[Shike]

Quote:

Originally Posted by linuxworks /img/forum/go_quote.gif that's really old fashioned thinking, that only 'business big machines' need data integrity protection.

Hardly. It's still modern and for good reason.

Quote:

I have no idea what agenda the ram companies have (that you claim) but they certainly have no business talking people out of this benefit. perhaps their margins are higher for non-ecc ram or that they have a lot more of it in stock and they want to push that? those are sales entities. you want to trust them to advise you??

This is a joke if I've ever heard one. Conspiracy theory much? If anything ECC based memory has a higher profit margin.

Quote:

the google article does show that ram errors are not 'unusual' and they happen all the time.

Yeah, which is associated with motherboards and not the RAM. Let's not exclude the fact that they use cheap hardware and run hotter than normal.

Quote:

putting your head in the ground and ostriching is not helping matters.

I'm not. Why must you plug your ears?

Quote:

do you drive without your seatbelt? you're only going "down the street", so why even use it? what are the *chances* of getting hit in such a short distance?

Invalid analogy on so many level, I'm not even bothering with it.

 

[FallenAngel]

Quote:

Originally Posted by Shike /img/forum/go_quote.gif How about practically every RAM manufacture saying ECC is only necessary for commercial servers? Including Crucial, which I linked to earlier in this thread.

"Only necessary", hmm... lets think about that. Could it be because it's only "necessary" when it actually matters? Well, I'd say that simply assumes that environments outside of "commercial servers" are OK with a few mis-calculations here and there, an application slowdown, little application error or crash, no big deal... it's not like you're doing anything that actually matters right? I'd say that's exactly what they mean by "only necessary for commercial servers"; non-critical environments, like "normal" PCs, can use any memory they like because they aren't doing anything important enough to want data accuracy.

Quote:

Originally Posted by Kurotetsu /img/forum/go_quote.gif What most posters in this thread are arguing is "Go ahead and buy ECC memory only if there's no difference in price against non-ECC". What linuxworks is arguing, and what everyone is against, is that ECC memory is REQUIRED for a home desktop computer and should be purchased regardless of price. People are saying to stay away from it because, usually, ECC memory is quite a bit more expensive than non-ECC memory, and that premium isn't justified for a home computer.

I don't think so, I haven't read anything of the sort from him. I only see somebody arguing that if a person wants fault-tolerance and data integrity from their ram because they actually "care enough for it" should use ECC memory. Sorry if I'm starting to sound condescending or like a jerk; when people start being rude and throwing around uneducated opinions and ignorant assumptions (not saying you do here), it brings that out in people.

Nobody is saying that ECC is "required", but it's designed for a reason it's a valid reason.

As for significantly more expensive - that was years ago. You can pick up 4GB (2x 2GB) of PC6400 ECC memory from NewEgg for $76, comparing to non-ECC, I'd say you're not losing any money.

 

[Shike]

Quote:

Originally Posted by FallenAngel /img/forum/go_quote.gif "Only necessary", hmm... lets think about that. Could it be because it's only "necessary" when it actually matters? Well, I'd say that simply assumes that environments outside of "commercial servers" are OK with a few mis-calculations here and there, an application slowdown, little application error or crash, no big deal... it's not like you're doing anything that actually matters right? I'd say that's exactly what they mean by "only necessary for commercial servers"; non-critical environments, like "normal" PCs, can use any memory they like because they aren't doing anything important enough to want data accuracy.

You're assuming. You also don't seem to comprehend that the errors rarely if ever occur on a consumer machine without a memory module going bad.

Consider this. Someone that cares about data integrity should be doing regular backups anyway. On the occasion that something valuable is corrupted (very rare if RAM is good, we're talking less than winning the lottery), they have a backup anyway.

The only time these errors become statistically significant to want and need to stop them is when using commercial servers where there's a better chance it will happen and has a large monetary value attached to need and avoid downtime (resorting to backups).

Quote:

As for significantly more expensive - that was years ago. You can pick up 4GB (2x 2GB) of PC6400 ECC memory from NewEgg for $76, comparing to non-ECC, I'd say you're not losing any money.

It costs more and you do lose performance.

 

[Kurotetsu]

Quote:

Originally Posted by FallenAngel /img/forum/go_quote.gif Nobody is saying that ECC is "required", but it's designed for a reason it's a valid reason.

Of course its valid...for servers, workstations, and any system that is running mission critical software. Something that home desktop computers don't do. And as linuxwork's own link pointed out, memory errors are more likely to occur due to a shoddy motherboard, not because of non-ECC memory.

Quote:

As for significantly more expensive - that was years ago. You can pick up 4GB (2x 2GB) of PC6400 ECC memory from NewEgg for $76, comparing to non-ECC, I'd say you're not losing any money.

