TR Forums

To say the least, I'm disappointed. You have $500+, requiring an even more expensive socket/platform to take advantage of their memory throughput and then you don't offer ECC optionally!? Why on Earth would you not offer it on these but instead of something like an i3-3225? Any insight into this?

Was looking at building a server for a small office and was going to go with an E series. Also noticed the E5 series, which are much more expensive, have a few more cores, faster memory options and more L3 cache, but serious overkill for this project. ECC being a must for what I'm doing, are there any better options really available? I have in the past built AMD FX-8130 systems due to their ECC support. My only purpose for going with an E series was ECC like they previously did, what gives?

E3 I suppose??? /sigh, Intel your segmentation just makes no sense to consumers, but does for your pocket book.

The "Xeon lite" chips (Core i7 38xx/39xx or 48xx/49xx) have never supported ECC.

The Core i3-3225 most likely doesn't support ECC. It's more likely that is a mistake in the Ark.

You want a Xeon E3 V3. Not only are they cheaper than the "Xeon lite" chips above, but they support ECC. (E3-1220 V3/E3-1270 V3)

They have models of the E3 V3 with integrated graphics if you need that. Just looks for models with a 5 at the end.

Thank Ryu, I noticed those E3, E5 and E7 with ECC. I could have sworn that the older E series chips on Skt 2011 supported ECC. Only seems to make sense when your talking about workstations doing things like CAD to have ECC. If the plain old consumer chips from AMD (Entire FX line) support ECC, you have to wonder why they don't include support on their (Intel) higher end chips. I guess they are the same silicon for the Xeon E3, E5, and E7 chips, but it sort of leaves the other guys out of the loop who buy a Sandy Bridge-E which is supposed to be for enthusiast.

Xeon E3 it is for this project I guess. I hope to god your right, can't imagine why the i3-3225 would have ECC support, must be an ARK typo. Thanks for the heads up Ryu

Ryu, actually i3's and celerons DO support ECC memory. I imagine they do because they are popular chips to use in SOHO servers.

Unfortunately they don't offer hardware AES support for encryption.

Irrelevant info pulled. Accurate info below.

As an aside you'll want a C22x series LGA1150 board.

Ryu Connor wrote:

There are some i3 CPUs that support ECC, that is totally true. Core i3-330E, i3-2115C, i3-2310E, i3-2340UE, i3-3115C, i3-3120ME, i3-3217UE, i3-4100E, and i3-4102E. The following chips all have something in common, they're all soldered to the motherboard. (In the disclosure of pedantry apparently old Nehalem based i3 chips could do it under very specific conditions.)

All i3 parts support ECC memory. It is not a mistake in the ARK. The wikipedia article is incomplete/wrong. The FreeNAS community knows this and those guys have been using this fact for a long time. I'm building my own Mini-ITX C226 based NAS system with 6x3TB WD Red drives with 16 GB of ECC DDR3 and an i3-4340. 12 TB in RaidZ2 FTW.

Yeah, I just noticed every entry in the Ark repeats it. Looks like you have to pair it with a server chipset. Motherboard should end up more expensive than the CPU.

moridinga wrote:

Ryu Connor wrote:

There are some i3 CPUs that support ECC, that is totally true. Core i3-330E, i3-2115C, i3-2310E, i3-2340UE, i3-3115C, i3-3120ME, i3-3217UE, i3-4100E, and i3-4102E. The following chips all have something in common, they're all soldered to the motherboard. (In the disclosure of pedantry apparently old Nehalem based i3 chips could do it under very specific conditions.)

All i3 parts support ECC memory. It is not a mistake in the ARK. The wikipedia article is incomplete/wrong. The FreeNAS community knows this and those guys have been using this fact for a long time. I'm building my own Mini-ITX C226 based NAS system with 6x3TB WD Red drives with 16 GB of ECC DDR3 and an i3-4340. 12 TB in RaidZ2 FTW.

I wouldn't say all i3's support ECC. None of the Haswell i3's do. Some SB and most IB i3's do however. I have an i3-3225's,i3-3220T's, Celeron G1610's in deployment with C216 chipset motherboards all using ECC. It is certainly not limited to soldered on only versions.

Deanjo wrote:

I wouldn't say all i3's support ECC. None of the Haswell i3's do.

Unless the ARK is wrong, Haswell i3's do.
http://ark.intel.com/compare/77481,77770,77771,75463

moridinga wrote:

Deanjo wrote:

I wouldn't say all i3's support ECC. None of the Haswell i3's do.

