ECC memory & Socket A

[Guest]

I have been trying to determine which socket A motherboards are available that support ECC PC133 DIMMS. Does the Asus A7V ? I know that the manual does not say either yes or no... The Abit KT7-Raid I know does not support ECC memory, according to the manual. I need to know that the ECC function is actually implemented; I understand that some motherboards can use ECC memory but do not necessarily implement the error detection and correction functions. I actually need to have them function. My applications are long running, scientific number crunchers, its not just about losing a pixel here and there. I have been looking for a socket A board (for Athlon 1 G) that will support at least 1.5 G of memory and have ata100 ide drive support. Somthing along the lines of the A7V. I have emailed many board manufacturer sites, including ASUS, but have still (after weeks) had no reply. I don't expect that I will. In any case I would greatly appreciate any info on this. If some one knows for a fact that the A7V does support ECC memory and functions...let me know and I'll be out the door to get one. Thanks

 

[Guest]

I face the same problem: finding boards with ECC memory support.

I think you will not find a modern bd with an AMD proc and ECC memory. Most (all?) use VIA chipsets: the KT, KX and now the KM. The KT and KX do not support ECC; the KM (used on the Asus A7V) looks the same, although I have not found its full datasheet on Via web site <A HREF="http://www.via.com.tw " target="_new">http://www.via.com.tw </A>. Via has traditionnally put its effort on the mass no memory checking market.

I have not looked at the older AMD750.

We are left with Intel chipsets and procs. The following page gives the needed information <A HREF="http://developer.intel.com/design/chipsets/embedded/index.htm " target="_new">http://developer.intel.com/design/chipsets/embedded/index.htm </A>. Our choices are the 440BX, on the way out, or the new expensive 840.

Let us note that ASUS is offering CUSL2 bds as "Entry-Level Server" bds. This is rather bold and unusual, since servers by IBM, Compak usually have ECC memory.

Last September, Tom presented the 815/815E as "heir to the BX". Well, this is partly incorrect, because of the absence of ECC memory support on the 815/815E.

Please correct me, anyone, if I am wrong.

Gilles

 

[Guest]

The AMD750 chipset does support ECC, AFAIK, but this is only found on Slot-A motherboards and not Socket-As.

I've also had this exact same problem. I intend to build a Linux box and have been told that ECC is more important for Linux systems than Windows ones. Does anyone have any views on this? General forum opinion seems to be that ECC is unnecessary for home users. Does anyone disagree?

Cheers.

 

[Guest]

AFAIK, the problem ECC solves is NOT a function of the OS, but of the physical nature of our components, i.e., the loss/destruction of data on the memory chips due to oddities of nature (super novas and other such things I frankly don’t understand). Therefore, I don't see how Windows, Linux, or even DOS for that matter, are particularly relevant. In a nutshell, it’s the suseptibility of our components to the physical phenomenon X the number of those components X the amount data being pumped through them (i.e., exposure). More data and/or more components means more risk, thus 24/7 servers w/ stacks of RAM are prime targets. But nowadays, I keep my 256MB "client" workstation running 24/7 too!

The real issue is the “cost” in lost or corrupted files, inaccessibility, time-consuming troubleshooting, etc., for NOT using ECC for YOU. Frankly, under that criteria, I’ consider my workstation mission critical too! As CPU speeds increase and apps demand more RAM (>256MB especially), I believe there will be some threshold passed that will require ECC or similar protections.

As someone new to overclocking, I'm watching these guys push their Duron/Athlon on say, the Abit KT7/A, then having all kinds of unexplained failures. Seems to me (and perhaps I'm overly cautious) that using ECC during shakeout would remove one variable! Of course, I'd like the option to disable it for performance (about 2% hit) as well.

But apparently I'm in the minority. The topic has virtually no traction in the general community. And finding a board that supports it, among the “hobbyist” level anyway, seems rare.

I’ve always wonder just how many of those unexplained BSOD’s in Windows might have been related!

 

[stable]

Also be careful to realize that this is a tad different in the DDR RAM world. There is unbuffered NON-ECC, unbuffered ECC and then to top it all off REGISTERED RAM. While the registered RAM requires an additional clock cycle, we use ONLY registered RAM in Servers and in workstations requiring accurate intense mathmatical models and high uptime reliability.

It should be noted that while Asus claims that the A7M266 supports ECC RAM, currently IT DOES NOT. The system will either crash if you install it, or it just won't recognize the RAM as ECC (depending on the board you have in your hands).


Steve Benoit

Stable Technologies
'The way IT should be!'

 

[Guest]

> I’ve always wonder just how many of those unexplained
> BSOD’s in Windows might have been related!

Well, exactly. I've heard that memory errors are sometimes the cause of just such crashes in Windows. Regarding Linux, I have read on other sites that Linux uses memory more efficiently and therefore "pushes it" harder, making memory errors more likely. This may just be unqualified boasting by Linux lovers, though!

nædus.com