Of course, you also need a motherboard that actually supports ECC. The cheapest one on Newegg with explicit support for ECC memory (using Newegg's search function) is this one:

Newegg.com - ASUS P5BV-C LGA 775 Intel 3200 ATX Intel Xeon Server Motherboard - Server Motherboards

for $155. I bought my motherboard, a Gigabyte board with Ultra-Durable 3 which has all the features and expansion I could ever need, for $100. The cheapest non-ECC DDR2 4GB memory kit on Newegg is this:

Newegg.com - Crucial Ballistix Tracer 4GB (2 x 2GB) 240-Pin DDR2 SDRAM DDR2 800 (PC2 6400) Dual Channel Kit Desktop Memory Model BL2KIT25664AR80A - Desktop Memory

For $61. So that's a total $70 premium for ECC memory. Admittedly, that's not a whole lot, and for someone who thinks they need it I guess its worth the money (but, frankly, if data integrity is THAT important to you, you'd be better off buying a pre-built workstation using server-grade parts). But, personally, I'd rather use that $70 for CDs or a backup hard drive.

 

[Ntropic]

Ballistix stopped being good after 1Gb D9GMHs stopped being produced. A moot point now, however.

RIP my beautiful 16FD3s.

 

[chris719]

I find this thread hillarious. DRAM bit-error rates are ridiculously low. If you aren't doing extremely high precision floating point calculations where you use the result to make a safety critical decision you don't need ECC.

ECC is single-bit correction and not really all that useful, it's there more for historical purposes, because it's easy to implement, and it allows for higher profit margins per DIMM. If you really cared about integrity you wouldn't be using unregistered DDR anyway.

But it's ok, I'm sure some Linux fanboy / IT guy knows more than the engineers that work for JEDEC member organizations.


If you understand how memory addressing works, then you'll understand why the likelihood that a machine is STABLE when it in fact it has a significant number of bit-errors is really small. In the case that an error occurs, I'd guess the probability of an invalid branch target address is millions of times higher than corrupting even a single-bit on your CD rip.

 

[Eagle_Driver]

Quote:

Originally Posted by sonci /img/forum/go_quote.gif OK I just upgraded my PSU to a bequiet! brand, and its really good, with very stable rails, in iddle or loading, I was thinking of upgrading my 1Gb RAM, to at least 2gb, and since my board support ECC nonREG RAM, I was thinking of going for that, cause old DDR ECC is really cheap, Would you think I`ll get any advantage or I`m just slowing down my pc, I use Foobar2k with a very large library, would error checking matter somehow

In your particular application, ECC isn't necessary. It's only with very large databases (not music) that ECC might be helpful.

Also, no current system uses DDR memory any more: The currently popular memory types are now DDR2 and DDR3 memory--neither of which is compatible at all with your current motherboard. This is the reason why DDR memory now costs double that of either DDR2 or DDR3 memory on a per-GB basis. Memory for old systems is relatively expensive compared to that for newer systems right now.

 

[sonci]

OK,
on my side I got the ECC Kingstone which was 5$ more then the non ECC brother, though this thread is really interesting,
I never thought of the ripping problem, but that is a good point,
To all people that find strange worrying about ECC Ram, should read this thread>
http://www.head-fi.org/forums/f9/doe...en-you-438739/

I mean there are people here, who rip the same CD every year, because windows files degrades

, why not get the ECC?

 

[linuxworks]

Quote:

Originally Posted by Eagle_Driver /img/forum/go_quote.gif In your particular application, ECC isn't necessary. It's only with very large databases (not music) that ECC might be helpful.

ECC is helpful anytime DATA (of any kind) sits in ram. while in ram it can get 'hit' with errors and if you have no detection/correction, you will never ever know.

"I never installed ECC and my system is fine" I hear people say. how do you KNOW? you don't know. that's why you buy insurance (ecc) to add an extra level of protection.

every single bit 'counts', of course. and the duration of how long data stays in memory before being written to disk is not relevant; data errors could occur in seconds (or less) after a cd track (block) was read into ram and before its synced back to hard disk.

it makes sense to care about data being written to disk (as, once you have a bit error, it 'stays' that way on disk and won't ever fix itself) but also it makes sense to care about data just *in transit* from disk to memory and then out to spdif or a dac. even though the bit error that might occur is not being 'archived' to disk, it still is a bit error and will affect playback thru the dac.

there was a time you could argue that ECC was not *cost effective* but you could never argue that it wasn't *effective* or useful.

now, cost is not the issue since the differential is so small. and so, for chipsets that support it, you can have extra data integrity protection for almost no extra cost.

seems prudent to avail yourself of all things that could possibly help keep data intact.

system uptime is irrelevant. data errors don't "wait" for months just to strike. they can happen even seconds after a cold boot.