Unless the ARK is wrong, Haswell i3's do.
http://ark.intel.com/compare/77481,77770,77771,75463

Missed that, thanks for correcting me.

Ryu Connor wrote:

Yeah, I just noticed every entry in the Ark repeats it. Looks like you have to pair it with a server chipset. Motherboard should end up more expensive than the CPU.

How important is your data to you? If you are building a file server, you want good data integrity and spending $150 on a motherboard shouldn't be too much of a sacrifice. You are already spending a premium on the ram after all and you do want to keep the power consumption down.

Just an observation. My statement is not meant to infer any particular form of conclusion. I own a $400 motherboard so I am not really one to get involved in cost debates.

So there are some slim pickings for the C222 chipsets. Some SuperMicro/ASRock and Asus motherboards. But I'm noticing that for instance on the P9D-X, it states that integrated graphics on the E3-12xx v3 Xeons (1245 in this case as an example) are not supported. So why would you include a VGA port on the back if all of the chips that have IGP aren't supported for their IGP?

The P9D-X would be ideal if I could use the IGP so as not to spend more money on discrete graphics and need to find one without a fan to avoid having a dead fan in a server. Seems silly, but its one of those things that I feel probably work fine, but isn't "officially" supported. Any insight into the Xeon E3 V3 series IGP use on C222/C224?

Welch wrote:

So there are some slim pickings for the C222 chipsets. Some SuperMicro/ASRock and Asus motherboards. But I'm noticing that for instance on the P9D-X, it states that integrated graphics on the E3-12xx v3 Xeons (1245 in this case as an example) are not supported. So why would you include a VGA port on the back if all of the chips that have IGP aren't supported for their IGP?

The P9D-X would be ideal if I could use the IGP so as not to spend more money on discrete graphics and need to find one without a fan to avoid having a dead fan in a server. Seems silly, but its one of those things that I feel probably work fine, but isn't "officially" supported. Any insight into the Xeon E3 V3 series IGP use on C222/C224?

The P9D-X actually has its "own" GPU onboard, rather than using the CPU's IGP - the back panel/front panel VGA are hooked up to the "Aspeed AST1300 with 64MB VRAM" rather than to the CPU IGP. Not sure of the reasoning behind it, other than these are server boards and as such you *never* want a discrete GPU, hence the inclusion of a basic GPU rather than having to limit it to just Xeon CPUs with IGPs.

If cost is a concern, you could still go with a Socket AM3+ board and an FX-x3xx and get your ECC support. An Asus M5A97 R2.0 motherboard is only 95 bucks.

Edit: Yeah, it lacks an IGP. But there are enough choices for inexpensive low-end fanless discrete GPUs that IMO this isn't a big deal. Also note that unless you have an IGP with its own VRAM (like that Aspeed mentioned above), the IGP will steal memory bandwidth and impact the performance of the server.

Mentawl wrote:

Welch wrote:

So there are some slim pickings for the C222 chipsets. Some SuperMicro/ASRock and Asus motherboards. But I'm noticing that for instance on the P9D-X, it states that integrated graphics on the E3-12xx v3 Xeons (1245 in this case as an example) are not supported. So why would you include a VGA port on the back if all of the chips that have IGP aren't supported for their IGP?

The P9D-X would be ideal if I could use the IGP so as not to spend more money on discrete graphics and need to find one without a fan to avoid having a dead fan in a server. Seems silly, but its one of those things that I feel probably work fine, but isn't "officially" supported. Any insight into the Xeon E3 V3 series IGP use on C222/C224?

The P9D-X actually has its "own" GPU onboard, rather than using the CPU's IGP - the back panel/front panel VGA are hooked up to the "Aspeed AST1300 with 64MB VRAM" rather than to the CPU IGP. Not sure of the reasoning behind it, other than these are server boards and as such you *never* want a discrete GPU, hence the inclusion of a basic GPU rather than having to limit it to just Xeon CPUs with IGPs.

Perfect, I overlooked that. What I've done in the past when building an AMD based server was throw a fan-less discrete card in there that cost around 30-40 bucks enough to have display and called it good. They have worked well but I figured if its got IGP why not use it. I'd assume because the majority of the Intel Xeon E3 and other E5/E7 don't have IGP unless you buy the specific model that has it. In this case I can purchase the E3 1240 instead of the 45, only a $5.00 difference anyhow, less junk on the silicon should (in some cases) mean less power use overall and less heat.

just brew it! wrote:

If cost is a concern, you could still go with a Socket AM3+ board and an FX-x3xx and get your ECC support. An Asus M5A97 R2.0 motherboard is only 95 bucks.

