Ecc Ram

[-=[U.D]=- Raverbaby]

Right i know you get ECC and non ECC ram and that ECC is usually for servers, but whats the difference?!

 

[fatbusinessman]

ECC basically means that the memory has built-in error checking on it, so instead of throwing out a duff bit occasionally (which is itself rare anyway) it'll catch that and try again.

As errors are so rare in any half-decent brand of memory, ECC isn't worth the extra expense for anything other than a high-performance (and probably safety-critical) server.

 

[Shovel]

It's also worth noting that a lot of consumer level motherboards don't actually support ECC. I think that still holds true.

Essentially as Fatty says, it's really only for so called "mission critical" systems, where a single error would cause catastrophic system problems (read: financial loss/environmental disaster).

You can also argue against it for hardcore gaming PCs - since the fractional amount of time it takes to do the error check would - technically - decrease the performance of your system. Albeit by a proportion so small only Wij could... ahem.

 

[Gurnox]

It's also worth noting that a lot of consumer level motherboards don't actually support ECC. I think that still holds true.

Essentially as Fatty says, it's really only for so called "mission critical" systems, where a single error would cause catastrophic system problems (read: financial loss/environmental disaster).

You can also argue against it for hardcore gaming PCs - since the fractional amount of time it takes to do the error check would - technically - decrease the performance of your system. Albeit by a proportion so small only Wij could... ahem.

If you want to buy an AMD FX-51, you need ECC memory. Otherwise, if you can cope with your machine crashing once in a while, don't bother. It's not worth the extra expense and will give you a, very minor, performance hit.

 

[Shovel]

Out of interest, what actually happens if a memory error occurs in a non ECC system? Does it just cause the program running to stutter and do it again?

 

[Gurnox]

Out of interest, what actually happens if a memory error occurs in a non ECC system? Does it just cause the program running to stutter and do it again?

Without going into too much detail, it would normally crash the system or, at best, the application addressing the affected memory space.

Bad things, basically.....

 

[Xavier]

If you want to buy an AMD FX-51, you need ECC memory. Otherwise, if you can cope with your machine crashing once in a while, don't bother. It's not worth the extra expense and will give you a, very minor, performance hit.

No, you need non ECC registered memories for the A64FX, you should only really use ECC for opteron.

*pokes at 2Gb non-ecc registered DDR400 in Athlon64 FX-51 beside him*

 

[Gurnox]

No, you need non ECC registered memories for the A64FX, you should only really use ECC for opteron.

*pokes at 2Gb non-ecc registered DDR400 in Athlon64 FX-51 beside him*

Oops. I bow before Xavier and beg forgiveness for my sins

 

[Xavier]

Without going into too much detail, it would normally crash the system or, at best, the application addressing the affected memory space.

Bad things, basically.....

Not quite true, ECC is capable of detecting and correcting single-bit errors, but anything more than that will still take an system using ECC memories out...

ECC isn't the be-all and end-all of memory stability, Chipkill is the next step up, consisting of augmented ECC memories... improving stability drastically.

An impartial piece on ECC, with a little info on Chipkill is available here.

Xav

 

[Danya]

Makes me glad I bought non-ecc instead of splashing out.

 

[fatbusinessman]

Out of interest, what actually happens if a memory error occurs in a non ECC system? Does it just cause the program running to stutter and do it again?

Strictly speaking, the answer to this is "it depends on the program". If the program has been coded to be robust, then it'll probably just cause a stutter. However, I suspect very few programs are coded to be robust.

 

[Danya]

Depends what you mean by "coded to be robust". It also depends what get corrupted. If the corruption occurs in the program code you're pretty much fucked, it's going to crash and burn. If it's in the data you might get away with it either generating an incorrect result, having a slight display issue, or nothing happening at all. Or it might crash and burn. Checking for bad memory is not something most programs do due the the infrequency of occurance and the overhead of actually doing it.