is hungrily looking at our NAS boxes, duh.

Nop, you only need a foot of cable...

You mean in addition to wires connecting stuff on the breadboard?

Yep :)

 

[danb35]

Well, I was being a little facetious, but having brought it up, let's see where it goes. Obviously the assumption is that you already have a 'scope, and it really should be a digital scope. If you already have a differential probe as well, great. Failing that, why wouldn't CH1-CH2 with AC coupling enabled work?

Or you could roll your own differential probe for probably the same $15 budget--but that eliminates the "advantage" of simplicity...

 

[jgreco]

To tie up the mad scientist in the room that is hungrily looking at our NAS boxes, duh.

Incidentally, just in case it wasn't obvious, that was actually a compliment.

 

[Ericloewe]

Does it have to be copper twisted pair, or is the Cat. 5 part just what you had at hand?

 

[jgreco]

Does it have to be copper twisted pair, or is the Cat. 5 part just what you had at hand?

Can I upgrade it to fiber?