Low-power NAS build

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

lanac
Posts: 26
Joined: Tue May 12, 2015 4:08 am

Low-power NAS build

Post by lanac » Tue May 12, 2015 4:52 am

Hi guys !

I've read numerous threads on this forum before posting this one.
I'm about to renew my NAS hardware (Atom 330, old PATA drive, 1 PCI tuner card, old PSU, 1 USB smart card reader, 2x 3Tb recent HDD) as it consumes about 55W in idle.
It runs headless on Debian (OpenMediaVault distro).

I would be able to use a 60W Pico PSU I already have but which is a little bit too short to power the current setup.

My requirements are :
1 PCI slot, at least 3 sata ports, headless so no need of a powerful video card, no transcoding, good local network streaming, low power.
The server is idling 95% of his time and is running, in the worst case, 18h/day

I picked this setup :
MSI H97M ECO board
Intel Core i3-4150
Kingston KVR16LN11/4 - DDR3L - 4Go
Sandisk SSD - SDSSDP-128G-G25 (didn't find power figures about it, maybe I'll switch to an Ultra Plus)

As an alternative, there is this boards too : Asrock Q1900M Pro3 but I don't know how Atom perform now...

Ultimately, this server will be put off-grid on a solar panel/battery, so every watt gained worths it :)

Do you think this build is correct or are there better alternatives in my case ?

Thanks for your help :)

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Sun May 24, 2015 6:43 pm

Based on your post in the Braswell thread, it's not to late to answer.

The parts you posted are massively overkill for a NAS.
A NAS has no use for anything but the cheapest Celeron. And of course a low-power board would be fine. Looking at the new ones currently selling here, you got the cheap C70M1 which isn't fanless and the fanless Q1900-ITX which is but kind of expensive right now with lots of shops showing no stock. There might be older or newer suitable products in your locale. You might want to wait for a suitable Braswell but the a Bay Trail would be fine.
I don't know OpenMediaVault but I wouldn't have a use for more than 1G of RAM running Linux. I would still buy 2G for future-proofing and a little extra caching but 4G really is a waste.
You don't need an SSD which would do little more than help with boot time... and you should only rarely be booting a NAS. My Debian home server is running from HDD but tweaked so as to allow the HDD to sleep more often. Running form HDD you can also use RAID1 for the boot/system partition(s). If you really want to run from SSD, pick a smaller one. I'd also pick something that is known to be reliable like an affordable Intel 320 from eBay.

xan_user
*Lifetime Patron*
Posts: 2269
Joined: Sun May 21, 2006 9:09 am
Location: Northern California.

Re: Low-power NAS build

Post by xan_user » Sun May 24, 2015 9:12 pm

what is the pci slot for?

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Sun May 24, 2015 10:59 pm

Oops! I forgot that the OP wanted an old PCI slot. I guess it's for the tuner.
I think there are old MSI Celeron 847 boards which are fanless, have 4 SATA ports and 2 PCI slots. They ought to be reasonably efficient. But they're not stocked here anymore (not by large retailers and their suppliers anyway).

lanac
Posts: 26
Joined: Tue May 12, 2015 4:08 am

Re: Low-power NAS build

Post by lanac » Mon May 25, 2015 7:34 am

Thanks for your replies :

@HFat : You got it right, I'm now waiting for Braswell boards from ASRock with the dc jack connector. The PCI was to keep my tuner card but I will upgrade it to a double-tuner from DVBsky. The NAS is shut down during the night when nobody is using it and I plan to allow it to shutdown when we are not home and he is not busy. Boot time factor may eventually become an issue but I'll see how the new hardware perform :)

@xan_user : PCI slot was for the Tuner, but I changed my mind and I'll upgrade to a PCIe 1x

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Mon May 25, 2015 7:52 pm

Maybe your old board can't do it if it's got a really old chipset but you can put desktop hardware in standby (suspend to RAM) nowadays. With a decent power supply, your NAS would consume very little power in that mode and be up much faster (after you send it a WoL packet) than if it were fully shutdown (even if you used an SSD).

lanac
Posts: 26
Joined: Tue May 12, 2015 4:08 am

Re: Low-power NAS build

Post by lanac » Mon May 25, 2015 11:14 pm

I learned something today thanks to you, I didn't know it was possible to wake on lan from a suspend state :)
So yes, the SSD is not necessary anymore !

