Low-power NAS build

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

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 10:08 am

HFat wrote:Let's not rush to conclusions. Some cheap measurement devices are especially bad at charaterizing small loads.
I won't disagree that measurement devices are not always accurate. But even a 50% error could not get you 5W from a 20-25W load. My tester is fairly cheap, but it is within 1W on known loads and matches what the serve the home review said about the board.
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 started in the same boat as you, but at least over here there were non-Atom options. There were a few boards with low power Celerons in the same price range. I have a board with a Celeron 1037U and despite its 17W TDP it uses exactly the same power as my D510 at idle. Paid the same as well, about $100 CAD. There were Celeron 847 boards before those.

I just checked and I could still get new Celeron 1037U boards, but the seem to being phased out for the Bay Trail boards. Not necessarily a bad thing, but it was nice to have the option of higher performance in the same price and form factor.
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.
I would say it is a combination of both. People have actually measured the load on the 12V CPU line on Haswells and seen how low they go. That is why there was the worry about "Haswell compatible" PSUs being needed.

Core and the original Atoms had the disadvantage of needing at least two major support chips, while the i series and D series Atoms only need the PCH or what was called the southbridge.

Part of the problem is finding power use numbers for Haswell systems that are not using crazy PSUs that are way out of their efficiency range. I tested a Haswell Pentium system with a DC-DC board and got 37W while running both prime95 and furmark at the same time, plus the system had a 3.5 drive. I was only testing to see if the 60W brick could cope with the system so I didn't write down the idle and I don't have convenient access to that system anymore.

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 10:24 am

HFat wrote:If you only loaded the CPU, yes. But that would make most TDPs grossly... irrelevant.
My apologies if this was not clear, but I hope it was implied that we are talking about otherwise identical systems. Only changing the MB/CPU and RAM if needed. Same disks, PSU, etc, no separate GPU.

When I tested my D510 vs my 1037U it was exactly the same other than the CPU/MB and RAM type. Same OS disk. Got the same idle power.

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 10:41 am

washu wrote:
HFat wrote:If you only loaded the CPU, yes. But that would make most TDPs grossly... irrelevant.
My apologies if this was not clear, but I hope it was implied that we are talking about otherwise identical systems. Only changing the MB/CPU and RAM if needed. Same disks, PSU, etc, no separate GPU.
My point was "CPU" TDPs now include a GPU's power enveloppe. And it can burn a lot of watts. That number wouldn't be very useful for divining efficency even if Intel didn't use the same one for whole product ranges.
washu wrote:Core and the original Atoms had the disadvantage of needing at least two major support chips, while the i series and D series Atoms only need the PCH or what was called the southbridge.
Which has little impact on the issues discussed here in practice. Some older boards manage to be significantly more efficient than most of the ones featuring a CPUs which doesn't have this disadvantage. Just because a part isn't integrated to the CPU doesn't mean it's got to be grossly inefficient.
Idle power consumption has little to do with the CPU. I don't know if it ever had (P4 systems maybe?). It's efficiency at CPU load that depends on the CPU technology (excepting the aforementionned very slow CPUs that have extremely low load consumption).
washu wrote:I started in the same boat as you, but at least over here there were non-Atom options. There were a few boards with low power Celerons in the same price range. I have a board with a Celeron 1037U and despite its 17W TDP it uses exactly the same power as my D510 at idle.
Maybe you haven't read me carefully enough. The D510 isn't a mobile CPU and of course cheap D510 boards aren't going to be very efficient. I've never used a D510 board for server duty.

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

Re: Low-power NAS build

Post by lanac » Fri May 29, 2015 11:17 am

Guys :)

My server is idling 90% of its time. When he is not, he is just streaming files through the network or updating itself... I really don't need to race to idle and I don't transcode media files.
I don't need 30Tb of storage as I delete or move old series/movies to a USB disk.
Baytrail-D boards seem to fit my need with a relatively low cost compared to a Haswell board.
I hope Braswell board will be a little more efficient in the idle state but I'll switch when they'll be available :)

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 11:40 am

HFat wrote:My point was "CPU" TDPs now include a GPU's power enveloppe. And it can burn a lot of watts. That number wouldn't be very useful for divining efficency even if Intel didn't use the same one for whole product ranges.
Unless you are loading the GPU heavily the total power use will be minimal. Intel is very good at power gating, the GPU is going to be largely powered off and clocked down in most cases. If your use cases are heavily loading the GPU then processors under discussion arn't the right ones.

I won't claim it to be 100% accurate, but my i3 in server duty is using 0.03W for the GPU according to the monitoring software. Compared to the 1.5-2W for the whole CPU at idle it's a rounding error.
Maybe you haven't read me carefully enough. The D510 isn't a mobile CPU and of course cheap D510 boards aren't going to be very efficient. I've never used a D510 board for server duty.
My apologies, I did miss the "mobile" part of your statement. We've been talking about desktop hardware, even if low powered versions.

