Why isn't ECC memory used more?

[thewhat]

Why isn't it used more in consumer PCs?

The price of ECC RAM compared to regular RAM is not much higher at all. Are the ECC compatible CPU/chipset/motherboard that much more expensive to produce?

The only other downside besides price is, according to Wikipedia, a 2-3% performance hit, which again is not that much.

OTOH, you avoid errors, which, again according to Wikipedia, affect 8% of modules per year. Even if this figure is for servers which run all the time, it's not that low of a number, IMO.

 

[VirtualLarry]

This. I cannot fathom why Intel has not mandated ECC throughout the system, including the RAM. It's just good fundamental computer science. ESPECIALLY with RAM quantities increasing so much. It greatly increases the odds of bit-errors in systems. Not to mention, the solar sunspot cycle is going to be at the 7-year high point this year.

AMD does support ECC on their CPUs, BIOS and motherboard design willing. But it is in their memory-controller. Intel actually goes out of their way to disable the ECC support on their consumer CPUs.

Why does Intel care more about its profit margin (from selling "server" CPUs, with ECC), than with the end-user's data-integrity? It's shameful!

 

[Mark R]

Two reasons:

1. ECC RAM costs 12% more to produce than normal RAM. This eats into the already razor thin margin for home computers, where consumers are extremely price sensitive (unless you can give them something unique - like apple does).

2. Intel disables ECC support for home CPUs, and only enables it for their premium Xeon CPUs. It costs them nothing to enable it, but they disable it to encourage businesses to buy their high-profit Xeon CPUs over their low-profit core/pentium CPUs.

 

[thewhat]

I saw now that something like an Asus P8B WS and a Xeon 1240 are actually comparable in price to a non ECC alternative. Too bad that ECC supporting motherboards like that (with extra "non-server" features) are kinda rare.

I have some more questions:
1. what does does buffered vs unbuffered ECC do? how is one better than the other?
2. is the locking of Xeons (so that you can't overclock them) purely a strategic/marketing decision or does ECC perhaps require a specific fixed clock to work properly?

 

[paul878]

Intel disable to make the PC cheaper, hence they sell more cpu and chipsets.

 

[Stayfr0sty]

Its server memory no need for it on home desktops.

 

[exdeath]

Intel disable to make the PC cheaper, hence they sell more cpu and chipsets.

But the feature is already there physically regardless... it costs nothing to enable/disable it.

 

[paperwastage]

But the feature is already there physically regardless... it costs nothing to enable/disable it.

more like what Mark R said... Intel disables it so that it can sell 2 lines: Consumer(without ECC, cheaper) and Business/Enterprise (ECC, more expensive)

if they enable ECC for consumer, they'll lose $ from the business customers who'll switch over to the consumer brand... and they make up money by raising prices for both Consumer and Business end

I saw now that something like an Asus P8B WS and a Xeon 1240 are actually comparable in price to a non ECC alternative. Too bad that ECC supporting motherboards like that (with extra "non-server" features) are kinda rare.

I have some more questions:
1. what does does buffered vs unbuffered ECC do? how is one better than the other?
2. is the locking of Xeons (so that you can't overclock them) purely a strategic/marketing decision or does ECC perhaps require a specific fixed clock to work properly?

for part 1, read wikipedia. http://en.wikipedia.org/wiki/Unbuffered_memory

basically, registered mem provides another layer in the architecture

for part 2... its part marketing decision part target audience. what are Xeons put in? Mostly servers in server racks... do you want more heat / TDP in a 42U cabinet (when you OC)?

also, most xeon builders want stability over anything, and Intel can only guarantee the chip to work up to the specified ghz

 

[exdeath]

more like what Mark R said... Intel disables it so that it can sell 2 lines: Consumer(without ECC, cheaper) and Business/Enterprise (ECC, more expensive)

if they enable ECC for consumer, they'll lose $ from the business customers who'll switch over to the consumer brand... and they make up money by raising prices for both Consumer and Business end

In other words hurting consumers in order to rip off businesses :awe:

 

[paperwastage]

In other words hurting consumers in order to rip off businesses :awe:

well, it's not like you're running any mission critical component that requires strict data checksums.... if you do, i don't consider you a consumer

i guess the $ for consumers(extra cost in RAM, latency) isn't really worth the benefits of ECC...