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Tue May 26, 2015 12:52 am

Something else that came to my mind: I see you have two 3.5'' drives which probably have three platters or more each. Make sure the DC-powered board you're waiting for supports that!
I'm using a DC-powered board for my home server but I'm running slow 2.5'' drives with it... fat ones with many platters but they're still slow 2.5'' drives. 3.5'' drives need quite a bit of power to spin up, especially if they have many platters turning fast. My board is only rated for a 26.2W draw on the SATA power connector(s) and the amount of USB power is limited as well. I tried booting with a single-platter 3.5'' drive and it worked fine but I didn't dare to try a more demanding drive (nevermind two at once).

lanac
Posts: 26
Joined: Tue May 12, 2015 4:08 am

Re: Low-power NAS build

Post by lanac » Tue May 26, 2015 3:29 am

It seems that ASRock supports 3,5" HDD, they estimate 4 disks with 2 slots or DRAM consumes ~79W for their Baytrail-D boards.
It's more than enough for me as my disks consume ~15W when spinning :)

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Tue May 26, 2015 4:18 am

Make sure you understand the difference between spinning and spinning UP. Maybe that's what you meant because there are drives like WD Greens which do not use too much power to spin up but going by specs you could easily top 80W by spinning up 4 faster drives (never mind the rest of the hardware). Maybe ASRock uses staggered spin up on its DC boards but if they don't I would certainly not recommend trying to power up a system sporting four 3.5'' drives with a 60W adapter, whether through a pico or a DC board.

xan_user
*Lifetime Patron*
Posts: 2269
Joined: Sun May 21, 2006 9:09 am
Location: Northern California.

Re: Low-power NAS build

Post by xan_user » Tue May 26, 2015 7:29 am

Id use slower 2.5" laptop drives if power consumption is a main concern.

quest_for_silence
Posts: 5275
Joined: Wed Jun 13, 2007 10:12 am
Location: ITALY

Re: Low-power NAS build

Post by quest_for_silence » Tue May 26, 2015 7:55 am

xan_user wrote:Id use slower 2.5" laptop drives if power consumption is a main concern.

But the quietest 2.5" drives are also much smaller (averagely 1/3) than 3.5" ones, aren't they?

xan_user
*Lifetime Patron*
Posts: 2269
Joined: Sun May 21, 2006 9:09 am
Location: Northern California.

Re: Low-power NAS build

Post by xan_user » Tue May 26, 2015 8:10 am

true. but it is a delicate balance when you want to use solar. What kind of solar system are we talking about anyway? will this "NAS" be running on DC system or AC?
If this is really a "NAS", then noise isn't the big issue, its power use. Right?
how many users will be accessing files at the same time? is the network 100 or 1000?
OP's last build had 6TB, how much will be needed in the new one?

lanac
Posts: 26
Joined: Tue May 12, 2015 4:08 am

Re: Low-power NAS build

Post by lanac » Thu May 28, 2015 1:12 am

I'll keep my disks, they are recent and 3Tb (Raid 1) is enough for me.
Most of the time, they are sleeping so they don't use much power.

The NAS will run on a DC system fed by 2x240W solar panels + a battery. It should be enough even during the winter.

It is a home NAS so it is lightly used, mainly to store movies, TV shows, photos, scans. We are 2 users :)

Noise is not an issue, the NAS is located in the basement.

The whole network is 1000 (CAT 6 FTP cables grounded), all equipments are 1000 except TV clients and the printer.

Abula
Posts: 3662
Joined: Tue Nov 02, 2004 12:22 pm
Location: Guatemala

Re: Low-power NAS build

Post by Abula » Thu May 28, 2015 7:39 am

One thing to consider, is that we are slowly moving toward 4k, i dont expect in the next few years to become a standard, none the less, for me server hardware is upgraded much less than the personal pcs, thus thinking a little on the future would be ideal, given that there is no way of future proofing a setup, you can have more empty hdds slots.

I started my server with 4tb, slowly moved toward 12tb, now im at 29tb and 75% full, starting to consider 4x 6tb wd reds, to have some room till 2016, but if 4k comes sooner, then i still have empty hdd slots that i can still grow into, i cant say my setup is future proof, but i have ways to deal to whatever happens short term.

destiriser
Posts: 35
Joined: Sat Mar 14, 2015 8:57 am

a

Post by destiriser » Thu May 28, 2015 10:24 am

a
Last edited by destiriser on Tue Mar 01, 2016 3:50 am, edited 1 time in total.

