Re: Speed freak needs help

[michaelg@Xenon.Stanford.EDU (Michael Greenwald)]

moon (David A. Moon) writes:

>							Symbolics still makes 
>Lisp machines, but again without having measured them I think you would find 
>performance on the latest Symbolics machine similar to or less than the 
>Quadra.  (The Symbolics clock rate is half the Quadra clock rate, but the 
>architecture is not much more than twice as efficient.)

I don't know the numbers, this is a non-rhetorical question: How does
the Symbolics single float implementation stack up against the Quadra?
(I'm assuming that the Quadra conses its floats, while the LispM uses
immediates).  If (and it's a big "if"), the application in question is
float-intensive, and the GC is taking a significant part of the
overhead, then perhaps the Symbolics machines (or any machine with
immediate single floats) can give a reasonable performance boost.

But for a factor of 10 or 20, the answer (if any) probably lies in:

>Are you sure the algorithm can't be optimized?  Sometimes that's the fastest 
>and cheapest way to get more speed.  But maybe you already tried that.