Tuesday, April 9, 2013

Free SolidWorks from Performance Constraints (FSWPC-13-#7a)

(Hardware Testing - Graphics Cards-SolidWorks Functionality)


If you have followed our efforts over the years to isolate performance factors you know that we have always had difficulty in breaking down the specifics of video cards as well as video related aspects of SolidWorks.


This year we tested an interesting array of video cards. The NVIDIA Quadro 600 that came with our BOXX 4920 extreme. The NVIDIA Quadro 2000, 4000, and K5000 were given to use by NVIDIA. Finally, we tested an NVIDIA GeForce GTX 650Ti that Adrian picked up at Best Buy.


Adrian and I tested graphics cards differently than our other hardware so that we could isolate specific graphic intense functions. We also tested against three different model types; a large assembly, small assembly and a single part file with a reasonable number of features.


The following graphics intensive functions were each tested individually across all 5 NVIDIA cards:



  • 5 different display styles

  • Shadows

  • RealView

  • Transparency

  • Lightweight and Large Assembly Mode

  • 5 different level of detail settings

  • 3 different Anti-Aliasing settings

  • 5 Different Image Quality Settings

  • A comparison of all of the above items set to the best possible setting vs. everything being set to the worst possible settings


This testing yielded some pretty interesting results. In this article we are going to focus on the use of different SolidWorks functionality and the performance ramifications that occurred across our 3 model types regardless of which graphics card is employed.


When analyzing display styles we found, as we expected, that the best video performance is obtained when working in Shaded.




What was interesting is how quickly the performance degraded and we moved from style to style; with Hidden Lines Visible and Hidden Lines Removed styles being, on average, 110% -- 140% slower than Shaded. I was also surprised that a simple Wireframe is roughly twice as slow as Shaded.


We found the effects of Shadows very interesting as well. While we thought the extra calculations required determining where the shadows fall in relation to the light source and other surfaces of the model would cause a drag in performance what actually occurred was virtually no performance degradation at all. (This may not be true if RealView was being used but we did not test that combination of functions.)




As expected RealView Graphics caused a significant drag on performance slowing the benchmark down by roughly 33%.


Transparency is a drain on performance as well but more significantly than we would have estimated. We turned on several transparent parts and faces in each of our tested models and found that on average Transparency caused a 40% performance loss for graphics related functions.




As always though; engaging Lightweight and Large Assembly Mode improved performance markedly. An approximate 70% improvement was realized when using these powerful tools.


In the cases of Anti-Aliasing and Level of detail it was surprising that neither caused any measurable difference in performance. We have determined that this is due to the fact that these settings are for improving image quality in the case of Anti-Aliasing and prioritizing image quality and performance while spinning, panning, and zooming. The reason we found that this occurs lies with the benchmark itself. The bench mark cannot accurately simulate the movement of the mouse. Instead of clicking and holding the mouse button and dragging it across the screen, the benchmark paints separate screens for each coordinate along the path the mouse would travel. The effect of this is that SolidWorks does not have an opportunity to use its feature blocking technique (set by Level of Detail) that removes much of the model detail while maneuvering the model.


If there is a moral to this story it would be "Simplify". Use the functions that are least likely to cause any detrimental effects first. This will help you to recognize the full benefits of SolidWorks with little cost. Unfortunately the same cannot be said for RealView, Transparency, and Display Styles each of which cause tremendous detriments in performance and should be used sparingly.


It is very important to realize that when discussing performance of graphics cards speed can be a misleading term. Our testing measured the cards' speed in painting a certain number of frames in a simulation of spinning our test models from one location to another. The faster it paints the frames the faster it reaches the final position of rotation and finishes the test. In everyday modeling life the final rotation position would be reached at the same time by every card. The performance difference would manifest itself in slower tests painting fewer frames in the given time and thus producing a "choppier" visualization of the spin.




Also, SolidWorks has some built in tools to work around slower performing cards to still allow the user to experience a smooth spin. This usually involves removal of visualization detail to allow the card to paint faster than would be possible if all of the detail remained in the model.


Please
check back to the CATI blog as we will continue posting our
series of articles that goes further into the details of each of our tests. All
of these articles will be stored in the category of Free SolidWorks from Performance
Constraints and links to each with their
release date are listed below:



Thanks,


Josh
Altergott, CATI Support Manager


Adrian
Fanjoy, CATI Technical Services Director





Posted by at
Labels:

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)