GMP compared to JAVA

Torbjorn Granlund tg at
Sat Dec 22 19:45:38 CET 2007

Nustenil Segundo wrote:

  For the result in C with n = 10005 bits the time was 2800 ms (a example).
  For the result in JAVA with n = 10005 bits the time was 10 ms (a example).

Paul Zimmermann replied:

  (a) JAVA is much more efficient than GMP
  (b) your computer is very efficient for JAVA applications
  (c) your JAVA exponent (q) has much less than 10005 bits

Nustenil Segundo replied:

  (a) and (b) can be true, but (c) is false.

OK folks, thanks to the information from Nustenil Segundo, we can now
unequivocally determine that java is 280 times faster on modexp then
one of the previously fastest known modexp implementation.  "JAVA" is
nothing short of a major breakthrough in implementation and/or
algorithm research.  Kudos to its authors!

Improving modexp from state-of-the-art implementations by just a few
percent is hard work.  Making it a factor of two would by many be
considered close to impossible.  And now we're facing an improvement
of 280x!

I humbly recommend java for all future bignum work.



More information about the gmp-discuss mailing list