Benchmarks of Quadro vs Softquadro and 2003 2004 (very long post)

Preface: I own an engineering business and we build and maintain all our
own systems. Performance is an issue that's close to me and the following
may be of some use to those contemplating different video hardware. All of
the tests were run using the typical software that runs in the background at
our office, like antivirus, motherboard monitor, trillian, etc. to fully
emulate our user environment. This means the benchmarks are as "real" as
they can be. Following the benchmark results are some observations from me,
someone who has used Solidworks faithfully day in and day out since the
beginning.

All benchmarks run from same machine: AMD Athlon XP 2400+, 1Gb ram, 40Gb HD,
WinXP Pro. All benchmarks used dataset "Standard General Test"

Procedure for swapping video hardware: Nvidia drivers uninstalled, cards
swapped, specific drivers installed, multiple reboots, opengl settings and
Solidworks functionality verified, then benchmarks run.

Resolution/color: 1280x1024x32bit
OpenGL vertical sync: always off
Use unified back/depth buffer: checked
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
SW2003 SP3.0
Geforce2-Ti 64mb (softquadro'd to "Quadro2Pro")
Nvidia driver: 23.11 Softquadro
Test Averages for 5 test(s).
Test Total = 216
Graphics = 34
CPU = 115
I/O = 66

SW2003 SP3.0
Geforce4-Ti4200 128mb (softquadro'd to "Quadro4-750XGL")
Nvidia driver: 41.03 Softquadro 4
Test Averages for 5 test(s).
Test Total = 209
Graphics = 30
CPU = 114
I/O = 64

SW2003 SP3.0
Quadro4-980XGL 128mb (true PNY Quadro4 board)
Nvidia driver: latest 45.23 (43.51 was marginally slower)
Test Averages for 5 test(s).
Test Total = 208
Graphics = 30
CPU = 114
I/O = 64

SW2004 SP0.0
Geforce2-Ti 64mb (softquadro'd to "Quadro2Pro")
Nvidia driver: 23.11 Softquadro
Test Averages for 5 test(s).
Test Total = 230
Graphics = 38
CPU = 115
I/O = 77

SW2004 SP0.0
Geforce4-Ti4200 128mb (softquadro'd to "Quadro4-750XGL")
Nvidia driver: 41.03 Softquadro 4
Test Averages for 5 test(s).
Test Total = 224
Graphics = 35
CPU = 114
I/O = 75

SW2004 SP0.0
Quadro4-980XGL 128mb (true PNY Quadro4 board)
Nvidia driver: latest 45.23 (43.51 was marginally slower)
Test Averages for 5 test(s).
Test Total = 212
Graphics = 34
CPU = 114
I/O = 64
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

I did much research into Nvidia Geforce boards using softquadro from
"Unwinder" - see www.nvworld.ru and www.guru3d.com. I also wanted to
compare my softquadro boards against a true top-of-the-line Quadro board
like the 980XGL. This information is readily available in numerous places
but this is my lone attempt to shove it into one place, using my own test
results on my own hardware.

I used the Solidworks benchmark downloaded from spec's website
http://www.spec.org/gpc/apc.static/sw2003.html and ran it using Solidworks
2003 SP3.0. Then I uninstalled 2003 and installed 2004 SP0.0, mass
converted all the benchmark files and ran the benchmarks again. This seems
a reasonable way to compare 2003 and 2004 because this is exactly how you
would approach the upgrade and conversion in a working environment. Somewhat
surprising results but not really; 2004 is marginally slower than 2003 in
both graphics and IO. By my calculations using a Softquadro2Pro, it's about
12% slower in graphics and about 7% slower overall - factorally less with
faster hardware. Not as large a difference as there was between 2001+ and
2003 but still noticeable.

