i3930K benchmark result with libgmp-5.0.4

Torbjorn Granlund tg at gmplib.org
Tue Mar 13 21:23:45 CET 2012

  Following is the new bench result with Turbo mode disabled.  But what
  is strange in this case is that the bench for the core i5 2500 (gmp
  5.0.1) reported on http://gmplib.org/gmpbench.html is a bit better.
  Are you sure the bench reported for the i5 2500 is also with turbo
  mode disabled ?  I hardly imagine the i5 2500 beating the i7 3930k.
One 2500 core can be expected to beat one 3930k core by about 33/32 or
about 3% on CPU bound tasks.  This is very close the the observed

GMP has good memory locality (except when mutiplying truly huge
operands) so can be expected to scale well with CPU clock.

The CPU vendors are trying to push each other off the shelf by providing
hundreds of models which are really not different under the hood.  But
almost all Intel cores are the same (sandybridge) while AMD is in
transition so have two common cores (stars family and bulldozer).

Both vendors also have low-power cores, but these have fewer models.


More information about the gmp-discuss mailing list