 

[Mark R]

I have some more questions:
1. what does does buffered vs unbuffered ECC do? how is one better than the other?
2. is the locking of Xeons (so that you can't overclock them) purely a strategic/marketing decision or does ECC perhaps require a specific fixed clock to work properly?

1. Buffered memory contains signal amplifiers (buffers) on the DIMMs.

On regular RAM, the memory controller circuit on the CPU must operate the RAM control circuits directly. If you have a lot of RAM (more than 2 DIMMs per channel) then the controller may not be strong enough to send a reliable signal to all the RAM chips, or the long length of the motherboard traces might distort the control signals. Similarly, signals from the RAM might get corrupted by going down long traces past several other RAM modules.

The buffers clean-up and boost the signals from the CPU, to ensure that the RAM gets a clean signal, and boost the signal from the RAM chips to ensure that it is received clearly by the CPU.

The buffers ensure stability in high RAM systems, and also help protect against data corruption caused by electrical interference on the mobo (only the data is ECC protected, control signals aren't - so your correct data might end up being stored in the wrong memory cell - so any interference on the control signals, and your data may be gone).

2. It's a business decision.

 

[thewhat]

Thanks for the explanation, Mark R.

Is there a way to measure the amount of such signal corruption?
IOW, how would one determine if buffered RAM is necessary or redundant?

For example, there's the 8% figure for modules affected by memory errors per year and from what I understand that is solved with ECC.

Would using buffered RAM on a workstation motherboard like the Asus P8B WS, which only has 4 slots (max two per channel), be completely redundant? Of course, that specific motherboard can't use buffered RAM, but what I'm asking is if that's just to reduce cost or if there's simply a lack of need for it, with so few modules.

 

[anikhtos]

now that is were things get funny
an i3 with a server chipset mainboard supports ecc ram!?!?!?
so there is a semi profesional aproach this time to get ecc support with intel consumer processors.
the funiest is i5 or i7 do not support ecc ram at all?!?!?!?
propably cause it would killed the sales of the xeon but intel feels safe for the i3 over xeons
e3-1220l cost 180 euro
i3-2100 cost 100 euro
asus p8b ws cost 170 euro
mainboard with p67 chipset cost 75 euro
so for 95 more euro you can have ecc memory for the consumer
http://www.asus.com/Motherboards/Intel_Socket_1155/product.aspx?P_ID=XGLe7ZgHTPXJPu7N&templete=2

 

[nanaki333]

adding ECC to desktop would be a huge mess. have you ever had to build a server, or upgrade memory in one? you have to get just the right number of ranks or it's not even going to POST. you have to actually RESEARCH what will and will not work on some server boards. i made the mistake a couple years ago thinking it was all the same and ordered the cheapest ECC with ranking of x2. turned out i needed minimum of x4.

imagine throwing this at a consumer. i go to microcenter and i see customers staring at the memory for like 5 minutes confused about just the SPEED of the memory. then buffered vs unbuffered? customers would stop building PCs and just happily over pay for apple.

 

[NP Complete]

The thing that bothers me about the whole "screwing the consumer" line of logic is that most of the time it comes across as a "I want more crap for free," and less of a "business are consciously trying to screw consumers."

Chip manufacturers have employed mechanisms of "soft SKUing" chips through disabling feature after manufacturing for a long time. This came from a realization that designing, testing, and manufacturing a chip is expensive. Making two chips is pretty much 2x the cost of making 1, even if you end up re-using a lot of the same design. There are some savings using less silicon on each chip, but those are far outweighed by having to maintain 2 different sets of manufacturing and testing equipment.

But if Intel is only making 1 chip, how does it sell into different markets? Not everyone wants to spend $1000 on a chip, and if all Intel did was put different stickers on chips without disabling features, who do you think would be dumb enough to buy a the same chip for 2x the cost.

In reality, if Intel chose to either design a different chip for each set of features, or if it only sold one chip with all available feature enabled, the cost to consumers would be much higher overall.

Features aren't "free" even if they're on the chip - Intel had to pay for them to be designed and validated.

The discussion really should be about whether ECC is worth the additional costs on motherboards and RAM, considering that the overall cost would likely be lower if it was enabled on each motherboard and chip. There's another discussion that can be had about whether Intel's pricing on each chip represents the value of that set of features.

