> It remains to be seen whether or not [identical results] will sell machines

[David G. Hough]

Once three different vendors of 32-bit IEEE architectures
provide identical numerical results on the same programs
(maybe three from among SPARC, SPIM, RIOS, HPPA)
then all 32-bit IEEE vendors will have to do the same, even
if most customer programs don't benefit much from it, and
even if many customer programs are actually compiled or
executed in a faster mode which is slightly different.
Customers will demand the capability.

I think this will work the same way that GCC is raising the
standards of C compilers.  There is no point bringing another
student project quality C compiler to market unless it's better
than GCC (or works on an architecture like IBM PC's that
GCC never will).  The minimum entry level for the workstation
and mainframe market will be GCC quality, and GCC is still
only in beta.  I'm still hoping that somebody is working on
GF77 or GF90.

Similarly once there is a way to get identical, unimpeachable
results on dissimilar 32-bit IEEE architectures, that will 
raise the minimum level of competence to that level.

That raises the issue of what the real impact of NCEG will be.
A written report will be useful documentation, I hope, for
a derivative of a GCC compiler and run-time library that 
implements everything in the NCEG report.  That implementation
will be proof (or disproof) of the consistency and feasibility
of the requirements in the NCEG report.  I've already got enough
to do for the run-time library to keep me busy for a while.

> As the computing world goes more an more parallel
> the idea that a sum reduction of an array of values will always yield
> the same result on every machine seems unattainable.

Kulisch would disagree.  Anybody who thinks that IEEE requirements
are a nuisance should investigate Kulisch's work, which has attracted
considerable interest (and implementations) from German manufacturers,
including IBM.  All he really needs to get going is a GF77 with
a correctly-rounded scalar product operator.