Mid Range to High End Graphics - Worth the $

[BOPdesigner]

Our workstations have nVidia FX1500 or FX1700 graphics cards. We are starting to have visualization performance issues with NX 6 and are considering upgrading to FX 3000 or 4000 series. Has anybody had experience with these cards that can comment on increased productivity with the higher end cards? How can you determine what the point of diminishing return on investment is? On the card specifications, is memory bandwidth the key characteristic to consider? We have played around with the visualization performance settings and nothing seems to be noticibly different with the cards we have. Is there a Minimum and Recommend spec for graphics cards to run NX?

 

[cowski]

Where is your display slowdown? a complex part? large assembly? photo-realistic renders?

32 or 64 bit OS? how much system memory do you have? processor? working on files locally or over the network?

In general, more system memory gives the most bang for the buck. You might search for other threads about hardware, I think there was one a while back that gave some info on a priority list of where to spend your money (memory vs processor etc).

 

[BOPdesigner]

Slowdown is with complex assembly manipulation. It is not a large assembly as far as component count is concerned. But the components are complex, mostly imported from STEP/IGS vendor files. Happens on both 32 and 64 bit OS. 4 GB on the 32 bit, 12 GB on the 64 bit. Xeon 5150 @2.66 Ghz (dual core) on the 32 bit and Dual Xeon 5260 @3.33 Ghz (4 cores total) on the 64 bit. Files are loaded over the network.

 

[cowski]

I should mention that I am running an FX1700, upgraded a year or so ago from a 3D labs wildcat something or other. I noticed a slight improvement when working with an entire assembly, but a BIG improvement when using the dynamic view section tool. The old card would just chug when you turned on sectioning, but the FX1700 shows no significant loss of performance for the same operation. The old computer had the same amout of memory but it was dual processor with a slightly faster clock speed than the new single processor quad core. Since view sectioning isn't multithreaded (to my knowledge), I have to attribute the speed up to the new card.

 

[hudson888]

http://www.tomshardware.com/charts/workstation-graphics-charts/SPECopc-Viewperf-10.0-UGS-NX,135.html

This is a long link to Tom's Hardware and it is the only useful graphics card selection tool that I know of. I believe that the Quadro FX 1800 is a comparable newer replacement to the older FX1500 and 1700 series. I'd personally stop spending money right about there unless you are totally convinced that you need to have what appears to be almost double the performance at more than twice the cost keeping in mind that this is still dependent on all the ancillary components of your system keeping pace.

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[BOPdesigner]

Interesting. Thanks Hudson. According to nVidias site, the FX1800 has 3X the memory bandwidth that our FX1700 has. Is that the main contributor to a cards performance?

 

[hudson888]

It is obviously one contributing factor to a card's performance but it is important to note that different cards do differently in different applications. Many good gaming cards are not suitable for CAD applications because the images are delivered differently by the system. Memory and the system that the card is running with also contribute to varying degrees. Beyond that like you I find benchmarks like Tom's pretty useful because more detailed understanding and analysis eludes me also.

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[DRyan34]

I've been working about 3 years now with an nVidia FX4500 card that has 512 MB memory. Being in the consumer products industry I do significant work with renderings and occasionally get into some assemblies. I've never had a video issue with this card. I'm also running 32-bit NX on a machine that has 4 GB memory. I've found that in addition to Visualization settings, you can also tweak the memory allocation from each source and this provides some relief on weaker graphics cards. Another suggestion is upgrading to a duo-core or quad-core processor to help your gpu process images quicker. Actually, this really helps cut down on part history update times. All in all, the investment into the high-end graphics card, about $1000 in this case, may seem a bit much but will bring benefits for a long time.

 

[BOPdesigner]

Thanks. I have been evaluating our FX1500/FX1700 graphics performance using those SPEC benchmarks and it seems that an upgrade to an FX 1800 might give us the most bang for the buck. The NX setting that I have noticed that makes the biggest impact is to use the the lightweight reference set on assembly load options. However I am not sold on it yet and at this point perfer the old Secne Reduction Settings in the visualization performance per NX5. There the components would slip into a bounding box display style as you manipulated the assembly, but then return to the solid detailed view when you stopped. With that being removed in NX 5 and replaced with using Facet Based visualization, you now have to switch components between facet and model references sets as you work on the assembly. This seems to require more work, more picks and clicks. Also, when I switch between reference sets for a component, all instances of that component change to the new reference set in the assembly. Is there a way just to change the reference set of the individual component of interest?

 

[hudson888]

What most people do is to either load "Structure Only" and/or set the default reference set to "Facet" either by default or manually whenever opening larger assemblies. It just makes good sense to do so and it is how the system is designed to be used with large assemblies. I would just add that you get huge benefits in doing so for ANY graphics card, and upgrade would simply extend the point at which you find assemblies becoming too large to open fully.

Other settings which help with performance with larger files are the use of Partial Loading and turning off automatic updates to wave geometry links. Neither of these have any great relationship to graphics performance, but it is ordinarily hard to tell the difference.

