Hi,
An aside: Please try to rename the subject like I did
above when the thread has strayed way off the original topic). Thanks.
At 06:34 AM 6/8/99 -0500, Mikus Grinbergs wrote:
>I have seen opinions on this mailing list that the
>existing 16-hour self-test might not be enough to bring hardware
>deficiencies to light. Someone suggested running the torture
>test - but is that more "rigorous" than the self-test?
It is more rigorous in that it can run for as long as you like,
whereas the full self-test takes "only" 16 hours if I recall correctly.
>And why
>should running the torture test for a week be any more definitive
>than already running Lucas-Lehmer on an assigned exponent?
The torture test is identical to a Lucas-Lehmer test except that
the end result is known and can be tested for correctness.
When you are done with a Lucas-Lehmer test your result will not be
proven right or wrong until a double-check or triple-check is done.
>My suggestion is that the QA people should "update" the built-in
>self-test to make it more likely to provoke hardware deficiencies.
>(Also, now that we are in the realm of "large" FFTs, is it still
>useful for the self-test to spend so much time on "short" FFTs ?)
I don't know how to make the self-test any more strenuous. I suspect
short FFTs are just as hard on the CPU and cache as the large FFTs.
The large FFTs will access a little more memory.
Best regards,
George
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
