Is today's memory more reliable than memory of the past? (desktops, operating system)

[eok]

There was a time in the past when all PC's had parity memory, and some of them had ECC memory. Was that because the memory was unreliable, and they added parity and ECC to try to reduce the unreliability? Does it imply modern memory is so reliable that it no longer makes sense to have parity or ECC?

What do operating systems, such as Windows 10, do with parity and ECC? Do they gracefully manage the parity errors, and report them to the user? Or do they just ignore them? ECC can correct a lot of errors but not all of them. Does Windows 10 ignore the ones that can't be corrected? Do most versions of Linux do it the same way as Windows 10?

What about cosmic rays? The more powerful, the rarer? How rare are the ones powerful enough to change the contents of a PC's memory? Are they rare enough that it doesn't make sense to have ECC or parity just to defend against cosmic rays? Even if you have so much memory that it multiplies your risk of being affected?

[Zymer]

Originally Posted by eok

There was a time in the past when all PC's had parity memory...

This is not correct. Parrity/ECC memory requires an extra bit and is therefore more expensive. Many motherboards did/do not support it. Typically, this was only used in machines that required a high degree of reliability.

Use of parity/ECC does seem to be increasing though, possibly due to the lower manufacturing cost (compared to what it was in the 90s, for the sizes you could get).

[eok]

Look at the specs for the original IBM PC, which came out in 1981, before any clones came out. It had 9 bit parity memory. The first clones copied it, so they had 9 bit parity memory too. Therefore, what you wrote is not correct.

[Zymer]

So, you want to go *that* far back, do ya? OK, I'll give you that one. I don't have to check the specs, I've still got one or three of them hanging around. But, I was in the business of building systems in the hey-day of the 'clone wars' and when everybody was trying to shave a few bucks off cost, the difference in the cost of parrity/ECC chips and MoBos and non made that a good place to cut. [Most] Average consumers didn't know or care. (After a hard (and costly) lesson I always fully informed my customers about such differences, and advised against making certain choices just to save a few bucks. I had a rep for building custom systems and I told them they were going to pay a little more, if they wanted to 'cheap out' I sent them elsewhere because it wasn't worth it to me- in the *very* beginning I tried to go uber-cheap to compete, but in my first year I ended up *replacing* every single cheap machine, and spending so many hours on 'tech support' calls that I wasn't making any money, I lost money actually. I wised up pretty quick.) "The bitter taste of poor quality remains long after the sweetness of low price has faded."

[eok]

When Intel came out with the first 8 bit chipset for home PC's, which was a long time after the first PC, they said the reason for omitting the parity bit was because most users didn't know what to do with parity error messages, and were confused by them. Intel said a home PC user would typically be doing something like playing a game. The parity error message would interrupt their game. But a random bit error in their PC memory, had a good chance of not interrupting their game, because it might be in an unimportant part of the memory.

But why did people then start using non-parity memory for office applications, etc.? Are memory errors in those considered unimportant too? If the data in memory includes accounts, etc., and that data gets written to a database, an account error might not be discovered till years later, and there might not be any way to know where it came from, or to even prove it was ever wrong.

[thelopez2]

I though memory errors resulted in BSOD and "Memory can not be written" or something like that message.

As far as when did it stop becoming standard on desktops, I'd say about the time AST disappeared around 1996. ECC is still common on Servers and Xeon Workstations.

[markjames68]

Originally Posted by eok

Look at the specs for the original IBM PC, which came out in 1981, before any clones came out. It had 9 bit parity memory. The first clones copied it, so they had 9 bit parity memory too. Therefore, what you wrote is not correct.

Your assumption about parity is correct, but having ECC/parity memory is not.

IBM motherboards (like the 5150, the first PC) had 9 separate socketed chips per bank (except for the first bank which was soldered). None of the chips themselves had parity memory, there was a parity chip. IBM motherboards could tell you if there was a memory error which chip/bank it was due to - chip 0-7 or the parity chip.

When memory modules came into play (i.e. SIMM or DIMM memory) - essentially a bank per module - they could then either have a parity bit or not, which means potential ECC functionality for those with parity, non-ECC for those without.

[markjames68]

To your other point about cosmic rays (which are rare) theoretically two bits could get zapped and then even a parity bit wouldn't help.

I do believe that manufacturing has improved, so memory errors are more rare, but nonetheless it's good practice to run a full Memtest after installing memory. At home none of my PCs have ECC memory except for one server that's on 24/7/365. For a regular PC that's why people should save their documents and reboot periodically. That's my experience based upon over 25 years of building and repairing PCs.

[davebarnes]

Is today's memory more reliable than memory of the past?
No.
Core memory is very reliable and does not lose data when the power is removed.
https://en.wikipedia.org/wiki/Magnetic-core_memory