Timothy Murphy wrote: >On Monday 20 November 2006 03:08, Jeff Vian wrote: > > > >>This description tells me that something is definitely not right with >>memory after the crash. Have you checked with memtest86 to see if that >>reports anything? >> >> > >Thanks for the suggestion - >I assumed the memory was OK since I have never had problems >with Windows XP or i386 Fedora. >But I've downloaded the memtest86+ rpm, >and will run this tonight. > > > >>You also might try reducing the speed of the memory and see if that has >>any affect. >> >> > >How do I reduce the memory speed? >Do you mean reduce the CPU speed with something like cpufreq? > > No. Run with additional wait-states. Like instead of 2/1, run with 3/2 or 4/2... >>What about cooling on the processor? Are you using the heat sink that >>was provided with the CPU or better? If so then it is not likely that >>it is heat related, although with inadequate cooling it is possible. >> >> > >Yes, I haven't touched the CPU or heat sink. >Unfortunately I'm not sure how to measure the CPU temperature. >I'm compiled a kernel with ACPI=>Processor=>Thermal Zone enabled, >but it doesn't seem to show anything in /proc/acpi on this machine. > >Again, many thanks for your suggestions. > >I must say the machine seems to be running faster under i386 >than it did under x86_64, >though that may just be an impression because I am not waiting >for it to crash! > > Unfortunately, that a processor didn't previously overheat with a different OS is less meaningful than you might think. Some distros, operating systems, and even compiler settings do a better job of exploiting the processor, meaning the code is compiled to make better use of instruction scheduling, or else might use operations that require dedicated logic on the CPU. (Better use often means more gate state changes on the die, which means more heat generated...) I've seen issues where code that was compiled to run with the i486 or i586 settings in gcc did fine, but when run as i686 would cause significant heating (this was on a Cyrix/Via CPU) ... even stranger things would happen if it was compiled as a pentium3 ... Even though the processor was supposedly p3 compliant. Do you have "lm_sensors" installed and configured? Try that. -Philip
