Benchmarks

Hi to all
where i can dowload a good benchmark to test sw 2006 on pc dual core and
xeon?
i hear that there is a benchmark called ship in a bootle.
thx a lot to all

Max

 

a good question...
'ship in a bottle' is a small test assessing general rebuild speed.
It really only gives a basic comparison between processors FP performance.
I think you will find a few links to it and various other simple tests
devised by groupies at the SW forum under the 'performance' section.
SIAB doesn't address working on complex parts, loading an assy, making
multi-page dwgs and the like that might be done during a days work where a
dual core might demonstrate real worth.
As we don't exactly know what parts of SW have multi-threaded potential
other than dwgs and a few scattered features smart old matt discovered it is
hard to gauge what the relative worth of dual core might be in actual work
using existing tests.
I don't think SPEC really addresses this either since use of the tech is
only emerging now with mainstream multi core hardware developments.
SW are very quiet about any extension to multi threaded capability for some
weird reason although they have stated in the past CAD is not suited to it.
Possibly as SW groupies we should devise our own dual core test.
Anyone feel like a challenge?

 

You use TSToolbox for assessing real world.

Dual core testing is not that difficult. Use perfmon and your favority
big assembly, part or drawing. Log both CPUs, download the data to an
excel spreadsheet and do your analysis.

 

Neil,

it.

Not true as a blanket statement. GM uses UG (under Unix) in a massively
multi-processor environment. I've done some work with GM engine models, and
there's no way "they" (GM) could get anything done without it. I've worked
with both simplified ( the complete engine minus some internal detail) and
the whole shebang. They use the simplified ones to actually drop into the
car model to check for interferances, fits, hose lenghts, drivetrain
alignments, etc,. The complete engines are just that, complete. I mean every
nut. bolt, screw, bearing, roller, pin, spring, belt,, everything. These
assemblys are very cumbersome even on a very high end dual processor PC.

UG uses Parasoild so it "is" being done as we speak. The big question, to me
at least, is, is this capability unique to the version of parasolid used in
UG ? Or is it in the OEM kernal, and is limited by implementation, or the
O.S. ??

Mark

 

SW is more multi-threaded than anyone is saying. Parts, which should be
the least multi-threaded of the 3 SW document types, will sometimes peg
both processors at 100%, and very often run around 70-80. It seems to be
true especially for different types of features.

Large assemblies and drawings must benefit a lot from dual proc /
multicore, although I haven't looked at that as much as parts.

Parasolid isn't the only consideration. There's also the solution of
sketch relations and assembly mates, which should be very parallel
process capable. Plus, what happens with multi solid or surface bodies?
Separate bodies should be capable of being processed separately.

 

Matt,

My dual Opteron systems do the same thing.

It's almost like the overhead involved in splitting the tasks, cancels out
any performance benefits.
This is definately an O.S. thing. Windows is way behind even the Unix
systems of 15 years ago in this respect

 

all,
Possibly it is the SW architecture itself then that does not lend itself to
multiprocessing rather than parasolid lacking the capability and hence their
quiet about it - a bit like splines and C2(?)....
I would have thought this would be good marketing fanfare as dual core
arrives so I guess there are issues they are not keen to publicise.
Could be that only some of the processing can be ever be done simultaneously
as it is implemented and they have about done all they can do?
Surely which ever CAD company manages to make the most of multicore will
have a significant sales pitch because of the significant time and hence
cost saving ...
Probably XP was not really built with multi processors in mind for the
masses but I think I read somewhere Vista has this as a underlying
consideration. So we can be hopeful of improvements for our needs.
It would be great if someone with dual core would do some experiments for us
along the lines TOP suggested and post some results.
Of course it would be even nicer if SW would talk about it as I keep asking
them to do... but then I am not a favourite customer... ;o)
I get the feeling 4 cores may be as many as will be utilised for mid range
workstns and then only taxed if you want to do other tasks at the same time.
Perhaps it is more sensible just to have another pc - your previously
retired workstn? running for those tasks and then 2 cores are enough?...
Will we be able to buy just a 2 core in a few years or will 32 be on
everyone's desktop doing mostly wait cycles?
Interesting to see how things pan out.
Hopefully those long frustrating waits for CAD jockeys will be a thing of
the past )