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[NXMold]

I just did a very informal seat-of-the-pants test between FX580, FX1700, and FX4600. No difference at all in rendering, rotating, or sectioning so we bought a boatload of FX580's for the price of one high end card. We don't do any fancy rendering... just regular modeling with transparency and big assemblies (several hundred parts, the few most complex parts are 10,000+ surfaces each). There are significant differences in other areas of the system, but what we bought is absolutely the best bang for the buck. Bumping up to the i7-975, fx1800, and 1866mhz memory gained another 5-6% calculating speed, at twice the workstation cost. Visualization performance seemed identical.

NX6.0.2.8 64-bit 6gb-1333mhz i7-920 overclocked to 3.2ghz SSD FX580

NX 6.0.2.8 MoldWizard

 

[BOPdesigner]

Strange. We will be evaluation some higher end cards soon as well. Is there an information option in NX that shows the number of faces, system memory, etc. that an assembly is using? I am curious to to see the stats for a couple of assemblies that are giving us problems. Also I am wondering if dumb imported solids consume more memory and might negatively affect graphic performance than if the same geometry were created native in NX. Can somebody comment about this?

 

[hudson888]

Take a look again at Tom's benchmarks which correlate with my own experience when using a decent amount of geometry as opposed to more general light use. MXMold may care to comment further on the content of his testing. What I have more usually done is to find both a single model and a large assembly that produce some noticeable effect when using the least rated or least expensive of cards that I'm wanting to test. Then I save each a a bookmark. I shut down UG and then I open the bookmarks and perform a range of tasks under a stopwatch, swap the cards and look for an improvement. I am very focused on bang for your buck approaches but I have also had experience of poor graphics performance in the past. I'm looking for noticeable improvements in the same order that Tom's charts are indicating or I simply wouldn't part with the freight!

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[BOPdesigner]

So I have a follow up question about this.

http://www.nvidia.com/object/IO_11761.html

Click on the chart on the top. Scroll down to the bottom to see the SPECViewperf results (3D Application Performance). Why is it that Pro/E and SW and Catia have roughly 2X the frames per second that NX has for a given FX card? In fact, SW has like 4X the fps that NX has. I took one of my assemblies that manipulates at about 4 fps with NX and translated it into SW 2007 (via STEP) and verified that it is much smoother there.

 

[hudson888]

Mind you if I'm reading that chart right then they may be using using NX-1. Which seems odd at this late date. I do know that aspects of graphics performance have changed over time for the later versions of NX as opposed to the earlier ones. Granted that Tom's hardware charts I mentioned earlier where any kind of parallel can be drawn show much the same thing. It would be interesting to know whether the correct interpretation is that other programs are better on graphics performance even while my direct experience indicates that some are much heavier on CPU demands.

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[looslib]

I would be very hesitant to read comparison between the CAD vendors performance in those charts. The actual data is manipulated differently by each system and the makeup of the tests are all different. The number after the CAD name is only the test version, not the SW version that the test was run on.

The only valid comparison is to say that the 1700 has 2/3 the speed of the 1800 when running NX.


"Wildfires are dangerous, hard to control, and economically catastrophic."

Ben Loosli

 

[wmalan]

Looks like the FX580 might be the best "graphics for the cheap". Not the fastest but NX compatible. I'm still on the fence with cam ($$$) consulting but that would really help with costs.

--
Bill

 

[BOPdesigner]

I have finished an evaluation with different graphics cards and have come to the following conclusions:

1) The test results form Tomshardware guide use SPECopc VIEWperf for NX 1. That is what the test was written with and it basically plays back the test program _without_ utilizing your current installation of NX. Starting with NX 5, Siemens integrated a different graphics engine for doing frame rate, scene reduction, etc. and so you can't accuratly predict the results you will get with this test anymore.
2) I can confirm NXMolds findings that there is no significant difference in NX 6 graphics performance across those different graphics cards (although I did not test the FX 4600). So I began to wonder.... I tested this with SPECapc for NX 4. So I repeated the tests with NX 4 and there I found a major difference between different cards. SPECapc uses your installed version of NX and you have to set the visualization performance settings beforehand (basically turn all options off that would improve performance, like disable transluency, etc.).

Now, before you say "well the test written for NX 4 is not valid for testing NX 6". That is a good point, however when using my own native NX 6 assemblies to get a feel for the cards, I found no noticable difference in performance there either. I STEPed an assembly out and imported it into NX 4 and found that frame rate peformance is 2 to 3 times better (again with all performance improvement settings turned off in both NX 4 and NX 6).

Attached are some results of my findings. I have a couple of PRs open on the issue.

Also, in the NX 6 Visualization Performance large model tab, enable the fixed frame rate slider. Fast or Slow, I don't see a difference with NX 6. Please report back if you discover something otherwise.

 

[hudson888]

For the price I'm sticking with my original recommendation. Go the FX-1800.

Best Regards

Hudson

www.jamb.com.au

Nil Desperandum illegitimi non carborundum

 

[jelmerra]

I use an FX5600, main reason is is&v simulation. The fx1700 is a big disappointment in is&v it is virtually impossible to rotate the model when doing the simulation. Apart from that it will remove details from the scene when you zoom in. For ordinary cad stuff difference is not very noticeable.