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 5:13 am

destiriser wrote:
lanac wrote:As an alternative, there is this boards too : Asrock Q1900M Pro3 but I don't know how Atom perform now...
This is not Atom, but Celeron, there's also a similar Pentium board.
You are incorrect. A J1900 is an Atom based CPU. More specifically, it is a Silvermont architecture Bay Trail Atom that is branded Celeron. The Pentium versions are the same, just a bit faster. The Avoton / Rangeley Atoms you linked are also the same Silvermont architecture.

There is also the current Haswell based Celerons and Pentiums which are the more traditional low end of the current mainstream architecture. They are much faster than the Atom/Silvermont based ones, but higher power and cost.

I think it was really stupid of Intel to have two current architectures with the same names. There are Celerons & Pentiums which are Atoms and Celerons & Pentiums which are Haswell.

destiriser
Posts: 35
Joined: Sat Mar 14, 2015 8:57 am

a

Post by destiriser » Fri May 29, 2015 5:36 am

a
Last edited by destiriser on Tue Mar 01, 2016 3:50 am, edited 1 time in total.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Fri May 29, 2015 6:39 am

Well, there's a few D510 boards I need to sell... I'd make them cheaper than half of the original price for you. :-) If you're in Europe that is (otherwise shipping might be too costly).

And yeah, the Haswell Celerons are generally much faster, especially at single-threaded operations. With the older Atoms (like the D5xx), performance could vary quite a bit depending on the operation. I don't know to what extent that's still the case.
While the Bay Trail Celerons seem very efficient, the power consumed by the rest of the system drags their efficiency way down in practice. The actual efficiency compared with a Haswell system will depend not only on the exact task but also on configuration (underclocking and so forth).

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 7:35 am

destiriser wrote:Are they really that much faster?
Yes, they are. A simple passmark puts the fastest Silvermont Celeron J1900 at 1885 and the slowest Haswell Celeron G1820 at 2866. That's not the whole story as the Silvermont has 4 cores and the Haswell only 2. Per core the Haswell chip is 3X as fast. In addition, the Silvermont GPU only has 4 EUs at 854 MHz and they are a generation older. The Haswell has 10 EUs at 1050 MHz. Not that you would be gaming on either GPU, but the Haswell one is much faster.

As an example, I have a new laptop with a Silvermont Z3775, passmark of 1328. It's basically a slower, lower power version of the Celeron J1900. My old Dell clunker with a Core 2 T7200, passmark of 1174 is still noticiably faster at end user tasks because it is much faster per core.

For a simple NAS the Silvermont is probably the better choice just for power savings. However, add anything CPU demanding and the Haswell is the better choice. It's not uncommon do do transcoding on a NAS like with PLEX and the Haswell would win hands down. In addition to speed, it has newer SSE support and Quicksync support which would help. The Haswell version also offers a much wider range of motherboards which can have more SATA ports if needed for more drives.
The right way to compare things is per Watt and per dollar at the same time, Atoms are way more power efficient and have much smaller max power usage than desktop processors, price is small too and if you care about noise, they are also fanless, another cost drop.
Things are not that simple. First off, the price is not much different. The discussed Asrock Q1900M Pro3 is $100 here. A Celeron G1840 (even faster than the G1820) and an ASRock H81M-HDS is $135. $35 gets you a much faster system and more SATA ports as well. You could probably cheap out on the MB a bit more and get the prices closer, but I wanted to make it comparable quality.

Second, power isn't so drastic either. At idle, the Silvermont would win, but not by much. For sure under 10W difference, probably under 5W. At load the difference would be more pronounced, but remember that the Haswell chip is getting much more work done for the watts used. A long running task such as video transcoding might actually be more efficient on the Haswell because it completes sooner. I don't know for sure in this case, but it was true for the older Atoms. They were so slow that they would actually use more power in the long run completing compute intensive tasks. Race to idle is a valid power saving method.
I think it's great actually, because people read too much not rational criticism of Atoms and don't buy them and I like to buy used stuff for half the price :)
I think Atoms are great in the right situation and I own several. However, anything compute intensive is not for them.

destiriser
Posts: 35
Joined: Sat Mar 14, 2015 8:57 am

a

Post by destiriser » Fri May 29, 2015 7:55 am

a
Last edited by destiriser on Tue Mar 01, 2016 3:50 am, edited 1 time in total.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Fri May 29, 2015 8:03 am

