The ECC enabled Atoms

uOpt

[H]ard|Gawd
Joined
Mar 29, 2006
Messages
1,656
What's your take on those? Anybody using them?

I am so disappointed. For so many years I waited for a really low power but ECC-enabled platform to emerge. So it's finally here, but there's just so many things that don't suit my uses. The only common board seems to be the Supermicro X9SBAA-F-O. It has the following nags:
  • Uses S0-DIMM, not real dimm. (BTW, do these CPUs do dual-channel RAM)
  • VGA out only
  • No PCIe slots (can the CPU handle that, anyway?)
  • 4 SATA ports, not 6, and nothing real to be done about it given the I/O selection
  • It costs $220. That includes the CPU, but I was really aiming differently
  • No extra slot for a boot drive such as a CF card or mSATA or whatever

I would really have loved to keep up a system for occasional fileserving use, such as backups, with harddrives that are powered down most of the time but a mainboard that stays up 24/7 without driving up the summer power bill.
 
What's your take on those?
It's a slim mITX board meant for system integrators. Not many of those here that will have any interest in this.

  • Uses S0-DIMM, not real dimm. (BTW, do these CPUs do dual-channel RAM)
  • VGA out only
  • No PCIe slots (can the CPU handle that, anyway?)
  • 4 SATA ports, not 6, and nothing real to be done about it given the I/O selection
  • It costs $220. That includes the CPU, but I was really aiming differently
  • No extra slot for a boot drive such as a CF card or mSATA or whatever
It's not the motherboard that's lacking:
http://ark.intel.com/products/71267/Intel-Atom-Processor-S1260-1MB-Cache-2_00-GHz

Single-channel, single-DIMM, 8 PCIe 2.0 lanes (probably shared with storage). Dual Gbit Intel NIC's and an IPMI, so not surprising this has only one DIMM socket, 4 SATA and no PCIe slots. There are just no PCIe lanes left.

It's also more than a year old: http://ark.intel.com/products/family/71263

VGA only is common on enterprise hardware and CF-card or mSATA are not.

You want to look at the NEW generation:
http://ark.intel.com/products/77980/Intel-Atom-Processor-C2530-2M-Cache-1_70-GHz

http://ark.intel.com/products/77987/Intel-Atom-Processor-C2750-4M-Cache-2_40-GHz

And a SuperMicro motherboard featuring that 2nd model:
http://www.supermicro.com/products/motherboard/Atom/X10/A1SAi-2750F.cfm

- Up to 32GB DDR3 1600MHz ECC SO-DIMM in 4 DIMM sockets
- Quad GbE LAN ports
- IPMI with dedicated LAN
- 2x SATA3 and 4x SATA2 ports
- 1x PCI-E 2.0 x8 slot

I would really have loved to keep up a system for occasional fileserving use, such as backups, with harddrives that are powered down most of the time but a mainboard that stays up 24/7 without driving up the summer power bill.
Then get a Core i3 4130 or Pentium G3220 with an H81 mITX motherboard. It's not like those CPU's aren't energy efficient, TDP = Thermal Design Power, it's to determine what cooler you'd need, not what power it uses at idle.
 
Last edited:
don't want to derail your thread, but wouldn't some kind of low power haswell chip suit the bill for this as well?
Personally I'm thinking of some Pentium 3220T undervoltaged/underclocked(*) with ECC ram on a 6 port LGA 1150 board, so either Asus or AsRock.

*) that's the part I have no clue about and isn't being touted on the 'faster-higher-louder'-net that much either so one could educate his-self.. ;)
 
You don't need ECC if you are using a filesystem like ZFS, which has parity checking built into the file system, so memory errors will almost never cause your data to go corrupt.

Mind you this doesn't hold up for the likes of NTFS or other popular default OS file systems.
 
You don't need ECC if you are using a filesystem like ZFS, which has parity checking built into the file system, so memory errors will almost never cause your data to go corrupt.

I don't agree with that view at all. All that it takes is for a block address to be switched when actually writing one and poof you go all over your metadata.

And if you are fileserving nothing keeps it from being corrupted between coming out of TCP checksumming and being committed into ZFS.
 