Yep did that last time , for whatever reason I'm wanting to test the waters with an Intel based server or at least price it out in comparison. Nowadays, short of the chipset/software it doesn't seem like motherboards for servers are that much more reliable as they are made with much of the same parts.

Some asrock boards support ECC when every other data point including intel's not-very-accurate-and-haphazardly-updated ARK says it shouldn't. (yeah, my E3 1265Lv2 that didn't come with the originally listed 4000 GPU says 'SUP INTEL)

The worst thing about ECC support for the last decade or so? Confirming it is actually working. There are half a dozen places to check, some of which will lie to you when it IS working, and others that will lie to you when it ISN'T working. You could also end up with a configuration where detection is working but not the nearly as important correction of single bit errors.

To be honest, I've come to the sad reality that the only completely verifiable way to test is tape a data pin on an ECC dimm and boot it up. If it is 100% functional the system will correct every dead bit, otherwise I doubt it even POSTs.

I'd place a lot of blame on Intel's {arrogance|dominance|apathy|monopoly} letting the situation remain stupid for 10+ years. AMD was better in supporting it mainstream but regressed with the FM# sockets. The motherboard OEMs are mostly useless which is par for the course, the typical bios is written by a room full of illiterate monkeys on typewriters anyways.

I am a slightly interested party as I currently have a 4820K in an asrock board with ecc kingston dimms. It is a placeholder for an eventual E5-16xx but some hardware info dump tools seem to think ECC is already working.
I however remain extremely skeptical. I also have ungainly hands way too large to attempt the pin covering method.

Bauxite wrote:

The worst thing about ECC support for the last decade or so? Confirming it is actually working.

Windows can tell you.

Welch wrote:

Nowadays, short of the chipset/software it doesn't seem like motherboards for servers are that much more reliable as they are made with much of the same parts.

There is room for debate in that.

Mentawl wrote:

other than these are server boards and as such you *never* want a discrete GPU,

That's a bunch of poppycock. There are several reasons why you would want to use a discrete GPU in a server situation such as GPGPU, video streaming encoding, RemoteFX, etc.

Just to add to the confusion, it isn't uncommon for desktop motherboard specs to indicate that the motherboard "supports" ECC DIMMs, when all they really mean is that the board will POST and run with ECC DIMMs installed (but is actually ignoring the ECC bits).

My SOP for a number of years now has been to download the manual for a motherboard I'm thinking of buying, and check the BIOS options for anything related to DRAM ECC. This is a very good (but still not 100% reliable) indication of functional ECC support.

Ryu Connor wrote:

Bauxite wrote:

The worst thing about ECC support for the last decade or so? Confirming it is actually working.

Windows can tell you.

Only if you have errors, and you might be waiting awhile to rule out the lack of hits as a false negative. Like well over a year going by the google whitepaper.

The takeaway from the recent study is you are more likely to have no problems at all ("Only 8% of DIMMs had errors per year on average") but when you do, they hit hard. By then learning that the misreported-as-functional ECC hasn't been working and you finally noticed you have a backlog of corrupt data...kinda too late.

Welch wrote:

The P9D-X actually has its "own" GPU onboard, rather than using the CPU's IGP - the back panel/front panel VGA are hooked up to the "Aspeed AST1300 with 64MB VRAM" rather than to the CPU IGP. Not sure of the reasoning behind it, other than these are server boards and as such you *never* want a discrete GPU, hence the inclusion of a basic GPU rather than having to limit it to just Xeon CPUs with IGPs.

As far as I understand it, this on board VGA is related to the IP management interface. I guess a video card needs to be present to render the output from the BIOS and later the OS and then pipe it out over the management Ethernet port. But it can also do output over the analog VGA.

For example, the board which I'm getting for my NAS system (http://www.asrock.com/server/overview.a ... =E3C226D2I), as three Ethernet ports, two Intel ones for the server activity, and one for the IPMI. I'm looking forward to trying this IPMI stuff out - apparently you can remotely mount install media (ISO files, etc) and setup the system without ever plugging in a mouse, keyboard, or monitor.