The G1820 doesn't have a TDP of 53W. Intel's numbers are misleading.
washu wrote:I don't know for sure in this case, but it was true for the older Atoms. They were so slow that they would actually use more power in the long run completing compute intensive tasks. Race to idle is a valid power saving method.
Not necessarily. Usually, the assuption of testers is that power consumption drops to zero once the task is completed. That's not race to idle! When you have CPUs such as some of the old Atoms that consume a very small amount of power at load (as long as you didn't load the GPU), in a true race to idle they can be VERY slow and still consume less power... as long as the operation doesn't cause the other components of the system to consume enough extra power to make a difference. In fact, due to PSU inefficiencies it wasn't unusual for a low-power system to consume less at load than a mainstream system at idle! In that case, racing to idle would obviously be a loosing strategy. As usal, the result of any test will largely depend on details.
In the real situations, what's going to be more efficient is often going to be determined by what (if any) other services the system provides while the computation takes place. Race to idle can be made irrelevant either because you're actually racing to standby/shutdown (in which case the faster archictecure typically is more efficient) or because your system isn't idling all that often anyway (in which case the low-power architecture is often more efficient).

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 8:29 am

destiriser wrote:Celeron J1900 has TDP of 10W, Celeron G1820 53W.
That doesn't say anything about how much power they actually use. Intel ranks pretty much all it's dual core CPUs the same. From the G1820 up to the i3-4370. I have an i3-4150 and with prime95 in max power mode it only uses an additional 25W from the wall. The G1820 would be less.
It's absolutely pointless to compare processors designed for different use in this maneer.
Both the J1900 and G1820 are current low end desktop CPUs, it is perfectly reasonable to compare them.
It's like comparing the newest Snapdragon to Haswell i7, if it didn't have a bigger passmark it would be completely ridiculous. The only objective performance measure is per watt.
No, performance per watt is by far not the only objective performance measure. Raw performance is a perfectly valid measure. A snapdragon simply cannot handle the same workloads that would call for an i7. There are people at my work (3D artists) who would pay huge money for a faster CPU even if it used 10X the power of current ones and they already have power hungry Xeon workstations. If you told them they had to use a snapdragon in their workstation they would quit on the spot. In the big iron world there are POWER CPUs that make Haswell Xeons look cold for only a tiny performance gain. People buy them. If you need speed you need speed.
Celeron doesn't have this passmark score big either, it's really slow for today's desktop computing.
Which one? The G1820 is a fine desktop CPU for non-demanding tasks. At a previous job I had to use a Celeron G540 in a spare developer desktop because I was out of standard replacements. The user complained mightily about the indignity of being given a "slow Celeron". By the end of the day they didn't want to give it back because for everything but compiling it was faster than their 4 core AMD.
The difference in the tests I found was 5W between Haswell and very old Atom D525! Haswell with 10W, Atom with 5.
Now you are just making stuff up. There is no way you could have actually tested this and gotten those results. From the wall there is no way a D525 will only pull 5W. That is a 15W from the wall system at the absolute best. You can't be referring to just CPU watts either as a Haswell is well under 10W idle.

destiriser
Posts: 35
Joined: Sat Mar 14, 2015 8:57 am

a

Post by destiriser » Fri May 29, 2015 8:38 am

a
Last edited by destiriser on Tue Mar 01, 2016 3:49 am, edited 1 time in total.

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 8:45 am

HFat wrote:In fact, due to PSU inefficiencies it wasn't unusual for a low-power system to consume less at load than a mainstream system at idle! In that case, racing to idle would obviously be a loosing strategy. As usal, the result of any test will largely depend on details.
I see the point you are making, and it was likely true when you were comparing the old Atoms to the Core 2/First gen iXs of the time. But now a Haswell CPU idles at less power than a desktop Atom at load. Again, assuming other conditions are similar race to idle certainly does apply in with the modern CPUs.

As you and I have both pointed out, the TDP of most Haswell CPUs is grossly inflated. Depending on workload I could easily see a G1820 only 10-15W more than a J1900 at load.

destiriser
Posts: 35
Joined: Sat Mar 14, 2015 8:57 am

a

Post by destiriser » Fri May 29, 2015 8:53 am

a
Last edited by destiriser on Tue Mar 01, 2016 3:49 am, edited 1 time in total.

fjf
Posts: 192
Joined: Wed Apr 06, 2005 9:44 am
Location: Europe

Re: Low-power NAS build

Post by fjf » Fri May 29, 2015 9:00 am

A good and cheap NAS is the HP n54l. I have one. A bit noisy, but can be silenced.

