Has Anyone Ever Overclocked ECC Memory?

[Gas Pipe Jimmy]

Getting good information on this question seems to be pretty difficult so I'll ask here.

1. Are there any newly-realized reasons to now go with ECC RAM in a non-server environment?
I have heard the arguments that it "slows a machine down", but in looking at some hard data on that, it seems to only make a difference of between 1%-2% in a modern computer.
Also, Microsoft, about a year ago, "encouraged (http://www.tgdaily.com/2006/02/06/ecc_for_vista/)" PC system builders to use ECC RAM in new Vista-Ready machines as a way of increasing system stability on PCs.

I also just read a paper from 2002 that states that a machine running with 2.5 GB of RAM can expect a soft-error in every 75 hours of use. If you leave your machine on 24/7 that's about once every three days or about 120 of them a year with effects ranging from not noticeable to "shredding the contents of your harddrive (http://cracauer-forum.cons.org/forum/viewtopic.php?t=18&sid=68db92cab88a1887f4f930c8c812129b)".
Here is the paper titled: Modeling the Effect of Technology Trends on the
Soft Error Rate of Combinational Logic (http://www.cs.utexas.edu/users/cart/publications/dsn02_dist.pdf).
I was wondering if on an overclocked machine if ECC would be more of a help than a henderence?

So, has anyone here recently used ECC RAM in a PC, and if so, were you able to overclock it or did you have any problems other than an extended or long boot up time?
Might ECC RAM help with stability on overclocked memory?

What's the ground-truth on this one?

[Kab]

Normal SDRAM DDR or DDR2 ECC RAM is not made for speed past its rating, but made for parity checking thus used in crucial environments i.e. servers, and for stability to eliminate any faults. Due to the extra error correcting code chip of 8bit its definitely slower anyway as all the data that runs through it is checked and fixed (mostly). I haven't heard of anyone wanting to OC yet go for ECC RAM - performance is attained with the "performance" suited RAM which is made specifically for OC'ing allowing extra voltages, heat etc withstanding. ECC RAM also costs a lot more than normal RAM, available in limited forms and I don't see many mobos that support it and support OC'ing it too. I doubt you'd get much OC with ECC RAM anyway TBH --

I've tried ECC RAM yes many times, different forms (i.e Rambus which is pretty quick). But never to OC no and its slower than a $300 XMS2 PC2-6400 definitely. These sticks are made especially to OC and run faster than 6400MB/s. ;)

I couldn't answer further to than that :) Why is it you're after ECC RAM strictly? Is this for a server or workstation?

[Stranger]

I was wondering about this issue myself--more out of random curiosity than anything else--its a bit like putting racing slicks on a schoobus--kind of getting away from the "safety first" purpose of using ECC Ram in the first place.

[Gas Pipe Jimmy]

I've tried ECC RAM yes many times, different forms (i.e Rambus which is pretty quick). But never to OC no and its slower than a $300 XMS2 PC2-6400 definitely. These sticks are made especially to OC and run faster than 6400MB/s. ;)

I couldn't answer further to than that :) Why is it you're after ECC RAM strictly? Is this for a server or workstation?

Thanks! I apperciate the detail in your answer.

I was wondering about this issue myself--more out of random curiosity than anything else--its a bit like putting racing slicks on a schoobus--kind of getting away from the "safety first" purpose of using ECC Ram in the first place.

I know the two concepts (speed vs. safety) might sound mutually exclusive but if I can get 2GB of PC2-6400 without too much of a price penalty ($25 a stick), on a board that will support ECC memory AND overclocking (Intel 975BX2), then unless the things won't over clock at all, I would have the best of both worlds. I was just wondering if that might not be one more safegurd against memory errors.
I just don't want to get the WORST of both worlds if I did this. :D

This might be a daft idea, but I'm just looking for concrete reasons why it would either be unworkable or counter-productive.

[Stranger]

This might be a daft idea, but I'm just looking for concrete reasons why it would either be unworkable or counter-productive.

Sometimes daft ideas are the best kind--fun, even if they turn out to be impractical. I am interested to know what you find out.

[Gas Pipe Jimmy]

I'll start calling memory vendors tomorrow and see what they say.
In the past, Micron/Crucial has been really cooperative in answering questions.

[Adamantine]

I had some Registered PC2100 ECC ram back in like 2001 when PC2100 was the fastest DDR available. I remember OC's with DDR1 was maxing out around 160 or so (I believe it was because the PCI bus wasn't locked and the 166FSB wasn't out yet), and I was able to get my ECC ram up to 155. I don't think the bitch ever crashed due to a RAM error either, just heat until I got a better HSF.

[uOpt]

1. Are there any newly-realized reasons to now go with ECC RAM in a non-server environment?



Not new, but if you don't want your harddrive contents scrambled.

http://forum.useless-microoptimizations.com/forum/ecc.html (ETA: you've already seen it)


I have heard the arguments that it "slows a machine down", but in looking at some hard data on that, it seems to only make a difference of between 1%-2% in a modern computer.



No. My benchmarks have been run with, ECC and ECC/registered and there is no noticable slowdown compared to other platform factors.

http://forum.useless-microoptimizations.com/forum/crabench.html
http://forum.useless-microoptimizations.com/forum/crabench-memory.html


Also, Microsoft, about a year ago, "encouraged (http://www.tgdaily.com/2006/02/06/ecc_for_vista/)" PC system builders to use ECC RAM in new Vista-Ready machines as a way of increasing system stability on PCs.

I also just read a paper from 2002 that states that a machine running with 2.5 GB of RAM can expect a soft-error in every 75 hours of use. If you leave your machine on 24/7 that's about once every three days or about 120 of them a year with effects ranging from not noticeable to "shredding the contents of your harddrive (http://cracauer-forum.cons.org/forum/viewtopic.php?t=18&sid=68db92cab88a1887f4f930c8c812129b)".
Here is the paper titled: Modeling the Effect of Technology Trends on the
Soft Error Rate of Combinational Logic (http://www.cs.utexas.edu/users/cart/publications/dsn02_dist.pdf).




That's nonsense. Last time I did the calculation based on Micron data of error rates I came to about 1 error/year.

For intact RAM, the real question is if one of your RAM sticks is slightly sick.



I was wondering if on an overclocked machine if ECC would be more of a help than a henderence?



Note that DDR/DDR2 ECC only protects the data bits, but not the control lines. If you overrun the control lines with speed, then you lose equally badly.

FB-DIMMS are fully protected but they aren't doing that overclocking thing very well anyway.



So, has anyone here recently used ECC RAM in a PC, and if so, were you able to overclock it or did you have any problems other than an extended or long boot up time?



I overclock my ECC permanently, but only slightly, 220-230 MHz instead of 200.

The real-world benefit of faster memory is very very small, it's really not worth the risk to do much more.

I basically pick the RAM speed closest to 200 MHz but not lower than 200 MHz for AMD64s.


Might ECC RAM help with stability on overclocked memory?



As I said, that's questionable since the control lines are not ECCed.



What's the ground-truth on this one?

Whatever I say :)

[Gas Pipe Jimmy]

Thanks for the great info uOpt!
I took your advice and made my purchases. :D