Non Atom mobile based boards do exist, they just aren't cheap.
Which has little impact on the issues discussed here in practice. Some older boards manage to be significantly more efficient than most of the ones featuring a CPUs which doesn't have this disadvantage. Just because a part isn't integrated to the CPU doesn't mean it's got to be grossly inefficient.
Idle power consumption has little to do with the CPU. I don't know if it ever had (P4s systems maybe?). It's efficiency at CPU load that depends on the CPU technology (excepting the aforementionned very slow CPUs that have extremely low load consumption).
We will have to disagree on this one. Sure comparing a older mobile based system to a newer desktop one may show a win for the old one, but comparing desktop for desktop the CPU has had a big impact on idle. As I said before, reviewers have actually measured the 12V CPU current on haswell chips and found it lower than previous gens. Crappy PSUs actually have trouble dealing with such a low load on the 12V line but work on older boards because the 12V load is higher.

Again, using desktop parts on comparable MBs, I've seen a steady drop in idle consumption. Over generations from P4->Core 2->Lynnfield/Sandy Bridge (about the same for both)->Haswell (never had an Ivy desktop). My main server runs 24/7 and usually gets my last desktop after I upgrade. I've measured the power consumption of the boards rather extensively to make sure they are efficient enough for 24/7 operation. I've rejected a few as being too inefficient, and the lack of savings from going Lynnfield -> Sandy Bridge was surprising.

I think many missed the big drop at idle from Core2 -> Lynnfield because they were checking Core 2-> Bloomfield which required the power hungry X58 QPI chip. Even if it wasn't being maxed out, eliminating a 20W+ TDP northbridge that the Core2 required did help lower overall power consumption. That same northbridge is why the first desktop Atoms (270 & 330) were so inefficient. Sure the mobile northbridges were not too bad, even beating out some of the integrated functionality in later desktop chips, but removing them entirely helped quite a bit.

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 11:59 am

washu wrote:Unless you are loading the GPU heavily the total power use will be minimal.
Which is the most important reason why TDP is irrelevant in that case.
washu wrote:As I said before, reviewers have actually measured the 12V CPU current on haswell chips and found it lower than previous gens.
People have of course been measuring the lines feeding the CPU for quite a while... and found it to account for a small fraction of the system's idle power consumption. It's been small enough for years that it doesn't matter much how small it is exactly.

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 3:11 pm

lanac wrote:Guys :)

My server is idling 90% of its time. When he is not, he is just streaming files through the network or updating itself... I really don't need to race to idle and I don't transcode media files.
I don't need 30Tb of storage as I delete or move old series/movies to a USB disk.
Baytrail-D boards seem to fit my need with a relatively low cost compared to a Haswell board.
I hope Braswell board will be a little more efficient in the idle state but I'll switch when they'll be available :)
I guess we went a bit off topic :-)

My recommendation for you specifically is:

- Do you really need need the tuner? If no -> ARM based NAS for even lower power. Get a Synology or other good pre-built NAS.
- If yes to above, will you transcode the video from the tuner? If no -> Baytrail MB.
- If yes to above, you will transcode from the tuner -> Haswell Celeron.

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

Re: Low-power NAS build

Post by lanac » Tue Jun 02, 2015 4:34 am

washu wrote:
lanac wrote:Guys :)

My server is idling 90% of its time. When he is not, he is just streaming files through the network or updating itself... I really don't need to race to idle and I don't transcode media files.
I don't need 30Tb of storage as I delete or move old series/movies to a USB disk.
Baytrail-D boards seem to fit my need with a relatively low cost compared to a Haswell board.
I hope Braswell board will be a little more efficient in the idle state but I'll switch when they'll be available :)
I guess we went a bit off topic :-)

My recommendation for you specifically is:

- Do you really need need the tuner? If no -> ARM based NAS for even lower power. Get a Synology or other good pre-built NAS.
- If yes to above, will you transcode the video from the tuner? If no -> Baytrail MB.
- If yes to above, you will transcode from the tuner -> Haswell Celeron.
PCI/PCI-e is not an option and I don't transcode so I'll go for a Baytrail or even Braswell MB as I'm not in a hurry !

Thanks everyone ;)

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

a

Post by destiriser » Thu Jun 25, 2015 6:33 am

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

Vicotnik
*Lifetime Patron*
Posts: 1831
Joined: Thu Feb 13, 2003 6:53 am
Location: Sweden

Re: Low-power NAS build

Post by Vicotnik » Thu Jun 25, 2015 6:38 am

destiriser wrote:Comparison of Haswell Celeron vs Silvermont Celeron
More like Gigabyte vs ASRock, since the motherboards make up most of the difference.

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

a

Post by destiriser » Thu Jun 25, 2015 6:56 am

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

rado3105
Posts: 29
Joined: Thu Oct 16, 2008 7:54 am
Location: Slovakia

Re: Low-power NAS build

Post by rado3105 » Tue Mar 01, 2016 12:53 am

I want upgrade my celeron 1037, mainly using virtualization. I want power consumption of mobo+CPU to be between 20-40W. What do you recommend? I am thinking of athlon 5350, intel G processors?

Post Reply