On Friday, August 21, 2015 11:00:16 AM Alan Grimes wrote: > Grant Edwards wrote: > > On 2015-08-21, Alan McKinnon <alan.mckin...@gmail.com> wrote: > > > >> Earlier I saw segfaults in gcc, and another poster pointed it out. > >> > >> When gcc segfaults, it is always suspicious mostly because the compiler > >> is an app where we know the devs take extraordinary measures to prevent it. > >> > >> The most common cause is faulty hardware (most often memory) as gcc > >> tends to use all of it in ways no other app does. The usual procedure > >> at this point is to run memtest for an extended period - say 48 > >> hours, or even 72 for an older slow machine. > > That is definitely good advice. I've run into this situation several > > times. A machine had bad RAM that didn't seem to cause any problems > > under "normal" operation. But, when trying to compile something large > > like gcc, I would see non-repeatable segfaults (it wouldn't always > > segfault at the exact same point). In those cases, I could often run > > memtest for several passes and not see an error. But, _eventually_ > > ramtest would catch it. Run memtest for a few days. Really. > > Yeah, I know there's a single bit error out at the end of RAM that will > appear on the third or fourth pass... > > I have already RMA'd half of the ram in this machine because it was > giving a whole fist-full of errors across two sticks... I run the rusty > old bus on the CPU ( SIX CORES!!!!) a bit harder than it was intended > in order to keep up with the new junk. My previous machine had ECC. =( > > I was advised to just jack the voltage a little bit and live with it. I > guess I'd better run more tests and see what the situation is.... > > It just doesn't seem reasonable to demand that every bit in a 32 > gigabyte memory bank be absolutely perfect....
LOL. It's perfectly reasonable. If it's under warranty, return it. And get a different brand cause it sounds like what you got is crap. If it's not under warranty and after running the test for an extended period as adviced you're sure that it's only a single bad bit at the highest end you can boot with the mem= option. Also see memmap and memtest. https://www.kernel.org/doc/Documentation/kernel-parameters.txt -- Fernando Rodriguez
