PDA

View Full Version : Impossibly slow openGL performance in modeler



MrFurious
07-05-2015, 02:35 AM
I have a single model of a room filled with furniture and decor totaling about 2.5 million polygons, not what I'd say is a huge poly count by todays standards. I'm getting around 1-2 frames a second openGL performance in Modeler, yet c4d, modo, max eats this file up for breakfast with room for seconds & a tall glass of guava juice. AND..... the version of Modo I'm trying this in is v601 which I purchased 4-5 years ago...and it's still wiping the floor with the current version of Modeler. In case you're thinking: 601 was a 'lighter' package so stands to reason- no... 901 is even faster.

I've tried this on 2 pc's including my workstation which is sporting a pretty gutsy (although somewhat dated) GTX 590. Tried in 11.6.3 ans 2015, no difference (so much for those --**AHEM!...**-- 'under the hood' optimizations, I for one am not convinced. It still looks, feels, has the same sluggish opengl performance as previous versions since v9.6)

Needless to say this is a workflow --**K I L L E R**-- I've reduced all the openGL options in modeller to barebones settings and still slow as molasses. Switching from single to quad viewports = 27 seconds!! = more time staring motionless at a screen than actually getting work done = stress & frustration = missed deadlines = angry forum posts.

Attached is a download link if anyone wants to try this on their system but warning it weighs in at about 108MB rarred.
http://www.mediafire.com/download/2khx6h5ajzy21ol/room.rar

OFF
07-05-2015, 04:12 AM
On my configuration (Quadro 4000) Modeler performance on this object is quite high.
Modeler does not have a very high performance on game cards.
But with professional graphics almost as other modeling packages.

MrFurious
07-05-2015, 04:41 AM
I wouldn't be so quick to call the GTX590 a 'gaming' card. My 'consumer level' graphics card is a GPU rendering powerhouse, sought after by graphics professionals and leaves the Q4000 for dead in that regard. In any case.. your reply says nothing about modeler's performance RELATIVE to other packages for high res objects. I mean, try the same model with your Q4000 in say modo, c4d or max and you may see a huge difference. What frame rates do you get on your Quadro with this model?

inkpen3d
07-05-2015, 05:04 AM
Hi there,

Loading your model onto my system takes ~1min, and then ~20secs to display in all 4 viewports. After that though, I can rotate and zoom into the model in the various viewports fairly smoothly in both wire frame and textured views - there is some jerkiness, but it's perfectly usable (i.e. it's not freezing whilst updating the viewports). However, I do get the ~27 sec delay you've experienced when switching from single to quad viewports.

Assuming that you're using C4D and Modo on the same machine(s) it's got to be the clunky old code in Modeler that's at fault. The "Hydra" geometry engine that's used in Chronosculpt has far superior performance when manipulating large poly-counts. Likewise, "Core" had good performance (but let's not go down that well trodden path!).

I'm no expert on this subject (so someone else might want to chip in here) but I understand that Modeler is notoriously bad at utilising GPUs, so the CPU and available RAM are probably going to be more influential on performance. How much system RAM do you have on your machines? My 3GHz workstation (with an old NVIDIA GeForce GTX 460 graphics card) has 24GB of RAM, so I've still got plenty of headroom, even though, when loaded, your room model results in Modeler using an extra ~2.5GB of RAM. I don't have C4D or Modo on my system, but I do have Rhino 5, which interestingly only uses an extra ~1.7GB of RAM with the room model loaded and manipulates the model fairly smoothly, which hints that its got a more efficient mechanism for handling the model data (and/or it better utilises the GPU).

OFF
07-05-2015, 05:15 AM
I have worked on projects with a huge number polygons. Very often it has been done in several programs - so I had the opportunity to compare their performance. Yes, usually Modeller inferior in performance to its main competitors, but not so much when working with professional graphics.
Does not really matter how much the memory on your game graphics card, or how much it produces FPS in games or how fast it made on GPU computing - in the modeler it will behave worse than the most cheap professional video.
In other programs, as far as I know it has been made specific code completion, allowing to bring the performance gaming graphics card to professional graphics cards.
But Modeler that it has not.
No, I have not yet measured FPS in modeler - Now here comes the big scene rendering, so I do not want to force the computer to excess.

Lewis
07-06-2015, 05:55 AM
I think there is some misunderstanding in testing here.

- Do you test OpenGL tumbling or moving/rotating models ?

Tumbling 2.5 mil polys in OpenGL is no problem for modeler that's fast even on moderate gaming card (with VBO turned on OFC) but EDITING geometry with that polycount is going to be slow on ANY GPU/CPU in modeler simply 'coz LWModeler geometry engine is dead SLWO. And no, NT/LWG3D didn't do "under the hood" changes and fast optimizations for modeler in LW2015 (nor many years earlier) so not sure where did you got idea it's going to be different than in LWM 9.6 :)

BTW only thing what will work better in Quadro GPUs than GTX GPUs (yes all GTX are considered Gaming market GPUs regarding of their price) in LWM is double sided materials. Nvidia killed of that feature in drivers around GTX 2xx series so if you have double sided surfaces in modeler it's going to work slower than single sided surfaces while Quadro still works normal in both conditions. Also Quadro has more performance in Wireframe modes. But all that is just for Tumbling, for editing it's still dead slow on both. Also smoothing angle on surfaces (even if is jut 1) helps for faster tumbling.

