ECC support on low-end consumer 1155 CPUs? ITX boards?

[crashtech]

For me, the majority of people doesn't need to overclock because a new, modern machine is overkill for the vast majority of task, yet people still spend more on bigger Motherboards, Power Supplies, unlocked Processors, aftermarket cooling, etc, just to do it. I don't see where is the problem that some people values stability enough to consider ECC. Maybe stability is not fashion enough to showcase it, hum...

Look, I'm not gonna argue the right of anyone to use ECC if they want to. It's AOK with me.

But I've never had a system that was unstable due to non-ECC RAM. Regular old RAM stays error-free for as long as I have the patience to test it. If someone's system has a problem with BSODs etc., you can bet it is not due to properly functioning non-ECC RAM, and users deserve to know that before committing if they don't have a high enough awareness of what's involved.

Implementing an automatic backup plan is a better use of resources than ECC for the majority of users, imo.

Or you can use ZFS File System.

I like that idea, though probably the only way to practically deploy it for the average user would be a preconfigured NAS. Have you been recommending easy to use ZFS solutions along with your advocacy of ECC?

 

[Vesku]

I like that idea, though probably the only way to practically deploy it for the average user would be a preconfigured NAS. Have you been recommending easy to use ZFS solutions along with your advocacy of ECC?

That's right a low TDP stock clocked processor, ECC RAM and ZFS is a perfectly fine choice for a NAS.

Since it seems like FM2 ITX with ECC support is a no-go, what about 1155 ITX boards and CPUs? I thought I had read that some i3s (and Celerons?) had ECC support enabled.
Would like to possibly purchase a rig like that for a mini-NAS.

 

[crashtech]

This talk of ZFS has got me interested in building a FreeNAS box now.

 

[zir_blazer]

But I've never had a system that was unstable due to non-ECC RAM. Regular old RAM stays error-free for as long as I have the patience to test it. If someone's system has a problem with BSODs etc., you can bet it is not due to properly functioning non-ECC RAM, and users deserve to know that before committing if they don't have a high enough awareness of what's involved.

That's the problem. This is because most people don't understand what ECC is supposed to do, and dismiss it as unneeded. Read here.

Basically, there is some chance that a bit spontaneously turn from 0 to 1 or viceversa, due to cosmic rays background radiation, bla bla bla. Cosmic rays are also one of the possible reasons for silent data corruption on a Hard Disk, that are what ZFS is mean to be reliable against, but there are more causes for those ones. This data corruption can happen on perfectly functional Hardware, not just defective parts, in what case you would need to replace them anyways.
There are statistics that tells you about an "incidence rate" that every 1 GB, there is a percentage chance of a bit to change its value over a given amount of time. That rate seems to be very wide and broad, because it could either happen often enough to a issue, to rare enough that it will never, ever happen on your lifetime. Still, it CAN happen to you.

What happens when that 0 turns to 1? Very probabily, nothing. Worst case scenarios would include data corruption on a file you're working with that you save back to the Hard Disk, that an application crashes, or that your system BSODs. How do you know if an specific issue was attributable to this? Simple, you don't. But you can assume that, if at least one time in your life you had a random weird BSOD, crash, or whatever, that were never able to reproduce again, it COULD have been caused by this sort of data corruption, and that ECC could have corrected it and your system would still be working like if nothing had happened.

My machine is on 24/7, and my record uptime has been 50 days, with 4 GB of common non-ECC RAM. Usually, buggy Drivers due to a non-standard GPU configuration (Radeon 4200 + 2x 5770), memory leaks in some games and such, and worst of all, my electric utility power company brownouts or blackouts, are the cause of my downtime issues. Do I need ECC? Chances are that no. But, it is still a "just in case" measure. In the same way that some people have fire extinguishers in their houses in case of fire, or a gun to protect from intruders if they ever break into their home, for me, ECC is that, a "just in case" measure, that may or may not be needed, but after all, its a mere preventive measure.
The chances that ECC corrects an error that saves the day from a BSOD or crash when I could be doing some mission-critical are virtually 0, but still, as a "just in case" measure, I could consider ECC as an extra feature, if it is at the right price. But why people consider that its fine to pay tons of money to overclock to get extra performance deep in the ever-dimishing returns, but disregard a stability-improving feature as "unneeded", is something I can't yet understand.