Before any Solidworks zealots chime in let me say that I used to be and
could again become a Solidworks zealot but allow me to vent for a second...
new features and glossy brochure "improvements" do nothing for my business's
billable time. If I needed Cosmos I would buy it and if I did need it, the
included version would do nothing for me. It has already done nothing for
me even though I would have liked it to. If normal day-to-day functions
take more time then there is a problem, as there was/is with 2003 and now
2004. It took me a few weeks to realize it with 2003 but when projects
began to take more time, normal rebuilds became 3 min ordeals, Solidworks
shutting down by itself, etc. it was easy to see that something evil was
happening. There are enough postings to prove my opinion so let's not
rehash. I can prove it to any app-engineer that visits our office and I can
equally prove it to them with a visit to their office. I'd like to know how
I can justify to a client that their project is going to have to have
additional days quoted due to software "improvements". I can't; therefore,
Solidworks has cheated me out of billable time by making their software take
longer than it used to and I can't bill for the difference. The last time
this happened was about 1998 or so with a company called Autodesk. Vent
off.

My findings can be compared in terms of what this or that board can do by
visiting a nice benchmark chart at (cad.de?)
http://solidworks.cad.de/int_swxbench11.htm.

My two year old GF2-Ti board (with Softquadro) was an admittedly fast board
for its day but I never expected it to perform at nearly the same level as
the new Quadro4 boards. This really blew me away but I suspect what's
really happening here is that not only is Solidworks becoming incrementally
slower with each release (proven) but nearly completely fails to take
advantage of newer hardware capability. This makes perfect sense when you
consider how much slower the video got between 2001+ and 2003 and now again
with 2004. Also, consider the last really fast version of Solidworks was
2001+ and you realize that with respect to OpenGL performance Solidworks has
frozen themselves in time at about year 2001. You could argue that Realview
negates this logic but Realview is not a measure of speed within the
modeler. It is merely a specific set of calls to new hardware. Hardly an
optimization, which we desparately need.

The GF4-Ti4200 board, when using "Softquadro 4" drivers appears to offer the
same speed **for Solidworks** as the true Quadro4-980XGL. The only caveat is
that to enable Realview you must also use Rivatuner's NVstrap utility to
convince your system at boot-time that the board is a "Quadro4". Without
NVstrap you achieve the same speeds but no Realview. The tradeoff to using
NVStrap is that your system is unable to gracefully return from a power
management event like Standby-Resume or Hibernate-Restore. If you never use
these power management functions then no problem, you'll achieve full speed
and Realview capability of the Quadro4-980XGL for about a fourth the price.
If those power management functions are important to you then you have two
other options; live without Realview and just use "Softquadro 4" drivers or
pay for the real Quadro4 board. Since I have yet to see any "real" benefit
to Realview, other than it's "cool" factor, the choice is acceptable to me,
either way.

- Eddy Hicks
www.solidlogicdesign.com

 

<clipped benchmark blah, blah, blah>

What's all the fuss about benchmark times? Yes, you've "proven" that
SolidWorks 2004 is slower using a sterile, controlled, single point of
reference. But benchmarks don't tell the whole story. New features and
commands are there to increase productivity, and they do. Even the "glossy
brochure improvments" can be of value. I'll use the much-bashed "custom
skins" as an example. Try creating a skin using the logo of your favorite
football team (Hook'em Horns), or put a picture of your wife/kids on there
for you to look at while you work. Doesn't do anything to increase the speed
at which SolidWorks rebuilds, but it can make the work experience that much
better (happy worker = more productive worker). We used to make drawings
three views at a time automatically - now with "pre-defined views" we can
make them as-many-views-as-we-want-at-a-time. Depending on your
requirements, that can certainly wash out the 7% (proven) slowdown.

Multi-bodies, eDrawings, handles, folders, rollback improvements, 3D content
central, 3D Instant website, contour selection, mirror parts in assemblies,
I could go on and on. These are all features and improvements added to
SolidWorks since the 2001+ - the fastest SolidWorks ever (proven?). While
some of these features still could use improvements, they are all capable of
increasing your productivity as a designer, drafter, or engineer.

