![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Hey all, I had an interesting experience and thought I'd share it.
I rebuilt one of my systems, using a DFI DR-Expert board and 2x512 OCZ gold, Athlon X2. Booted Windows to do the initial overclock (Windows OC tools are so much better than what's available in Linux). Found 2700 Mhz to be 24 hour Prime95 stable, no errors running Memtest, either. Then I installed Gentoo Linux...
Everything was great until I went to recompile the toolchain. For non-*nix users, the toolchain is the system compiler, libraries, and related utilities. Needless to say, the toolchain is the most important software on the system because everything else is built by it. Anyway, I got a file not found error when compiling gcc the first time. I couldn't find any config mistakes, so I went ahead and tried it again, and it worked. Suspicious...
A bit later, the box was in the middle of compiling glibc when it segfaulted. I was worried now, and I restarted the glibc compile. It was successful. I knew then that I had some kind of hardware problem.
To make a long story short(er), I had to clock the memory back down to stock in order to get a clean toolchain compile (turned out the CPU was fine at 2700). I believe that a toolchain compile is especially sensitive to memory errors, because the toolchain essentially must recompile itself at least twice, which would tend to magnify any problems with the generated binaries.
So memtest and prime don't mean a damn thing, really. Go figure.
I rebuilt one of my systems, using a DFI DR-Expert board and 2x512 OCZ gold, Athlon X2. Booted Windows to do the initial overclock (Windows OC tools are so much better than what's available in Linux). Found 2700 Mhz to be 24 hour Prime95 stable, no errors running Memtest, either. Then I installed Gentoo Linux...
Everything was great until I went to recompile the toolchain. For non-*nix users, the toolchain is the system compiler, libraries, and related utilities. Needless to say, the toolchain is the most important software on the system because everything else is built by it. Anyway, I got a file not found error when compiling gcc the first time. I couldn't find any config mistakes, so I went ahead and tried it again, and it worked. Suspicious...
A bit later, the box was in the middle of compiling glibc when it segfaulted. I was worried now, and I restarted the glibc compile. It was successful. I knew then that I had some kind of hardware problem.
To make a long story short(er), I had to clock the memory back down to stock in order to get a clean toolchain compile (turned out the CPU was fine at 2700). I believe that a toolchain compile is especially sensitive to memory errors, because the toolchain essentially must recompile itself at least twice, which would tend to magnify any problems with the generated binaries.
So memtest and prime don't mean a damn thing, really. Go figure.
