PDA

View Full Version : Lots of surfaces = 5 minute wait for a response from Modeller !



adk
06-28-2013, 12:00 AM
I've noticed that when importing (medium res - 1mil ) FBX models from places such as Sketchup / Esri City Engine, which both produce a bucket load of surface names, modeller basically becomes unresponsive for minutes on end.

270 Surfaces for the one I'm working on now
175 Image maps
290 MB worth of texture memory

If I load the original model & make all viewports wireframe - no problem.
Change the perspective viewport to Textured Wire - I can go and make a coffee before modeller wakes up
Try to select all surfaces - make another coffee
Change all diffuse parameters to something - make a cup of tea.
I'm basically becoming a caffeine addict.

Turning off all the fancy display features in GL does make the original swap from wireframe to textured wire a bit faster, but not a great deal.
Turning all materials to manually update = nothing really

Crunch that down into a few surfaces and it's not an issue.
(I run a 2 year old HP z600 here at work, Win 7 x64 with a decent Quadro card & 16gig RAM so it's not such a bad setup)

I know that it's a bit to deal with loading all those texture maps but come on - this is 2013.
Layout is not a problem. MAX chews this up in a few seconds & I'm yet to test this on some other apps, which I'm sure would be fine with this too.

Anyone else experience this issue ?

More that happy to package this FBX up and send to NT for comments.

3DGFXStudios
06-28-2013, 02:29 AM
Yup I've experienced the same thing with a CAD model that I converted with rhino. Having a lot of layers also causes the same kind of slowdowns. If you cut past it into on layer everything is a lot faster.
I agree this needs to be fixed.

Andy Webb
06-28-2013, 05:08 AM
Yes I've been recently struggling with this, a very major slowdown.

I was sent an FBX model out of Maya which took 15 minutes to load, had 3000 layers and about 300 surfaces, I gave up in the end, LW just crawled.

Just trying to make a selection was a life changing experience :D

I tried Axis scale on it and the wait was 5-10 minutes before I could use it.

JonW
06-28-2013, 03:55 PM
Same thing if I have a complex architectural model with lots of architectural drawing (plans, sections, elevations etc) as layers.

Not a lot you can do, Modeller needs fixing!

In the meantime you could use an SSD (if you can twist the accountants arm) & also go to BIOS & turn off Hyperthreading if your computer has it (this does not cost anything!) With HT off it does help a bit to squeeze a little more single core performance out of your system.

If you leave HT off then you will loose about 20% rendering performance, but you can turn it back on for your final render. If you are back & forth between modelling & rendering you may be better sacrificing HT for a bit more Modeler performance.

Typically for one of my jobs I spend far more time modelling than rendering, so high single core performance is critical.

adk
06-30-2013, 12:52 AM
Cheers for the feedback folks. Much appreciated. I've already crunched my model to one layer so looks like there's not much more I can actually do.
I wonder at what stage with the numerous surfaces does modeller become a crawl? Is it simply the number of surfaces, or is there something else that's causing the slowdown. I assume it's the large amount of surfaces + textures that are giving it grief.
Would be worth knowing so that I can be as optimal with my objects as possible. This might be a job for another app tho as I don't fancy wasting hours waiting just to optimise this model.

Cheers for the hardware tweaks JonW - never knew about that !

JonW
06-30-2013, 05:27 AM
Make sure every surface has some Smoothing even if it is not needed just put it on 1%. This helps screen redraws a bit. This has been spoken about in other threads. I always do this for every surface now.

Which CPU do you have in the box? If it's not an X5570 2.93 GHz which has a Turbo of 3.33 GHz you only have a % of that speed. If it is not one of the top 3 CPU then you are worst off.


My W5580 (x2) box is 4 years old it runs at 3.2 GHz & on 1 core 3.46 GHz (not too bad for modeler, & an old box!).
http://ark.intel.com/products/37113/Intel-Xeon-Processor-W5580-8M-Cache-3_20-GHz-6_40-GTs-Intel-QPI

If I have HT on then Modeler will only use 1 core of the 16 cores of the 2 CPUs. So only 6.25% of the total CPU resources. HT off it is now about 13%. Turn on WTM while working in Modeler to get a feel for what is going on.

Lightwolf
06-30-2013, 06:53 AM
If I have HT on then Modeler will only use 1 core of the 16 cores of the 2 CPUs. So only 6.25% of the total CPU resources. HT off it is now about 13%. Turn on WTM while working in Modeler to get a feel for what is going on.
Ouch... With or without HT, Modeler will always use one Core either. The percentage difference comes from the task manager seeing twice as many CPUs (ven though half of them are virtual) but is not an indication of absolute performance that is available (or even relative performance across a session with HT and without HT).

