Do you need or want ECC RAM? (Poll)

[ActionableMango]

Whenever we talk about ECC RAM, we talk about 99.9999999% of people not needing it. I'm curious to see if anyone here actually needs it, and for those that do (if any), what software they use that needs ECC RAM.

For my own part I don't need it at all. Yet when I replaced my RAM, I replaced it with ECC RAM anyway. I can't really say why other than it's the "proper" RAM for a MP and I like the concept of self-correcting RAM.

For the purposes of this poll, I'd like to be generous in the definition of ECC RAM. ECC RAM can be ECC system RAM, ECC GPU RAM, a software emulated ECC function that uses non-ECC RAM (as is often the case with workstation video card drivers), or anything else that can reasonably be argued to serve the same function. In other words, the point is not about how the function is accomplished, but whether or not you need/want the function.

 

[MacVidCards]

We have offered the flashing of various Quadro and Tesla cards which have ECC ability. Almost NOBODY ever gets them.

So, anyone and everyone who needs this could have it for GPGPU but rarely do they.

 

[DoofenshmirtzEI]

It's a real pisser when you're tearing your hair out trying to nail down a phantom critical bug that showed up in your release build, and you can't because it's not in your code, it's caused by a bit flip.

ECC always and forever.

 

[zephonic]

I probably don't need it.

But I have never needed the airbags in my car, either. (touch wood)

I do feel better having both!

 

[flowrider]

^^^^With a Takata Airbag, your analogy is questionable - but yes, I feel the same way.

Lou

 

[AidenShaw]

It's a real pisser when you're tearing your hair out trying to nail down a phantom critical bug that showed up in your release build, and you can't because it's not in your code, it's caused by a bit flip.

ECC always and forever.

+1, and there's no question with ECC about whether random crashes (OS and apps) are due to memory problems.

 

[jimj740]

+1, and there's no question with ECC about whether random crashes (OS and apps) are due to memory problems.

Disclaimer: I like and use ECC memory...

But to be pedantic I will state that there are limits to correction and detection of errors with ECC, multi bit errors can occur that are neither corrected nor detected; albeit with far lower probability than single bit errors...

There are very few absolutes in this world!

 

[AidenShaw]

Disclaimer: I like and use ECC memory...

But to be pedantic I will state that there are limits to correction and detection of errors with ECC, multi bit errors can occur that are neither corrected nor detected; albeit with far lower probability than single bit errors...

There are very few absolutes in this world!

Agreed, but two things:

1. Corrected ECC errors will be logged, and visible from the BIOS or diagnostics utilities

2. I should worry more about an asteroid hit than an undetected multi-bit error (most multi-bit errors are detected, and usually have associated single bit corrected errors that are logged).

So, if I don't see panics or BSODs due to "Hardware ECC memory check error", and the logs don't show corrected errors - I think that we'll both agree that we should look at software bugs.

There was a study from Compaq long ago (probably about the time that 4 GiB was a huge amount of memory on a quad socket server) that said that one should expect one ECC error (a bit flip corrected by ECC) per month per gigabyte. This was expected with perfectly functioning hardware due to cosmic ray events. It was so long ago that it's not worth looking up, since memory chips now are much higher density.

I see this in my ProLiant servers - most of them have a front panel health display with LEDs for every component, including 24 LEDs for the DIMMs. Normally dark, but if there's an ECC error that DIMM indicator will light up yellow. I'll take note, clear the LED, and log the entry. If that DIMM on that system shows a history of ECC corrections, I'll investigate. Mostly, though, they're random and don't repeat.

BUT THE SINGLE BIT ERRORS DO OCCUR.

That's why I used my own dime to buy a Dell workstation for my home PC - I want ECC and don't want to worry about the RAM. (Six core E5-1650v2 - 128 GiB)

 

[jimj740]

[SNIP! edited for brevity]
There was a study from Compaq long ago (probably about the time that 4 GiB was a huge amount of memory on a quad socket server) that said that one should expect one ECC error (a bit flip corrected by ECC) per month per gigabyte. This was expected with perfectly functioning hardware due to cosmic ray events. It was so long ago that it's not worth looking up, since memory chips now are much higher density.

Actually the problem is much, much worse with higher density chips...

Intel has actually stated this as a likely limiting factor in chip design!

 

[ActionableMango]

Interesting results. I had expected most would say "no", a few people like me "prefer", and only one or two people with "yes, required".

 

[tomvos]

Well, the point is, ECC saves your day and you won't even notice it. Just think about the fact that many people say that their Mac Pro is their most relieable Macintosh. This reliability is not for free, it depends on certain components and reasonable default operating conditions (like a good power supply, sufficient cooling, etc.).

Many people here in this forum may be highly opinionated but they are not stupid. So many people here know that ECC did or will protect them from some problems. So the result does not surprise too much, doesn't it?

For the record: Yes, ECC is cool. I would even like to see these new mobile Xeon chips from Intel in a MBP. Which would allow ECC in the MBP (although I did not check if ECC RAM for notebook form factor even exists).