 

[Coup27]

For someone to doesn't know the ins and outs of this subject, what happens with errors on a standard desktop PC? My RAM doesn't have ECC but I don't ever notice any errors?

 

[nanaki333]

For someone to doesn't know the ins and outs of this subject, what happens with errors on a standard desktop PC? My RAM doesn't have ECC but I don't ever notice any errors?

you'll have a lot more crashes with programs and you'll probably get some nice windows error popups saying it's unable to access memory location blah blah. if that happens, you run a program like memtest86+ to find the offending DIMM.

 

[NP Complete]

you'll have a lot more crashes with programs and you'll probably get some nice windows error popups saying it's unable to access memory location blah blah. if that happens, you run a program like memtest86+ to find the offending DIMM.

I think "lots more" is a bit of an over statement. An often quoted google study puts DIMM error rates at 8% of DIMMS have >= 1 error/year. Another slice is 25,000 to 70,000 errors per billion device hours per Mbit. The google study is here: http://research.google.com/pubs/pub35162.html.

While this is important for mission critical systems that are handling lots of traffic, in reality this most likely means that you get a crash every 2-3 years, or when you rebuild a system you get a bad stick that makes your system crash a lot (in which case you run memtest86).

 

[nanaki333]

I think "lots more" is a bit of an over statement. An often quoted google study puts DIMM error rates at 8% of DIMMS have >= 1 error/year. Another slice is 25,000 to 70,000 errors per billion device hours per Mbit. The google study is here: http://research.google.com/pubs/pub35162.html.

While this is important for mission critical systems that are handling lots of traffic, in reality this most likely means that you get a crash every 2-3 years, or when you rebuild a system you get a bad stick that makes your system crash a lot (in which case you run memtest86).

having just 1 crash is a lot more than never having a problem before a ram upgrade

 

[Coup27]

Ah, I see. This error comes around once in a blue moon and certainly not on any of my home build stock PCs for a very long time. All of my systems use Intel, Asus and Corsair for the main building blocks. Memory errors are a lot more common on works cheap Dell PCs.

 

[Lonyo]

Simple way to find out is to undervolt your RAM, and then you will see.

 

[Phynaz]

you'll have a lot more crashes with programs and you'll probably get some nice windows error popups saying it's unable to access memory location blah blah. if that happens, you run a program like memtest86+ to find the offending DIMM.

ECC doesn't fix defective chips. A defective ECC DIMM will behave the same way as as a regular DIMM.

 

[VirtualLarry]

adding ECC to desktop would be a huge mess. have you ever had to build a server, or upgrade memory in one? you have to get just the right number of ranks or it's not even going to POST. you have to actually RESEARCH what will and will not work on some server boards. i made the mistake a couple years ago thinking it was all the same and ordered the cheapest ECC with ranking of x2. turned out i needed minimum of x4.

imagine throwing this at a consumer. i go to microcenter and i see customers staring at the memory for like 5 minutes confused about just the SPEED of the memory. then buffered vs unbuffered? customers would stop building PCs and just happily over pay for apple.

It used to be that way back in the day too, when buying SDRAM.

Single-sided or double-sided? Low-density or high-density? PC100 or PC133? 128, 256, or 512MB? You had to know the limitations of your memory-controller on your motherboard.
Mobo manuals had guides, on how to install SIMMs/DIMMs, depending on single- or double-sided sticks.

 

[VirtualLarry]

For someone to doesn't know the ins and outs of this subject, what happens with errors on a standard desktop PC? My RAM doesn't have ECC but I don't ever notice any errors?

Without ECC, that's the thing, you CANT detect the errors, unless they are blatant, and cause crashes. (Errors in code pages.) Errors in data pages, will be happily ignored, and slowly but surely they will cause silent data corruption, in your programs (hope you're not doing your taxes!), on your HD, in your graphics, etc.

 

[Cerb]

ECC doesn't fix defective chips. A defective ECC DIMM will behave the same way as as a regular DIMM.

Really? You'll be able to see logs of errors with a non-ECC DIMM? I have had an ECC DIMM that just locked a machine up, but never gave data errors, so yes, it happens, but that's not the general case.