Proposals for SPEC 3.0

[Larry Yang]

Now that we're talking about SPEC, I'd like to bring up something that's
been bugging me about the way the data is presented.

Is it clear to anyone out there why the SPECmark data is presented the
way it is?  That is, a bunch of points are plotted on a SPECmark vs.
benchmark graph, then, for some reason that is beyond my understanding,
the points are connected.

For me, I see absolutely no reason to connect points that are completely
unrelated.  Now, if the x axis were laid out to go from less FP-intensive
on the left to more FP-intensive on the right, then there is sort of
a relationship between the points, so connecting the dots might be
appropriate. But how do you define something to be more or less
FP-intensive?

My problem is that connecting the dots seems to accentuate trends that
have very little meaning.  Has anyone else thought about any better
ways to present SPEC data?

-- Larry