Nullack
10-22-2007, 02:11 AM
Hey everyone. I was interested in seeing Tony post in another thread that an overclocker using 8GB at that RAM density should be very careful about the chance of it erroring out. Another guy was recommending ECC RAM. Xtreme Systems has alot of good info and I hope I can better understand the situation from the experienced amongst you.
With 4GB and 8GB RAM densities the rate of soft error failure increases greatly over say 1GB. I have an IBM research paper that looks at this. I havent found anything else thats reliable and actually describes how the research was done. When people were talking about ECC ram being unecessary for desktop computing it was back when there was 512mb typically not 4gb to 8gb. And for those of you that are as old as I am, you would remember all personal computers had parity checking and that ECC is an extension of that. I fully accept the design of the modules was bad back then and they succumbed to radiation alot more than modern ones but my point remains that this used to be handled before vested interests decided to make a small extra bit of money on pc sales instead of selling reliable computers.
1. So does anyone have good solid research they can point too about exactly what the typical metrics are on the frequency of soft errors in modern DDR2 modules?
2. Has anyone tried overclocking unbuffered ECC RAM modules (electrically I cannot see why not but it would be pot luck from Crucial and Kingston)?
I have some scientific and engineering tasks I use my workstation for that makes ECC RAM nice. What would be terrific for me is to be able to shutoff the ECC in the BIOS for benching or gaming (dont see me bothering alot for gaming given its only 1%-3% performance overhead but anyway) and then putting it back on for real work during the day. I have my eye on the X38 chipset for this.
Thanks in advance :)
With 4GB and 8GB RAM densities the rate of soft error failure increases greatly over say 1GB. I have an IBM research paper that looks at this. I havent found anything else thats reliable and actually describes how the research was done. When people were talking about ECC ram being unecessary for desktop computing it was back when there was 512mb typically not 4gb to 8gb. And for those of you that are as old as I am, you would remember all personal computers had parity checking and that ECC is an extension of that. I fully accept the design of the modules was bad back then and they succumbed to radiation alot more than modern ones but my point remains that this used to be handled before vested interests decided to make a small extra bit of money on pc sales instead of selling reliable computers.
1. So does anyone have good solid research they can point too about exactly what the typical metrics are on the frequency of soft errors in modern DDR2 modules?
2. Has anyone tried overclocking unbuffered ECC RAM modules (electrically I cannot see why not but it would be pot luck from Crucial and Kingston)?
I have some scientific and engineering tasks I use my workstation for that makes ECC RAM nice. What would be terrific for me is to be able to shutoff the ECC in the BIOS for benching or gaming (dont see me bothering alot for gaming given its only 1%-3% performance overhead but anyway) and then putting it back on for real work during the day. I have my eye on the X38 chipset for this.
Thanks in advance :)