Until I landed a my present job, I was using a Dell 866Mhz desktop with 1GB
RAM. Now I have the pleasure of running on a 2.6G machine and yep,
SolidWorks runs way faster. Should I complain about the software now? How
much lost productivity did I have at the last job? Who's fault would it be?

I know that some people have many issues with SolidWorks. I also know that
one man's crap is another man's gold. You point to "enough posts to prove
your point", but in the world outside of this newsgroup I know for a fact
that there are many of people that would disagree with some of the
assessments here. Even some of the recent "2004 sucks" posts have had
responses that were quite the opposite.

I responded to this post knowing full well that I will get hammered. Go
ahead - but for me, SolidWorks is the finest tool I've used in 23 years of
design. It's fun to use, productive beyond anything else I've used, and I
welcome each and every new release with excitement.

Richard

<more clipped chatter>

 

I am merely posting benchmarks primarily for the purpose of comparing Nvidia
graphics cards and Softquadro performance. The 2003 vs 2004 comparison came
as an afterthought. I thought maybe 2004 was going to utilize the
Quadro4-980XGL better than 2003 did, in fact I was counting on it, but I was
wrong. As for user experience benefits, etc... would you rather get your
work done faster and get home to your kids sooner or take longer and look at
them withing the Solidworks window. Frankly, that's rhetoric we can all
live without. You draw your own conclusions but when you add 2.8hrs to your
next 40hr job think about who's paying for it.

- Eddy

 

You missed my point. I am saying that benchmarks aside, and all things
considered, SolidWorks and getting my work done with SolidWorks is faster
than it was before. That's not rhetoric or marketing speak, it's what I
believe - and I think the facts bear me out.

The reference to the "skins" was to make another point. Just because
SolidWorks chooses to spend some time on something that is purely cosmetic
doesn't make it bad. If something makes my work experience better through a
little personal customization I'm all for it. Merely having this feature
available costs me nothing.

And before anyone shouts out - but if they weren't working on "skins" they
could be working on <fill in the blank>, let me say that I could retort that
if they weren't working on
features and functionality that I don't use - they could focus more on <fill
in the blank>. It's give and take, and I trust SolidWorks to provide the
tool that gets my work done.

And they have.

Richard

 

Mike, I've had the same thought because you never know. That's one of the
reasons I decided to test a real Quadro4 board, to keep me and the results
honest. I figured if it fixed the crashes the price tag would be worth the
piece of mind. I was hoping and half expected that 2004 was going to scream
with the Quadro4-980 and not crash. Problem is... it's not the hardware or
softquadro. Using the real Quadro4 both SW2003 and 2004 have managed to
bomb when used outside of the benchmarks, and the Quadro did anything but
scream with 2004. I'm serious man, someone better get it together at SW or
it's gonna be Autodesk all over again. I've been around this stuff since
the beginning, starting with Autocad in 1986. People are deathly loyal.
That's good to a point. I used to be loyal to Autocad until 3d became a
requirement. Then I was an SDRC Ideas fan until my fellow engineers at the
time couldn't pick it up easily. I remember the Autocad exodus in 1999 when
Solidworks showed everyone a better way, and a better company. There's also
an analogy to Pro/E and their 3d dominance back in 1999, but this isn't
about price. This is about corporate arrogance. SW shouldn't be resting on
their laurels. Users simply should not accept new features in the face of
deficiencies. Loyalty only extends as far as one of two points; either
until someone feels screwed, or until something else looks better. Either
one and the dollars start flowing in a new direction. The season is ripe
for a new contender. Not sure who or what but it could happen. Too bad
though. If SW were to fix the real software issues, make the bloat optional
at install, and start offering real performance gains instead of marketing
lightweight assys I would gladly start waving their flag high again. Until
then, buyer beware?

- Eddy