On Sun, Sep 11, 2005 at 12:57:59PM +0200, Paweł 'Róża' Różański wrote: > Well, you can compile kernel on another computer and make a deb > package (http://www.osnews.com/story.php?news_id=2949 may be helpful).
Well, yes, I figured as much -- but I don't believe it's the kernel. I've done a fair bit of testing now, and the problem is quite odd. It _only_ kicks in when the memory is configured (by putting it in the physical slots) in dual-channel mode, regardless of configuration otherwise. The place where I bought it seems as baffled as I am. :-) My motherboard (EPoX EP-9NPAJ) has four DDR-DIMM slots, where 1+2 and 3+4 form pairs -- the manual says that if I want the full 128-bit memory path with two chips (I have 2x1GB from Corsair, they're bought as a «matched pair» so they should really be identical), I must put them in 1+2, or 3+4. Both configurations are unstable, with classical memory corruption symptoms, as outlined in the previous e-mail. mprime reports excessive roundoff, gcc crashes, the kernel complains, etc. (after ~15 minutes of work or so). Yet memtest86+ can't find a thing, and it's usually good at finding this sort of thing. However, if I only use one memory controller -- regardless of which (I've been told the physical DIMM slots are hardwired to a specific memory controller on the chip) -- it all works. I can hit it with torture test for hours, compile stuff, etc. without any problems. Having a single chip (either one of them) in slot 4 works, using 2+4 works, using 1+3 works. IOW, I'm discounting the possibility of a broken DIMM slot, and also of defect RAM -- it doesn't add up. The machine is not overheating (it's in a cooled server room, keeping 18-20 degrees, and the CPU never hits 50 regardless of how much I load it), nor is it overclocked in any way. It has the latest BIOS update from EPoX (I needed that to get it to recognize the dual-core CPU). (Side note: Bonus points to EPoX for supplying bootable CD-ROM images of their BIOS upgrades, so I wouldn't have to install Windows or find a disk drive to do that sort of thing.) Does anybody have any ideas about what could go wrong here? I could of course just keep on running it this way (memory in 1+3), but should I upgrade it in the future, there's really no way out of using both channels, and I'm not really comfortable trusting my backups and whatnot to a known-bad system. :-) /* Steinar */ -- Homepage: http://www.sesse.net/ _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
