Re: performance of 21164-300 vs P6-200
Arno PAHLER (paehler@atlas.rc.m-kagaku.co.jp)
Wed, 12 Jun 1996 00:15:49 +0900
I just posted the code - as far as I can tell there should not be
any float-to-double conversions, in the timing loop I basically
only use (sorry for the Intel terminology) fadd and fmul - but on
Jim's suggestion I tried doubles on an IBM RS6K and the results
are a lot better - what surprises me right now is that tests 1
and 2 are faring so poorly then on a P6 - my suspicion is that
going float to double (note the sizes) makes them fall out of
cache - and what PC memory is like we all know.
Arno