DDR3 ECC: Any reasonable prices? Do I need ECC?

[sofakng]

Right now I have a small server running Hyper-V and it's using an AMD Regor II 250 with 8 GB of DDR2 ECC RAM.

It's also my primary file server so I was told to make sure I get ECC RAM to prevent flipped bits especially when moving around so much data. (I'm also using a Dell Perc 6/i card if that matters)

I'd like to upgrade to something that supports VT-d or IOMMU but all of those chipsets seem to require DDR3 which costs an insane amount of money especially for ECC and 12 GB is expensive as heck.

Do I have any other options for VT-d or IOMMU? Do I really need ECC memory?

 

[pjkenned]

How mission critical is the data, what do you use it for, and what else are you running? With that CPU I'd be guessing that you aren't doing encryption. Also, I'd guess that you are running a windows OS in Hyper-V not an OS with ZFS which would use the ram to cache.

I always figure a flipped bit taking out a MP3 file = $0.99 of damage, and odds are it would happen once a year at most. In that situation, I wouldn't worry about ECC as data replacement cost is lower than the ECC premium. On the other hand, if the server has word docs/ presentations that are used for work, a single word file can easily be a $50,000 work product.

Using 12GB or less of RAM you don't need registered memory and can probably do unbuffered ECC which is a bit cheaper.

On the VT-d front I've recently started using a Supermicro X8SIL-F with an Intel Xeon X3440 combo. With that I'm using some Kingston unbuffered ECC memory which runs about $125-130 for 4GB (2x2GB). Fairly reasonable given current memory prices.

Really great combo, however it may be a bit of overkill since it is a LOT more CPU than you have now and it has features like IPMI 2.0. The X3430 (slightly slower) can usually be found for about $200 so it is somewhat expensive, but really not that bad if you want a good all around solution and put a lot of stuff in Hyper-V on that box.

My small word of advice is, pay a bit more to build once versus continual upgrades.

 

[PCTC2]

Well, does the AMD Athlon II 250 even use the ECC functions? I'm pretty sure the memory controller on the AMD Athlon II X2 250 doesn't because it's a consumer-grade processor and lacks the enterprise features of the Opterons. Generally only Xeons and Opterons will use ECC. Since you're using a desktop processor, I'd say you can ditch the ECC.

If you buy a system with Xeon or Opteron class processors, you should definitely buy ECC-registered DDR3 RAM.

 

[RebateMonger]

Generally only Xeons and Opterons will use ECC. Since you're using a desktop processor, I'd say you can ditch the ECC.

ECC functions are controlled by the motherboard chipset, not by the CPU.

Over the years, many "high-end" desktop motherboards used "consumer" CPUs and supported ECC memory. A main difference between the Intel 865 and 875 chipsets, for instance, was the ability to handle ECC. Dell 400SC servers used desktop P4 processors, but used the Intel 875 chipset and ECC memory. My current Dell SC440 server is using a Q6600 desktop processor and refuses to work without ECC memory. My previous Dell SC420 used ECC only and sported a Celeron 2.53 GHz CPU.
Edit: This has changed with the latest Intel CPUs. See my posts further below.

Regarding ECC vs. non-ECC:

In a corporate setting, ECC is well worth the cost difference. Besides potentially protecting against data corruption, ECC greatly reduces downtime when troubleshooting server problems. With non-ECC memory, one of the first things you'd probably do is take the server down to run many hours of memory tests. With ECC, if there's a memory error, the memory controller should let you know.

 

[Voo]

ECC functions are controlled by the motherboard chipset, not by the CPU.

What? So it's not the memory controller that uses the checksums? Are you sure, I think that's rather surprising, but well not my expertise.

 

[RebateMonger]

What? So it's not the memory controller that uses the checksums? Are you sure, I think that's rather surprising, but well not my expertise.

I'm including the memory controller when I say "chipset". I'm trying to say that ECC has little or nothing to do with the CPU itself.

Edit:
While it's true that a "server" CPU generally wasn't necessary to run ECC, it does appear that Intel i7 and i5 CPUs won't work with ECC. Since the memory controller is now implemented inside the CPU, that means that, among recent Intel products, you need a Xeon to do ECC:

http://www.intel.com/support/processors/sb/CS-029913.htm

"Does either the Intel® Core™ i7 processor or the Intel® Core™ i5 processor family support Error Correction Code (ECC) memory?

Neither family of desktop processors supports ECC memory. Typically ECC memory is used on servers and workstations rather than on desktop platforms. This is due to the price premium and extremely low likelihood of a data error occurring even on memory not utilizing ECC." <===?????