 

[crashtech]

We're all entitled to be fanatical about something, I suppose. I can appreciate your passion about ECC even if you have so far failed to infect me with it.

 

[milee]

While data protection is a much much broader subject than just ECC memory, you need to think that the memory subsystem is one of the most important subsystems in computer technology (and I'm not talking about overall PC system RAM memory, just think about pretty much any large system component and you'll see they all do have some kind of memory).

End to end data protection is almost (never say never) non achievable. You protect the system memory, but what about the storage, what about the network interfaces, what about cables, what about internal power, what about external power, what about the software...? Show me just one data path that's 100% protected and I'll prove you wrong.

Of course, once you get professional, you'll gradually try to improve/increase the protection. Many times you're REQUIRED to do so by law. But there's still no guarantee. You can ride a 1997 car with minimal protection or you can cruise in your 2013 armored bulletproof Mercedes-Benz. You can still have a fatal accident in your MB.

I would personally never buy/build a system without ECC RAM for work except for real world test systems. But you can't expect average consumers to understand the implications and refuse to buy non-ECC configured systems.

 

[ShintaiDK]

I would love ECC on every PC. However I can also understand the implications of it. And these are somewhat rather high, considering the still limited benefit of ECC memory. At some time ECC memory will be a requirement. But until then it simply add ~10% cost to memory and lowers the speed a tad.

 

[Magic Carpet]

I would love ECC on every PC. However I can also understand the implications of it. And these are somewhat rather high, considering the still limited benefit of ECC memory. At some time ECC memory will be a requirement. But until then it simply add ~10% cost to memory and lowers the speed a tad.

Trying to find a good board is also a bit of pain. Intel is very reluctant to let go this mainstream, as I believe they don't want to lose their higher-margin server market.

I would personally never buy/build a system without ECC RAM for work except for real world test systems. But you can't expect average consumers to understand the implications and refuse to buy non-ECC configured systems.

Personal preference, that's fine. I have no issues building ECC or non-ECC systems.

Basically, there is some chance that a bit spontaneously turn from 0 to 1 or viceversa, due to cosmic rays background radiation, bla bla bla. Cosmic rays are also one of the possible reasons for silent data corruption on a Hard Disk, that are what ZFS is mean to be reliable against, but there are more causes for those ones. This data corruption can happen on perfectly functional Hardware, not just defective parts, in what case you would need to replace them anyways..

There is some truth to it. If you run a nuclear reactor or a huge commercial database, money isn't as important. The ordinary consumers, usually, vote with their wallets.

 

[ShintaiDK]

Trying to find a good board is also a bit of pain. Intel is very reluctant to let go this mainstream, as I believe they don't want to lose their higher-margin server market.

Why would Intel lose anything? You are not going to use desktops as servers in any business.

 

[Magic Carpet]

Why would Intel lose anything? You are not going to use desktops as servers in any business.

Oh believe me, small businesses would.

Cheapest hardware isn't necessarily much worse (as the price usually reflects that). Certainly, I wouldn't advise my clients to pay a big premium, if I could find something much cheaper and only slightly "worse". Again, I don't like to imply binary logic here (good or bad). There is more gradation to it.

Problem is features and price. Intel knows that.

 

[ShintaiDK]

Oh believe me, small businesses would.

Cheapest hardware isn't necessarily much worse (as the price usually reflects that). Certainly, I wouldn't advise my clients to pay a big premium, if I could find something much cheaper and only slightly "worse". Again, I don't like to imply binary logic here (good or bad). There is more gradation to it.

