> Currently, apps like mplayer and OpenSSL run much
faster on x86-32 > because such core routines are in hand-coded assembler. > For most other things, though, GCC 3.4 uses the extra 8
registers > to good effect. I think that is where we will see most of our improvements
is in doubling the registers, since so much of what gets done is recalculating
recent values. By not going back to main memory or even L2 cache, I think we
can get a significant performance increase (ie. An hour less run time for a 24
hour period). I guess I would have to run tests to see. Another problem is that
these programs can be memory hogs and the nodes only have 512 MB of RAM which
is another concern. Robert LeBlanc |