ECC common error rates

Kevin C

n00b
Joined
Mar 22, 2016
Messages
5
Does anyone have some statistics about (soft/hard) error rates in today's RAM?

Even better, if it is sortable by memory site, litography feature pitch, metres above sea level...
 
Thanks for the article, it was quite an interesting read. It suggests that memory errors are quite common. The only problem is that the article is a bit old (DDR1 and DDR2 DRAMs tested).
 
Thanks for the article, it was quite an interesting read. It suggests that memory errors are quite common. The only problem is that the article is a bit old (DDR1 and DDR2 DRAMs tested).
yea thats what i thought. It seemed really old...no clue if RAM error rate is the same on all RAM :/

I'll probably read it on the plane over monday.
 
It does appear that ECC is on the decline in areas that it used to be commonplace like Workstations etc. Personally all ram should be ECC equipped IMO.
 
It does appear that ECC is on the decline in areas that it used to be commonplace like Workstations etc. Personally all ram should be ECC equipped IMO.
if they only had fast ECC :/ The MTs and timings are total trash. They have 2/3s the bandwidth and double the latency of fast RAM -_-
 
if they only had fast ECC :/ The MTs and timings are total trash. They have 2/3s the bandwidth and double the latency of fast RAM -_-

I believe this is due to not overclocking the ram. What I mean with ECC ram you typically get official JEDEC timings not timings that are overclocked by the module maker.
 
I believe this is due to not overclocking the ram. What I mean with ECC ram you typically get official JEDEC timings not timings that are overclocked by the module maker.
its fine if they are not 3200 MTs but the timings are terrible. I don't recall JEDEC issuing timings but i could be wrong. I don't see why they haven't dropped a little with their timings. I would pay extra for sure.

2133 CL 15 is garbage. I would be happy with CL 10 :/ but 15 is rough. That is 40-60% higher latency verse regular desktop.
 
I wouldn't worry about ECC unless you modify large quantities of data. The only chance for permanent damage from a bit flip event is if you're saving the result to disk. Otherwise, the worst that will happen is you crash the system. And that's not happening any more often than it was ten years ago, despite DRAM density increasing by a factor of 32!

You should only be worried about the absence of ECC if you're running long-duration mathematical models or editing movies (i.e. multi-core workstations), or else in-memory databases with serious up-time requirements (i.e. server). For everything else, the majority of your I/O time is spent reading from disk. Memory errors are much less likely to hit you then.
 
Last edited:
if they only had fast ECC :/ The MTs and timings are total trash. They have 2/3s the bandwidth and double the latency of fast RAM -_-


Well I guess a drop from 233 FPS to 229FPS could be a worry for a few.

The articles and tests I've read in the past comparing show at worst a 2% drop and at best a 1% drop over standard ram.
 
Last edited:
Well I guess a drop from 233 FPS to 229FPS could be a worry for a few.

The articles and tests I've read in the past comparing show at worst a 2% drop and at best a 1% drop over standard ram.
its not gaming that i care about its encryption and a number of other things that RAM has an impact on. Its just silly to have 40-60% more latency. 2133MTs is fine (could be better) but the latency is crap.
 
its not gaming that i care about its encryption and a number of other things that RAM has an impact on. Its just silly to have 40-60% more latency. 2133MTs is fine (could be better) but the latency is crap.

I know where you are coming from but I think you are over worrying about it. I doubt you would notice.;)
 
Back
Top