Granted that applies only to VBO accelrated modes (Wireframe, Texture, Texture wireframe, Smooth shaded) other modes didn't get VBO acceleration around LW 9.2 when it was introduced (like wight or shading color) so they will work slower.

erikals
07-06-2015, 08:26 PM
tweaking polys is still darn slow in Modeler, always has been...

i'm "lucky" as i don't import ArchViz interiors, but if you do... well... not sure what to say, i heard rumors that an integrated "Hydra" engine could fix some of it...
but that could take time, if at all possible...

:/ sorry...

Snosrap
07-06-2015, 10:35 PM
Yep the object kills Modeler. But I can't really do anything with it in Modo either. Sure it tumbles alright, but transforms suck. LW can handle this fine if all the pieces were cleanly remodeled as individual models and then brought into Layout for placement. That is the prefered workflow anyway as anyone trying to building something like this as one big model is clearly making a mistake.

JonW
07-06-2015, 11:11 PM
Give every surface some level of smoothness, say 1% for surfaces that don't actually need any smoothness. It will help Modeler viewports a bit.

erikals
07-07-2015, 10:53 AM
testing the scene...


http://youtu.be/1aEksR8jLLk

best advice i can give you at the moment is >

- copy objects to separate layers
- use proxy objects

meshpig
07-08-2015, 01:24 AM
... 170 000 or so polys tied up in this sort of ambiguity? :)


128868

MrFurious
07-25-2015, 09:32 AM
Wow.. so much good advice here many thanks guys! I've also been researching this quite a bit myself as the openGL speed should really be addressed in future revisions of modeler. Layout too, but It's really showing it's age in Modeler.

@OFF: "Modeler does not have a very high performance on game cards...in the modeler it will behave worse than the most cheap professional video"
Is it true then that I can purchase a Quadro4000 from ebay for $175.00 and this will outperform a $2500.00 TitanZ in modeler? If so I'll have to reconfigure my GPU setup and dedicate a Quadro purely to drive my monitors and use GTX(xxx) for GPU rendering. I understand there's barely any difference between the GTX and Quadro cards except the ram.. and drivers. Apparently many GTX cards can be (soft?) modded so your pc sees them as Quadros, will then work with Quadro drivers and offer all the openGL speed benefits.

@inkpen3d: My CPU is an i7 4820K @3.7ghz, and 16 gigs or ram. I could probably use another 16 gigs but with this mesh open in modeler and a few other programs open. I'm only using about half of what I have, so lots of ram to spare. I've yet to give Chronosculpt a look, and with Core I opted to sit on the fence until something production ready came of it. But Core happened some time ago now and I have to wonder why today's modeler doesn't utilize some of the technology from Core. NT lost me at version 10, back on board with 11 and not quite ready for 2015, maybe if a future point release will impress.

@Lewis: Gday fellow OctaneWaver ;) pretty much all of what you said confirms what I'm finding on my end. Especially the double-sided polygon slowdown. Even in Layout if I turn DS off for everything, the speedup is immense. And interesting to learn about the older Gforce drivers supporting this in older drivers because I really only noticed this since upgrading from a GTX 260, so that explains a lot. I wonder would it really be that difficult for NT to add a global checkbox in the display options tab, or even a second double-sided checkbox (for openGL only) in the surface editor. Ok, so maybe it's a workaround for those using consumer cards with geforce drivers but I'm sure many of us would appreciate the speed boost. And how many newcomers would be sporting Quadro cards? my guess is very few. Re: the under the hood changes.. maybe I've been living in a different reality than you but I've been seeing those words, or words to that effect, being thrown around by NT for YEARS.. words of promise, words of hope. A cursory search returned this: [Rob Powers] "I will say that under the hood work has gone into foundational modeling architecture and we are serious about innovating with our modeling tools." to which I fully agree with one reply from that thread "Its not only about what is going on under the hood, its about what we use, see and feel and Modeler feels ancient."

@erikals: always happy to watch one of your vids. I replicated the layer separation you used in that video but my modeler is still nowhere near as fast as yours even with all surfaces single sided and at least .1% smoothing, what video card are you using?

"i heard rumors that an integrated "Hydra" engine could fix some of it..." I've been around long enough to know that LW & rumors = disappointment, I pay them zero attention now and at the risk of sounding preachy, would suggest you do the same. I don't blame NT for their tendency to be conservative in their communications and given LW's recent history I'd go as far as to say it's a smart play.. for now. I would like to see things improve in this area though and I'm optimistic they will.

