Benchmarking 2006

Here are the results of the first round of benchmarks of 2006. They were run
on an AMD XP3000+ system with 1GB ram. The STAR2.1 benchmark was run on
2003,4,5 and 6 setting type and constants libraries according to SW release
per instructions. Three runs were done on each release and the results
averaged.

2003 avg. 43.42 s / 5.07 s
2004 avg. 48.66 s / 4.97 s
2005 avg. 57.40 s / 5.19 s
2006 avg. 62.21 s / 5.56 s

Since the benchmarks were run without reboot between either tests or
releases a reboot was done to try and get the 2006 times down. After rebot:

2006 avg. 63.11 s / 5.55 s

In addition one of the STAR parts from 2003 was loaded into 2006. It
consistently got faster rebuild times than the part created in 2006. I'm
going to have to mull over this one.

This benchmark can be obtained at http://www.engtran.com/SWBM001.html.

Next in line is Patbench.

 

Thanks for posting these results.

SolidWorks, you continue to insist that 2006 is "faster" than 2005. This
point was brought up at the 2006 rollout. What are you views about these
results?

Best Regards,
Devon T. Sowell
www.3-ddesignsolutions.com

 

STAR shows one piece of the puzzle which is made up of many parts. The
piece STAR shows is regen times without the influence of graphics or
GUI. I think Patbench will be a little kinder and it will be with
great interest that I run it because of the new patterning
functionality in 2006. With that functionality people are going to be
using large patterns a lot more IMHO. After Patbench I'll run Ship in a
Bottle. Those two shouldn't take much time. But after that I'll try to
run the SPECapc benchmark, but that is release specific. I've never had
good luck getting it to run either. When I've exhausted benchmarks
I'll start with VLPs and VLAs* using the Rebuild macro or TSToolbox to
time it. I couldn't get TSToolbox to load on my XP machine though.

* VLP = Very Large Part
VLA = Very Large Assembly

 

OK, good point.

To your knowledge, has any version of SolidWorks performed better(faster)
than its previous version in any benchmark?

Best Regards,
Devon T. Sowell
www.3-ddesignsolutions.com

 

I think you will see that a little in Patbench. In general though, no.
My rule of thumb has been about 4% a year loss since '99. That would
make about a 21% loss in rebuild times on complex parts since '99. And
sometimes you see it in service packs. If you search the archives way
back I think 2000 had issues like that between service packs. But 2005
was the biggest single jump I ever saw.

Another thing to note about the STAR benchmark is that it uses a real
time timer. So if there are a lot of services running you will get
slower times or a lot of variation. That isn't necessarily a bad thing
as it can point to a problem in the system as opposed to SW. And that
is the kind of problem a user can fix. For example we ran the STAR
benchmark on this group a while back an people with seemingly identical
hardware got dissimilar results. That can point to a problem that a
user can fix.

The big picture on this kind of testing is that there are many factors
that go into modeling time. SW performance is just one. Bad modeling
practices, a system with a lot of junk running on it, a hokey network,
or even a virus or spyware app can all lead to long modeling times. The
benchmark can help point the way if there is data available for your
system and what it should be.

 

Patbench Results

Ran four times, threw out the first run as it was always longer.

2003 6.92 s for 8
2004 7.81 s for 8
2005 6.82 s for 8
2006 6.80 s for 8

In this case 2006 is a very slight improvement over 2005 and a bigger
improvement (2%) over 2003. 2004 was a dog. I did not log memory usage
in these tests. That is also an important consideration as patterning
uses a lot of memory.

Ship in a Bottle Results

2003 41.625/40.265/30.78/30.328
2004 42.45/40.88/31.54/30.91
2005 41.484/40.75/31.266/30.88
2006 41.55/41.09/31.55/31.39

The results are given in the order specified in SWBM001. Looking at the
fourth result, the one with the least graphics processing there is a
noticable but small difference between 2003 and 2006 with 2006 being
about 3.5% slower.

Overall 2006 is slower than 2003 but not by much except in STAR. 2006
does seem to handle patterning better.