I also said "almost never" because the chance of that happening is small. If it was such a big issue with desktop PC's, it would have been marketed towards power users already, but it rarely happens there. So when you have a checksumming file system, you have even less to worry about.

It also is a lot less data to be written, so the chance of a bit flipping is even smaller. We're not talking about something that is going to happen on a daily basis and unless you are talking about enterprise use and mission-critical data, you're overthinking it. You're more likely to have other issues being a major issue like having the PC stolen, lost in a fire or flood, electricity damage, motherboard or PSU failure, etc etc.

While you are free to do what you want, like stimulating the economy with going ECC, I can't agree with you that ECC is needed or important for home use. You're much better off buying a better PSU, better disks, better motherboard, better case or more beer.
 
While you are free to do what you want, like stimulating the economy with going ECC, I can't agree with you that ECC is needed or important for home use. You're much better off buying a better PSU, better disks, better motherboard, better case or more beer.

Until one of your DIMMs goes bad slowly. That messes up the statistics of error rates pretty good. Not that I find any recent statistics that I find convincing.
 
More parity checking = good. I don't think ECC is needed for most homes, but it all comes down to how important your data is (mine is not that important).
 
Until one of your DIMMs goes bad slowly. That messes up the statistics of error rates pretty good. Not that I find any recent statistics that I find convincing.
I've thought that a normal board would support ECC as long as the CPU does, but after checking yesterday that's not the case.. and as you seem pretty into the theme already, can I ask you something?

Is there any LGA 1150 mITX board with 6 sata ports that does support ECC from the
common manufacturers like Asus..?
 
I've thought that a normal board would support ECC as long as the CPU does, but after checking yesterday that's not the case.. and as you seem pretty into the theme already, can I ask you something?

Is there any LGA 1150 mITX board with 6 sata ports that does support ECC from the
common manufacturers like Asus..?

In the Intel world, you need a Xeon CPU even for single socket and a workstation board with Intel's authorized ECC enabled chipsets. The i3s discussed here are an exception but they still need a workstation board for ECC.

I won't buy Asus anyway, but Supermicro boards are pretty reasonably priced. Of course there's no overclocking.

In the AMD world they removed ECC support from the APUs but the regular lines supports ECC in both CPUs and chipsets, unfortunately only Asus chooses to implement ECC in the BIOS for regular consumer boards.
 
You don't need ECC if you are using a filesystem like ZFS, which has parity checking built into the file system

ZFS uses parity to store and validate redundant data. It also uses CRC checking to validate data and metadata. But ZFS uses all the memory it can get its hands on for such transactions, cacheing, etc. If a bitflip occurs, faulty data can be written to the array, checks or no. If a CRC check intersects a bitflip pending a scrub, then entire sections of data could be rewritten. If a bitflip intersects a resilver, then - worst-case - the entire array may be lost.

In small home systems (<=8G) , the chances of bitflips intersecting critical operations are pretty low, and the consequences of any dataloss usually minimal (but both still enough to make me start saving for a new board:
A system on Earth, at sea level, with 4 GB of RAM has a 96% percent chance of having a bit error in three days without ECC RAM. With ECC RAM, that goes down to 1.67e-10 or about one chance in six billions.
-- http://lambda-diode.com/opinion/ecc-memory - especially note the references at the end of the article).

In enterprise settings, of course, both the number of transactions and the amount of RAM involved are huge, and the chances of a bitflip occurring increase exponentially - and the consequences of such an event intersecting a scrub or resilver likewise huge.

And even home systems today are steadily becoming seriously big. Home storage solutions exceeding 20T aren't really exceptional anymore, and soon boards that support 32G will be the standard.

If you're running a ZFS array on a system without ECC memory, and storing critical data on that array (or just holiday pictures that you wouldn't care to lose), I'd be very nervous.

(You might, after reading the below specs, ask why I am not all that nervous: my smallish zfs pool is really a one-way-mirror for a JBOD, an experimental setup, and important data is backed up separately, so even if the thing becomes total-loss nothing is lost).
 
Back
Top