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 9:08 am

destiriser wrote:They are directed to different use cases.
No they are aimed at exactly the same use case: Low end desktop.
Snapdragon isn't designed for what i7 is, it could be.
Sure, and if it was just as fast an i7 it would use just as much power and likely quite a bit more. The snapdragon's poorer node process, ARMs instruction density penalty and the laws of physics make it almost certain that it would use more power than an i7 at equivalent performance. Modern i7s are very efficient for the performance they provide. Every other CPU in the same performance category uses much more power and are less efficient.
I don't think anyone would actually waste money on power for a little gain in performance, you can simply use few xeons.
What you think is not reality. If you need performance you need performance. Not every computational problem can be solved by throwing cores at it. Not every problem can take a few seconds longer.
I found few tests in the Internet, that's it. Do you have some other tests?
Sure. There was a toms hardware test that got a crazy 35W idle on a D525, but that was likely their poor setup.

This Test shows a 23W idle. I also have personally tested my board and got about the same. That is why I said 15W at the best if you had a very efficient PS, but that is likely pushing it. Interestingly it shows a G630 (a Sandy Bridge CPU) at 27W idle, only 5W more. Haswells are even lower than that.

washu
Posts: 571
Joined: Thu Nov 19, 2009 10:20 am
Location: Ottawa

Re: Low-power NAS build

Post by washu » Fri May 29, 2015 9:23 am

destiriser wrote:I only mentioned TDP to show that it's a completely different kind of a processor,
It's a different processor aimed at the same market and use case. The AMD AM1 and low end APUs are two more different processors aimed at the same market and use case.
f you want to compare it to haswell, then compare it to haswell U or Y. It will be closer.
Haswell U or Y are laptop/tablet CPUs. Different market and use cases. In addition they make the desktop Atom look even worse. An i5-4302Y has a passmark of 2936 vs 1885 for the Celeron J1900 for only 1.5W more. A Core M-5Y71 (Brodwell-Y) has a passmark score of 2962 for 5.5W less than the Atom. The Atom is a low power, low performance, low price CPU. It's not necessarily an efficient CPU.

I own a silvermont based Atom. I like it. It was cheap. If I could have bought a Haswell-Y or Broadwell-Y device for the same price I would have and got a faster and more efficient device. But that is not how the market work and why the Atoms exist.
And 15W isn't exactly that little.
It is if it saves you power in the long run or lets you perform a task in a more reasonable amount of time. As per above, if you want to pay for it you can do it with less power than the Atom.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: Low-power NAS build

Post by HFat » Fri May 29, 2015 9:25 am

washu wrote:Now you are just making stuff up. There is no way you could have actually tested this and gotten those results.
Let's not rush to conclusions. Some cheap measurement devices are especially bad at charaterizing small loads.
washu wrote:But now a Haswell CPU idles at less power than a desktop Atom at load. Again, assuming other conditions are similar race to idle certainly does apply in with the modern CPUs.
Like many, I've been using mobile Atoms for light server duty because they were offered as part of affordable and efficient boards. That hasn't been the case for non-Atom Intels in a long while. Again, other conditions are rarely similar. :-)

I assumed we were talking about the power consumption of the system (seeing that the CPU power consumption real-world relevance is limited to CPU cooling which isn't what we were talking about).
The reason you can get efficient Haswell systems now is not that the CPU idles at less power than the oldest "Core" CPUs but that the other parts included in typical systems have become more efficient. Back in the day, you could build a pretty efficient system by choosing the board carefully, underclocking/undervolting, avoiding 3.5'' drives and using a pico paired with an efficient brick... and you still couldn't touch the best mobile Atom boards. Now you can simply buy a NUC and beat (or at least match) them.
washu wrote:As you and I have both pointed out, the TDP of most Haswell CPUs is grossly inflated. Depending on workload I could easily see a G1820 only 10-15W more than a J1900 at load.
If you only loaded the CPU, yes. But that would make most TDPs grossly... irrelevant.
Also, underclocking the Haswell a bit would make that number even smaller without making the system slower than the J1900 system.

Post Reply