In article <7raral$e1h at net.bio.net>,
Hans Sluiman <h.sluiman at rbge.org.uk> wrote:
>Obviously, existing softeware needs to be rewritten to make use of the
>new Multiprocessing and AltiVec technologies that are built into the G4.
>However, the extent to which applications will benefit from the enhanced
>FPU speed of the G4 will depend on whether they use floating point
>instructions or integers. How do PAUP et al. work?
Maximum likelihood algorithms use floating point very heavily;
anything that speeds up FP will speed up DNAML and its cousins
roughly proportionally. Last time I profiled one of our likelihood
programs it spent over 90% of its time on FP calculations. I
have not looked at the PAUP* implementation but I would expect
this to be true for any ML method, since the calculations are
intrinsically FP.
I don't know if this is also true for parsimony implementations.
Mary Kuhner mkkuhner at genetics.washington.edu