 

[linuxworks]

Quote:

Originally Posted by chris719 /img/forum/go_quote.gif I find this thread hillarious. DRAM bit-error rates are ridiculously low. If you aren't doing extremely high precision floating point calculations where you use the result to make a safety critical decision you don't need ECC.

so, something very small 'doesnt count' as far as you're concerned?

"almost" is good enough for you? seriously? you really think that way?

and, floating point - what on earth does that have to do with this discussion?

THAT is the hilarious part; that you think that FP at all enters into this.

 

[linuxworks]

Quote:

Originally Posted by Shike /img/forum/go_quote.gif You're assuming. You also don't seem to comprehend that the errors rarely if ever occur on a consumer machine without a memory module going bad.

you are really wrong, here.

memory modules don't have to be 'bad' to get alpha particle hits.

ref: The effect of alpha-particle-induced soft errors on memory systems with error co

the ram is fine but they still get bit-errors. run memtest86 and it passes. hmmm, what's up with that? its not the ram, its the environment around it.

Quote:

Consider this. Someone that cares about data integrity should be doing regular backups anyway.

you miss the point; by the time the memory bit-error occurs, you wont' even know and you now blindly write that BAD DATA to disk? you've actually made things worse, dont' you see that?

backups != data integrity

Quote:

The only time these errors become statistically significant to want and need to stop them is when using commercial servers where there's a better chance it will happen and has a large monetary value attached to need and avoid downtime (resorting to backups).

again, you fail to understand. 'large money' and 'backups' are not germane here. this is ALL about quality of data; why should only 'big rich guys and companies' have hardware based data integrity subsystems?

re: speed, HP has this to say:

ECC memory chips add an additional hunk of DRAM (dynamic RAM) for every 8 existing DRAM parts to accommodate data for error correction. Every time 64 bits of data of standard data are transferred to the memory controlled, 8 bits of this ECC data tag along. The ECC circuit looks for errors in the 8-bit block; when it finds them, it performs error correction on the 64-bit chunk as well.

This checking incurs no performance penalty in the latest chipsets. In addition, any 2-bit errors and multi-bit errors (for most failure mechanisms) are detected, and system shutdown is initiated.

A look under the hood - ECC memory

arguing about cost and performance 'penalty' of ecc is a red herring, these days.

 

[xnor]

Cool, if you happen to have ECC RAM in your desktop machine please let us know when it shuts down while you're listening to music.

I have never heard or seen something like that...

 

[linuxworks]

Quote:

Originally Posted by xnor /img/forum/go_quote.gif I have never heard or seen something like that...

could be that there are MANY things you have also never heard of?

please stop making recommendations you know little about, then. you do the readers a disservice.

 

[Shike]

Quote:

Originally Posted by linuxworks /img/forum/go_quote.gif you are really wrong, here. memory modules don't have to be 'bad' to get alpha particle hits. ref: The effect of alpha-particle-induced soft errors on memory systems with error co the ram is fine but they still get bit-errors. run memtest86 and it passes. hmmm, what's up with that? its not the ram, its the environment around it.

No, you're quoting an article from the 1980's when ECC was a necessity to run properly. We're past those days.

Quote:

you miss the point; by the time the memory bit-error occurs, you wont' even know and you now blindly write that BAD DATA to disk? you've actually made things worse, dont' you see that?

Then the files are corrupt, and need to be restored. Simple as that.

Quote:

backups != data integrity

Sure it does, if done properly.

Quote:

again, you fail to understand.

Hardly.

Quote:

'large money' and 'backups' are not germane here. this is ALL about quality of data; why should only 'big rich guys and companies' have hardware based data integrity subsystems?

Because there's no need for it. You have yet to provide statistical analysis of why a consumer computer needs it, and have tried to remove context from various studies to further your false argument.

Quote:

re: speed, HP has this to say: ECC memory chips add an additional hunk of DRAM (dynamic RAM) for every 8 existing DRAM parts to accommodate data for error correction. Every time 64 bits of data of standard data are transferred to the memory controlled, 8 bits of this ECC data tag along. The ECC circuit looks for errors in the 8-bit block; when it finds them, it performs error correction on the 64-bit chunk as well. This checking incurs no performance penalty in the latest chipsets. In addition, any 2-bit errors and multi-bit errors (for most failure mechanisms) are detected, and system shutdown is initiated. A look under the hood - ECC memory arguing about cost and performance 'penalty' of ecc is a red herring, these days.

Does HP manufacture the RAM? What benchmarks have they provided for ECC of incurring no performance penalty?

That's what I thought.

 

[linuxworks]

I'm done with 'shike'. he seems to think backups are an integrity fixer. his profile says 'student' and so I'll assume he is one of those kids who thinks they have 'been around' a lot. clearly you're too young to have enough experience to lecture others about.


game over. but thanks for playing.

I'm outta here. enjoy your silent bit-errors, guys.