Problem is features and price. Intel knows that.

Server boards and Xeons for entry level servers cost close to what desktops cost. Its a myth that its supposed to cost much more.

 

[Magic Carpet]

Server boards and Xeons for entry level servers cost close to what desktops cost. Its a myth that its supposed to cost much more.

They may be comparable in price, but not in features, unless of course ECC is the only feature one is looking for. A good ECC board costs way more, than a good non-ECC board. You can't everything at a good price, can you.

ECC memory support should come with mainstream chipsets, as well, for the good of mankind. Chipset/cpu market is way too fragmented, just gives extra headache to system builders.

 

[_Rick_]

The delta isn't that big. I just looked at the cost of a Supermicro X10SAE board (C226 workstation board), and it's only around 200 euro. It comes with dual LAN and full vPro support, an extra SATA controller, ALC 1150 audio, and even PCI ports. Sure, you don't get SLI-certification on these boards, but it'll gladly take a Titan. 8GB DDR3 ECC 1600MHz CL11 isn't that expensive, compared to non-ECC. That board even has Firewire(400) headers, serial port headers, a bunch of USB headers...basically anything anyone would ever need. Even DP1.2.
Compared to a normal 200 euro Gigabyte/ASUS board, you will miss the fancy UEFI design and overclocking, but if you want ECC, then you won't OC anyway.
It's really not such a horribly bad deal as made out, especially with the Xeons coming down in cost too. Of course, there's no excuse for elevated prices either, as most of the stuff is the same as the desktop gear, with just a different serial number burned into some ROM, and some pathways not fused off...

 

[Vesku]

I would love ECC on every PC. However I can also understand the implications of it. And these are somewhat rather high, considering the still limited benefit of ECC memory. At some time ECC memory will be a requirement. But until then it simply add ~10% cost to memory and lowers the speed a tad.

True, at some point the process node size will pretty much demand the use of ECC features on all RAM due to risk of such a dense piece of electronics being more susceptible to environment induced errors.

If I build anything with server like functions though I get ECC because the moderate increase of initial cost on one component of the machine is quite reasonable when the data matters.

 

[sxr7171]

I would upgrade my NAS to ECC RAM if a cost effective solution is available. $900 MB is a bit much.

 

[Ayah]

I would upgrade my NAS to ECC RAM if a cost effective solution is available. $900 MB is a bit much.

i3-3220 = 100$
X9SCL+-F = ~100$/used, 200$/new
8GB 1.35V ECC UDIMMs = 45$/each. 180$/4 sticks.

voila, ecc! great for home nas and smb freebsd router/firewalls.
I'd imagine that the same trend should happen with Haswell dualcores when they come out.

 

[Cerb]

Well, if the person understands what little they are getting for their extra money and performance reduction, so be it. It's just that it is a lot quicker to say they don't need it, which is true in the majority of cases. Silent disk corruption is a bigger problem than memory errors, and there isn't a good way to address it yet with SATA drives. So everyone who thinks they need ECC should also feel SAS RAID array to be mandatory as well.

ZFS, SFV utils, par2, and one of these years, BTRFS. That's off the top of my head. More than a few cheap NASes and NAS software solutions also handle such corruption through software means. You can always trust that a disk is being truthful when it says it has an error, but omission of an error is not proof of being correct.

Disk corruption of standing data can be dealt with by having additional recovery info for that data, and works fine with SATA disks. Live data is another story. Software ECC methods don't perform, outside of embarrassingly parallel work. But, just like disk corruption, the checking part is orders of magnitude more important than correction, except for certain server uses. The correction is nice to have, but parity-checking is the big point.

You're also making 2 other major mistakes:

1. That's there's a real performance reduction. With unbuffered RAM, there is none. With registered, there's 1 clock cycle. One. It makes no difference on a modern CPU. Get some CAS 11 RAM, and you're already at that horribly low performance level (so low you'll need synthetic benchmarks to measure the difference).

