[RC5] Checking a bad proc..
blitz at macronet.net
Wed Jan 8 14:26:59 EST 2003
Agreed, familiar with that "bent-ium" prob of days past...thats why I
suggested benchmarking them on a different motherboard. That should rule
out a MB problem, and conflicts that might be arising from the dual CPU
original board. It (the CPU) might be bad, but I'd strive to proof it on a
known good system first. The overheating of the original board might of
caused the failures. Cant hurt to proceed in a logical fashion to prove or
disprove CPU failure.
At 19:30 1/8/03 +0100, you wrote:
>At 05:23 08.01.2003 -0500, you wrote:
>>CPU's are transistor devices, and after 40 some years of dealing with
>>them, they either work as advertised, or theyre dead. There is no middle
>>ground. It sounds like there are other problems, as I have never
>>seen/heard of a partial failure of a CPU...
>Modern CPUs are very complicated devices, and using them does not always
>use every little part of it. In fact floating point operations very often
>use some kind of tabling for low-level operations. For common operations
>they do not use everything of those circuits. (As most might remember
>there was an issue with the early Pentium CPUs when in very rare
>circumstances the FPU returned false values). I have seen some CPUs with
>damaged FPU and even some vector processors that did not work properly.
>All those were working fine in usual OS environments, but failed when
>doing some special operations. Those special circuits are very hard to
>check on Hardware base sometimes (like for tabling you must either have a
>calculatable table or a second one). But at least the x86 processors have
>very good check circuits for the integer and memory operations, that
>prevent the CPU from booting in case of failure. So I personally think
>that if a CPU runs, then it will run on RC5 as well, since it does only
>use integer arithmetics afaik.
>PS: According to the AMD spcifications their processors have their
>hot-spots at the FPU and Multimedia instructions.
>To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
>rc5-digest subscribers replace rc5 with rc5-digest
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5