Multi-Core CPUs & SolidWorks?

Also Seth Renigar.

I would also like to see a benchamrk that segregates performance for the
various "vertical industries" or whatever they're called now.

Unfortunatley, simply converting files doesn't really exercise SW. That
would be more of an IO test. (What's the point of an IO test when most of
the time spent saving has nothing to do with file transfer?) We would
probably need to run ForceRebuild3(false), which is more thorough than
ctrl-Q.

Rant: Even that would only exercise the CPU. We still need some meaningful
test of the video card. Unfortunately, by the time you have a model that
can tax current video cards, the SW UI is so bogged down that you can't
work any faster. Just today I added a feature that shows a rebuild time of
zero, even though it took almost 5 seconds for the UI to catch up.

 

Again, I use TSToolbox to time in-house models.

There are:

Ship in a Bottle -- fairly complex part that taxes both CPU and
graphics. By setting the graphics at full fast and full slow you can
get a feel for the contribution each makes.

STAR2.1 -- definitely checks out the CPU only. Interestingly Intel and
AMD typically will score very different on the modeling portion, but
still be quite close on the rebuild times.

PatBench -- large pattern check and on older SW releases checks the
maximum amount of memory SW will use.

SpecAPC -- good for tuning a machine and stress testing. As a
benchmark it sucks because they change it so often.

Perhaps I could turn those truncated icosadodecahedron models into
some sort of benchmark if the authors agreed to it.

IO is the hardest thing to check because the OS gets involved and with
caching it is hard to tell whether a file is coming from disk, network
or ram cache.

The other thing that is hard to evaluate, and you can see it in STAR
is that after SW is done building the model there is a big wait while
it writes to disk. This is not included in the current timing
algorithms. But it could be.

TOP

TOP

 

Again, I use TSToolbox to time in-house models.

Oh yeah? Well my watch has a chronograph on it!

I actually inadvertantly turned one of those into a stress test. Someone on
the SW forum asked how to do an extarnal thread as a library feature. Well,
I thought that sounded like fun, so I did it. I needed a test model with
faces oriented to all 8 quadrants (octants?) to make sure something
wouldn't flip, so I used the truncated icosahedron. Of course, the helical
cut took a little time to rebuild. What surprised me was that every
successive thread I inserted required exponentially longer to rebuild, so
that the 8th one took quite a while to rebuild. That was SW06, I think.

 

Now there is an idea. Take all the pathological models and turn them
into a benchmark.

TOP

 

Hmm. This has me thinking about the fact that reported rebuild time is
often far less than the time we have to wait. I think they call this
"friction" in the interface. I would be interested to see if this is
linearly proportional to rebuild time, if if veries between systems, and
how much it varies between versions. I did some subjective testing between
05, 07 and 08 b1 yesterday, and there seems to have been some serious
progress made since 05. A matter of 8-10 seconds vs 3-4 seconds to add a
feature that shows a rebuild time of zero.

 

Hmm. This has me thinking about the fact that reported rebuild time is
often far less than the time we have to wait. I think they call this
"friction" in the interface. I would be interested to see if this is
linearly proportional to rebuild time, if if veries between systems, and
how much it varies between versions. I did some subjective testing between
05, 07 and 08 b1 yesterday, and there seems to have been some serious
progress made since 05. A matter of 8-10 seconds vs 3-4 seconds to add a
feature that shows a rebuild time of zero.

 

Yup, a proprietary graphics standard will allow a monopoly to dictate
what platform CAD runs on. Right now there is a choice.

Several OS support serious CAD.
Unix - CATIA, UG, Pro/E, and perhaps some lesser known
Windows XP - The above plus mid range modelers and AutoCAD
Windows NT and 2k - Ditto XP but only older versions.
MaxOS - SolidWorks, and others (don't know if they use OGL or
something else)
Linux - Pro/E and others.

I invest in high end graphics cards with a much longer time horizon
since at least as far as SW is concerned it only takes a CPU and
motherboard upgrade to keep up at minimum cost.

If it ain't broke, why fix it? What in the world is wrong with OpenGL
as far as CAD is concerned? I have yet to see a post on the NG
complaining about graphics issues other than driver problems.

If it is MicroSoft throwing people and money on the problem don't bank
on it. They haven't gotten the OS right yet and it has been 14 or 15
years since New Technology (NT) was going to save the world from Unix
and MAC.

SW got it right when there were very few people working on the
problem. The bigger they grow the more problems there are. Getting it
right doesn't always scale with the number of people working on it.

TOP

If it ain't broke don't fix it. SW is moving away from letting the
graphics card do all the work anyway.

 

Top of the morning to you all, and I have for all...ONE word...for
user benefits...

<- - -CHOICE- - ->

It is the only thing that allows users to keep the damn suppliers on
their toes. MS does NOT promote choice on their OS. It is the
Redmond highway or noway they want promoted.

Microsoft became complacent in the late 90s once they became a super-
majority supplier and could virtually dictate terms to everyone. Now
they are trying to muscle in on audio, video, graphics, replacing
Java, .Net. I think MS is spread to thin, trying to "be all, end
all".

Frankly, I can't see the benefit of Microsoft's OS in today's web
world. I can see the value in various programs I use on the OS, like
SolidWorks, but the OS is a bitch if you take it online, and sort of
OK if you keep it off the net. That is not an OS I want to write home
about.

If the CAD guys start seeing better ways to run their products on
other OSs, I think we will see more viable options appear. Mac &
Linux use in some organizations and colleges is on a major up spike,
and it is not hard to see why, just in reduced maintenance time.

IT guys are seeing the Mac as a boon. Run Mac OS, BSD, Solaris,
Windows, DOS, Linux, and all from one box with several running at the
same time if needed.

Where is the "run anywhere" mantra when users want it?

Bo