Xeon as a worthwhile platform?

[Torbjorn Granlund]

Mirar <mirar@mirar.org> writes:

  > Before Opteron gets assembly support in GMP, it isn't going to be
  > any better than Athlon for running GMP applications.  It might in
  > fact be slower clock-for-clock.
  
  No, with the double size (1Mb) of the L2 cache, it's faster then an
  Athlon XP for the same clock speed for the same instructions for any
  test I've seen. Not by much, but anyway. :-)
  
The Hammer core is superior to Athlon, and should be faster for
almost any test case you can come up with.  But with the GMP
assembly loops, we have a case where something has been
meticulously tailored for a specific pipeline.

For Hammer in 32-bit mode, the configurations athlon-pc-xxx and
pentium4-pc-xxx will both work.  It is quite possible that
pentium4-pc-xxx will be better.  Also, GMP 4.1.2 has rudimentary
Hammer support by means of inline assembly macros in longlong.h.
A generic configuration without any assembly loops might
therefore be best.

-- 
Torbjörn