 

[Voo]

I'm including the memory controller when I say "chipset". I'm trying to say that ECC has little or nothing to do with the CPU itself.

But doesn't use AMD an integrated memory controller, just like Intel for Nehalems and Westmeres (clarkdale being the sole exception)?

So if the MC is responsible for the ECC features (and imho nothing else makes any kind of sense) and that's integrated in the chip, that would mean that it depends on the CPU if it does support ECC or not.. though I assume the MB also needs to support it.

 

[pjkenned]

To add to your fun, the Clarkdale i3's and i5's do "support" ECC, just not ECC functions. So you can use unbuffered ECC no problem with them.

On the other hand, AMD's CPU's do support ECC on die/package memory controller, but the AMD server motherboards (i.e. IPMI and etc included) do not use socket AM3! So you get ECC, but no BMC. Tyan makes a few decent AM3 server boards, but of the three or so models they make, they are relatively expensive and only one has a really good feature set.

 

[VirtualLarry]

The Biostar A760G M2+, which is the free motherboard you get at Microcenter when you buy their CPU + mobo combo, DOES support ECC. DDR2, unbuffered ECC.

Which, if you happened to pay attention to the Hot Deals forum, was available for $30-35 per 2GB DIMM recently. Possibly still.

I'm sitting on piles of that RAM right now, and waiting for MC to get more of those mobos in stock. Note that there are only two memory slots on that mobo, so server usage might be limited. All AMD AM2 CPUs support ECC in their memory-controller though. Unlike Intel, AMD actually gives you all of the CPU features, even on their lower-end and mainstream CPUs.

I only wish that Intel would learn from their mistakes. I ALWAYS used to run ECC memory, back in the day. There's no good reason why ALL computer systems shouldn't be running ECC. ALL data-paths should be protected by ECC. It's just good computer science. Bit-flips happen. Don't let them happen to you.

 

[RebateMonger]

There's no good reason why ALL computer systems shouldn't be running ECC. ALL data-paths should be protected by ECC. It's just good computer science. Bit-flips happen. Don't let them happen to you.

Many folks will remember the standard "parity chip" that was standard on virtually all memory intended for IBM clones until, I'm thinking, around the mid-90s. An extra memory chip on each row gave an extra parity bit (8 bits plus 1 parity bit) that could detect (but not correct) a memory error.

I'm surprised that we aren't moving back to parity bits or to ECC. Memory errors are not uncommon even now in these days of 1- to 4-Gigabyte-memory computers.

http://www.theinquirer.net/inquirer/news/1558338/memory-errors-frequent

 

[Voo]

Memory errors are not uncommon even now in these days of 1- to 4-Gigabyte-memory computers.

You think a 8% chance per DIMM getting at least one correctable error when running 24/7 the whole year in a (probably rather hot) datacenter, also considering the fact that google probably buys RAM the same way as it buys HDD, is that worrisome? Chances for a usual desktop user are probably a few promille.

 

[Emulex]

ecc is a must for servers - and for people who value their work product 🙂 no questions about that.

 

[RebateMonger]

Chances for a usual desktop user are probably a few promille.

Well, I see quite a few defective memory modules, both in my business and in my personal life. A close friend has been using a Dell 400SC for years now, with 1.5 GB of name-brand ECC memory. She does financial analysis and keeps thousands of spreadsheets.

Twice now in two years she's had the Dell refuse to boot because a memory module has encountered an uncorrectable error and the motherboard shut down because of the error. Without ECC, she might be saving files with errors and no idea they are happening.

 

[Voo]

Yeah but if you look at the data: 8&#37; chance of at least one error per year, while running 24/7 the whole year in most probably suboptimal situations and while juggling lots and lots of data around, seeems like you must be really unfortunate as a normal user to encounter one error if those numbers are correct 😉

Which doesn't mean that those won't happen and since ECC memory often doesn't cost much more, it can't harm (that's especially true for a financial analyst I assume). Maybe I'm also just lucky but at least I haven't had any problems I could lead back to memory errors (though that's probably not that easy)

 

[RebateMonger]

Maybe I'm also just lucky but at least I haven't had any problems I could lead back to memory errors (though that's probably not that easy)

Well, memory is probably more reliable now than it was ten or twenty years ago. It's also more likely that you'll recognize a memory problem now. Who knows what was causing those twice-a-day reboots we'd see with Windows 2.x, 3.x, or Windows 95? Memory? Hardware? Drivers? Or ????

For a business computers or servers, other than price there's no negatives to ECC.