2. That there's a lot to it, and any actual cost. Every single socketed Intel CPU has the feature on the chip. It's literally a marketing decision. We could have the option, and then make the decision ourselves, like users of AM2/3 sockets could. That would be best. Then, if I wanted it, I could spend 20% more on my RAM, and turn it on. it would technically need validation per CPU, but that would be a drop in the ocean (if it weren't, it wouldn't be enabled on cheap Pentium and i3 CPUs).

Compared to a normal 200 euro Gigabyte/ASUS board, you will miss the fancy UEFI design

Miss?! We already had GUI BIOSes back in the day. They were just as annoying as the new ones, though they rode the wave of 32-bit and Plug'n'Play, instead of EUFI. I'll gladly take a textmode interface over these gaudy inconsistent GUIs.

 

[Cerb]

End to end data protection is almost (never say never) non achievable. You protect the system memory, but what about the storage, what about the network interfaces, what about cables, what about internal power, what about external power, what about the software...? Show me just one data path that's 100% protected and I'll prove you wrong.

Any small amount of data with a good CRC backing it up. As long as it is read-only, it can be 100% protected. Give it some correction, and it can be protected end to end without needing a reread. This is a large part of why networking is generally so reliable.

When you edit the data, however, *poof* goes any such scheme, and then you are down to best efforts, due diligence, scrubbing unedited data, and so on and so forth.

 

[milee]

Cerb, you got that right (almost). Especially the "you are down to best efforts, due diligence...". Still, think about airplane control systems (2/3) or nuclear power plant systems (3/5) and you'll see that there's still no guarantee, just that the redundancy puts you into very low probability of disaster.

 

[pcunite]

This is an interesting discussion. Perhaps we need a new sub-forum for enthusiasts who are into stability vs. overclocking?

I'm curious as to what it really takes (parts wise) to load a file into RAM, make changes, and then write it back over and over non-stop for years and be notified when something incorrect occurred. However, what about when it gets sent over the WAN and back?

That is my biggest concern, bit flips and then nothing bad or obvious happens, just slow bit rot.

 

[ShintaiDK]

This is an interesting discussion. Perhaps we need a new sub-forum for enthusiasts who are into stability vs. overclocking?

I'm curious as to what it really takes (parts wise) to load a file into RAM, make changes, and then write it back over and over non-stop for years and be notified when something incorrect occurred. However, what about when it gets sent over the WAN and back?

That is my biggest concern, bit flips and then nothing bad or obvious happens, just slow bit rot.

Biterrors via WAN/LAN is another funny thing. IPv4 only got 16bit CRC on the header.

 

[cbn]

This is an interesting discussion. Perhaps we need a new sub-forum for enthusiasts who are into stability vs. overclocking?

+1.

Maybe Workstations/Servers? (Under Hardware and Technology)

 

[bobbozzo]

The MSI FM2-A75IA-E53 FM2 board supports Kingston ECC RAM.

It has 4 SATA + 1 eSATA, and a crappy Realtek NIC (and one PCIe-16 slot).

I'm building a FreeNAS box, either using this or a future Kabini board, if I can find a Kabini supporting ECC.
I'll add an Intel NIC.

Bob

 

[VirtualLarry]

The MSI FM2-A75IA-E53 FM2 board supports Kingston ECC RAM.

How sure are you about this? Since the memory controller for FM2 systems is on the CPU, and there are no FM2 CPUs that I know of that support ECC, then I don't see how this board can actually support ECC.

Running with ECC memory (without ECC protection) is another thing entirely. I'm running ECC RAM in my Q9300 rig, on a P35-DS3R v1.0. Basically, it just means that the BIOS is "tolerant" of ECC, and that it won't reject ECC DIMMs (some boards do this, just to be mean). It doesn't mean that the ECC is actually active and is being used.

 

[crashtech]

The MSI FM2-A75IA-E53 FM2 board supports Kingston ECC RAM.

Proof?