Does SW2008 take advantages of multicore processor? Any recommandation or
I'm not satisfied with the SW 2007 performance, because use only 2 of 4
processor in quad configuration. In that sense is the reason to build
supercomputer like microwulf?
Thx in advance
CAD is very difficult to parallelize. Think about the parent/children
relations in your feature tree, and you'll see that 2 or 4 cores are
of little help to rebuild feature quicker.
Some specific features which are fairly long to rebuild might be
partly parallelized. I'm thinking of shells, which could be possibly
divided into 2-4 parts each calculated on a different core, then merge
the results (which takes some extra time compared to a mono-thread
calculation). I'm fairly sure Parasolid works on such things, because
it's their business rather than SolidWorks. The day Parasolid (or
ACIS) will be multithread, all the CADs will take advantage of it.
AFAIK, at this point SolidWorks uses 2 cores only in drawings, where
the high-quality for printing is computedin the backgroud while you
work on a low-quality drawing on screen. Photoworks and Cosmos are
multithreaded, since parallelism is fairly straightforward for those
Actually the parasolid kernel is multi-threaded. For a long time this
was visible when doing detailed performance assessments of single
versus dual processors.
When horizontal modeling is used the history dependence is to a large
part removed. The handling of the fancy user interface graphics
certainly does not have to share the same processor that. One of the
things that is becoming more apparent with SWX is that seemingly
slight changes in modeling technique can cause HUGE changes in
solution/rebuild times. This was proven with the challenge I put out
some time ago to build a truncated icosahedron. The final geometry was
the same in each case and the method that solved the fastest was the
most complex. But strangely the model that was second fastest wasn't
all that different than some of the slowest models in that it had very
few features. The upshot is that there was no apparent way to see the
effects of decisions in how the modeling was done. For the competitive
group that participated some I know tried to reduce rebuild times.
But I really don't think that:
a. it is that difficult to parallelize many time consuming processes
especially in drawings where the model is already solved and the big
crunch is projecting lines.
b. it is that difficult to parallelize some of the new features we
have begun seeing. For example suppose parallel processing was used to
simultaneously solve for the fastest way. Or to solve several
components in an assembly. Or several views in a drawing?