Re:Comparing DDR5 Memory from Micron, Samsung, SK Hynix
by sliken on Tuesday February 15, 2022 @11:33PM
(#30356395)
Heh, dunno, the rates I've seen likely are 100x or lower than other common cause of mistakes, corruption, and confusion.
Do I want ECC, sure. Would bitflips ECC could prevent be in any in the top 10 reasons in year for file corruption, application crashes, and OS crashes etc... unlikely.
Re:Comparing DDR5 Memory from Micron, Samsung, SK Hynix
by notriddle on Wednesday February 16, 2022 @12:24AM
(#30356643)
You’re not wrong, but that’s not really the point. The rest of the computer industry needs serious reform, too. That doesn’t take away from needing ECC in “consumer” products.
Re:Comparing DDR5 Memory from Micron, Samsung, SK Hynix
by Dylan16807 on Tuesday February 15, 2022 @09:39PM
(#30355742)
You don't fix rowhammer with ECC, you fix it by tracking row activation counts and then refreshing nearby rows. Without any cheap-outs like using small hash tables.
Re:Comparing DDR5 Memory from Micron, Samsung, SK Hynix
by Dylan16807 on Tuesday February 15, 2022 @10:20PM
(#30355995)
I've read that paper, yes. It works by bypassing target row refresh, because all those chips are trying to be clever about their row tracking and failing.
Track row activation counts for real, no estimates and no hash tables, and you're not vulnerable to this.
People don’t care, because people don’t know. RowHammer-vulnerable RAM is like canned food with botulinum and gasoline with lead. In 100 years, we’ll all be flabbergasted that everyone in the industry wasn’t arrested for letting it happen.