Yes.. the only way to tackle a model like this is to break it up into smaller objects, but what your video didn't show was starting with the full model on a single layer, and the painfully slow process to get the geometry into separate layers.. viewport navigation, selection, copy and pasting. As Snosrap and yourself said, this isn't really the way to go about a project like this and I guess regardless if this is the correct application or not that's not the point. Given the relative geometry count it is painfully slow. The problem is having to play along in a 3ds max world. The only 2 options when I'm given a max file like this are; A: export the whole lot into an obj like the one in this thread, manually break it up into components/layers, setup pivot points and hierarchy's in modeler etc. Or B: (my default now) is export as an FBX, and drop it straight into Layout. This actually works surprisingly well! Provided scale and rotations are setup correctly in Max, once the FBX is imported into Layout it's pretty sweet.. Layout creates it's own project folder group and places everything in its appropriate folder (objects and scenes, not images though) Cameras are supported (with a little tweaking), all 3ds max groups are preserved as layer hierarchies and all pivots preserved as well! Textures translate perfectly so long as everything is UV mapped. The only thing OBJ has over FBX at this time is that the 3ds max obj exporter can flatten procedural textures to a single bitmap, this needs to be done manually with FBX. So... yeah it's FBX all the way from now on.

@Snosrap: "LW can handle this fine if all the pieces were cleanly remodeled as individual models and then brought into Layout for placement. That is the prefered workflow anyway as anyone trying to building something like this as one big model is clearly making a mistake" This room wasn't modeled as a single model... it's an obj exported from a 3ds max scene. Thankfully I've since learned that FBX does do what you mentioned and does it well. Interesting what you say about Modo, I just tried it once more and imported the room as a single mesh into Modo 601 and it's fast for me; tumbling, transforms, selections, pasting to new layers, all very quick except copying is bit slow and a slight lag when swapping transform tools (maybe one second). Realtime soft dragging like in erikals video, except with all geometry on a single layer. No prep work, no optimization tricks just import and go. I understand Modo is optimized for Intel CPU's perhaps if you have an AMD chip this might explain it?

@meshpig: "... 170 000 or so polys tied up in this sort of ambiguity? " That's right.... exactly. Welcome to the world of 'getting the job done'. Sometimes it's faster and easier to grab a stock object and not have to worry about the poly count or having to 'optimise', just drop in and go. It's a shame much prep work is often required when working with modeler to be able to work fluidly. (FYI in case you hadn't considered, the entire mesh has been deliberately distorted before posting here)

erikals
07-25-2015, 10:30 AM
@erikals: always happy to watch one of your vids.
thanks! happy to hear it! http://erikalstad.com/backup/misc.php_files/smile.gif


I replicated the layer separation you used in that video
but my modeler is still nowhere near as fast as yours even with all surfaces single sided and at least .1% smoothing,
what video card are you using?

right now i'm using a GTX 470, so kinda old, but i've used even older ones, and the OpenGL speed seems to be close to the same.
with LightWave it's more about the CPU i think, having a fast Quad Core CPU
a Dual Core with a very high speed would be the best for Modeler, as Modeler is only (unfortunately) optimized for 1 Core

afaik at least...

still, very strange, as i read you have >
i7 4820K @3.7ghz, 16G ram
i have >
i7 2600K @3.4ghz, 8G ram

beware that antivirus apps can eat CPU power, maybe try to turn that off, see if it makes a change

MrFurious
07-25-2015, 12:17 PM
Hmm.. so my system should fare well considering yours is lower spec. Will definaltely try disabling antivirus but something else just occurred to me. The GTX590 I'm using to drive my screens is in the bottom PCIe slot ie the one farthest from the CPU. I think this one only runs at 8x PCI lanes? So gonna have to try this again but from the top slot which apparently is 16x lanes. Sometimes the obvious eludes me until all other options have been exhausted!

erikals
07-25-2015, 05:24 PM
by the way, make sure you didn't use flat shade...
https://www.youtube.com/watch?v=0LWsRrHnul8

also try >
https://www.youtube.com/watch?v=8zDFeD7z4rQ

Snosrap
07-25-2015, 10:18 PM
Interesting what you say about Modo, I just tried it once more and imported the room as a single mesh into Modo 601 and it's fast for me; tumbling, transforms, selections, pasting to new layers, all very quick except copying is bit slow and a slight lag when swapping transform tools (maybe one second). Realtime soft dragging like in erikals video, except with all geometry on a single layer. No prep work, no optimization tricks just import and go. I understand Modo is optimized for Intel CPU's perhaps if you have an AMD chip this might explain it?
Yeah my chipset isn't the latest for sure - i7 920 2.67GHz - so that might be why it stumbles in Modo for me. Modeler just sucks at high poly counts. :) Hopefully NT can resolve that soon.

spherical
07-26-2015, 01:42 AM
I understand there's barely any difference between the GTX and Quadro cards except the ram.. and drivers.

Absolutely true.


Apparently many GTX cards can be (soft?) modded so your pc sees them as Quadros, will then work with Quadro drivers and offer all the openGL speed benefits.

This hasn't been possible since they closed that hole many years ago. Only thing you can do now, and these are becoming ever-limited (because they're losing sales of too-expensive cards for a limited market) is to hardware mod the card by bridging/cutting traces and surface mounting resistors where necessary.