Essentially, turning HT off is a placebo.

The only way to make sure that it affects anything is benchmarking. Just looking at the task manager doesn't give you the information you'd expect here.

Cheers,
Mike

Danner
06-30-2013, 07:38 AM
it's not just the number of images, the size also matters, so if you have some absudrly large images that you can resize to something more reasonable it will help.

JonW
06-30-2013, 08:48 AM
Ouch... With or without HT, Modeler will always use one Core either. The percentage difference comes from the task manager seeing twice as many CPUs (ven though half of them are virtual) but is not an indication of absolute performance that is available (or even relative performance across a session with HT and without HT).

Essentially, turning HT off is a placebo.

The only way to make sure that it affects anything is benchmarking. Just looking at the task manager doesn't give you the information you'd expect here.

Cheers,
Mike

I have timed the computer with HT on & off. Modeler is quicker with it off. I had some massive files which were taking ages to do things. It wasn't twice as quick but there was a difference. That's what then got my looking at WTM.

It would be nice if Modeler used a few more cores!

Lewis
06-30-2013, 12:31 PM
modeler is slow slouch when it comes to many surfaces (50+ is already "many" for modeler) or number of layers even if model is not heavy and oyu hav eonyl few textures you'l still wait some hefty time when switching surface or changing modes or just simple selecting. Sadly that's not changed for many many years although it's "known issue" so you/we can't do much to change that other than use different software :(.

adk
06-30-2013, 10:58 PM
Cheers again for your feedback guys
JonW - yeah I've been using the smoothing trick for a while now & in this case it's of little use I'm afraid.
Danner - what you suggest can speed things up… well sort of.

I had time to run some tests and this is basically the result ...

Created 3 tesselated balls each with 320, 720, 2000 polys & a corresponding number of individual surfaces.
No texture maps used in this case.

Load each one of those into modeller with Smooth Shade enabled only in the perspective view (rest are wireframe). Each one only takes a split second to load & there's only a minor slowdown on the last (2k) ball.

With each one loaded I open up the surface editor & multi select all surfaces & change luminosity / diffuse / smoothing an all surfaces. The first two are very responsive where as the last one (2k) takes a while to multi select all surfaces. Once done tho it's pretty responsive. Not stellar but workable.

With each loaded up I then drag a few verts around and see how responsive that is then switch back display modes from Smooth Shade via all the modes & back again. Again the last ball displays a minor slowdown but it's workable.

From this I guess I can conclude that it's not the number of surfaces.


Now to the FBX model … (1.2 Mil polys, 269 Surfaces, 175 image maps, 1.3 MB texture memory - pretty low now since I rescaled all the textures to fit within 56x56 pixels). Perspective view set to Smooth Shade.


Loading…
1 surface = 8 secs
269 surfaces = 210 secs

Once loaded…

Toggle Perspective viewport display … pretty snappy & workable. Final display in Wireframe.

Open up surface editor, multi select all surfaces & change luminosity / diffuse / smoothing all at once.
… pretty snappy & workable.
With all surfaces selected change their option (refresh preview) to manual.

With surface editor still opened toggle to Textured Wire mode.
… coffee time.
Try to multi change surfaces while in Textured Wire mode.
… coffee time.

Back to wireframe mode to drag some verts around. Quite slow & not really workable (we all know this already).
Toggle back to Textured Wire mode.
… coffee time.

One thing I have noticed is that even tho all my surfaces are set to manual refresh mode - when you try to toggle things such as smoothing, exclude from VStack, double sided - refresh mode turns to auto & response goes back to a crawl. Surely that's a bug !

If I disable all the viewport displays I can multi select all surfaces, change & toggle them instantly.
But turn one of the viewports on to Smooth Shade & above = coffee time.

I basically see little in terms of options to effectively work with any sort of high multi surface / multi textured object when the surface / texture count rises beyond a certain amount (don't know what that is yet). Texture sizes seem to make little difference.

JonW
06-30-2013, 11:16 PM
I turn off viewports if they are not really needed for these really heavy situations. Maybe you can work with 3 instead of 4, & where you can turn off 2 viewports.

Bump/spec maps, greyscale png to reduce file sizes?

adk
06-30-2013, 11:29 PM
I turn off viewports if they are not really needed for these really heavy situations. Maybe you can work with 3 instead of 4, & where you can turn off 2 viewports.

Bump/spec maps, greyscale png to reduce file sizes?

Cheers mate. Yeah I think I will go down to 2 viewports if need be & will also try to see if the knocking down of the bit depth helps - good idea.
Tho it's too much jumping through hoops for my liking.
Hopping into another app then back into Layout at the end would be be easier in this case.