PDA

View Full Version : Why there is this incredible speed difference?



AmigaNewTek
07-25-2011, 09:10 AM
http://vimeo.com/25913180

Netvudu
07-25-2011, 10:15 AM
Softimage is good at handling many polys, but the problem with Lightwave mainly comes from the fact that you are working with Modeler.
Layout accepts many more polys than Modeler. Iīve read some time ago that one of the problems with Modeler is that it draws once per view, so it is way slower. I donīt know if this is still true currently but I suspect it is.

Sensei
07-25-2011, 10:40 AM
But you're just spinning viewport in this video.
So, everything goes from VBO buffers.
When geometry is modified, it needs to be 1) put to undo buffer 2) transfered to VBO- optimal transfered just changed part.

Rayek
07-25-2011, 01:08 PM
It has to do with Modeler not using Vertex Object Buffers in OpenGL, I think. Because Blender has an option to turn on or off VBOs, I repeated the same experiment, and low and behold: with VBOs turned off I get slow performance (still twice as fast as Modeler). With VOBs turned on, there is just no comparison: with eight copies I still get much better performance than one copy without VBOs.

Read this for more info: http://www.spec.org/gwpg/gpc.static/vbo_whitepaper.html

Of course, without more info about the hardware used, the resolution of the screen, and so on, it is hard to make good comparisons.

But modeler's opengl is, as we all have realized a long time ago, very slow. Core had great opengl performance, I read - so that performance will hopefully be part of the next version.

Sensei
07-25-2011, 01:25 PM
In Modeler there is VBO on/off toggle in Display Options > GL tab.

But OpenGL can be multi-threaded or not multi-threaded. I am not sure whether LightWave uses multi-threading for transfering data to gfx card. I think so it's not, because on dual-core cpu in Task Manager while spinning viewport I see CPU usage 50%.

Using wired modes make things worser.

hrgiger
07-25-2011, 01:44 PM
Yeah, spinning doesn't impress me a whole lot. And it's not the big problem with LightWave. Try manipulating some of those components. You can spin a lot of polys in LightWave. But when you go to move points or polygons, it slows to a dead crawl.

Rayek
07-25-2011, 02:09 PM
Thanks, Sensei - I wasn't aware of that vbo option - though by default it is turned on, and still the view is quite slow.

Sensei
07-25-2011, 02:37 PM
Open Task Manager and start spinning viewport in other 3d apps. Do you see all cores are 100% while doing so? In LW it's just one core 100% and the rest is pretty much idle. And don't forget to check the same viewport settings, LW viewport in Textured mode and equivalent mode in other app.

Cageman
07-25-2011, 02:55 PM
Just a note here...

LWs OpenGL speed is dependent on Smoothing being ticked ON. If you have it off, you will notice a slower performance even with as much as 500k polygons. Make sure you set Smoothing ON in surface editor for all your surfaces. I had a scene here with 48million polygons and with Smoothing ON, it was still way faster than a 1 million poly object with Smoothing OFF.

:)

cresshead
07-25-2011, 03:01 PM
to be fair modeller code is old...try that with Core and you'll see much better performance.

Rayek
07-25-2011, 03:49 PM
Just a note here...

LWs OpenGL speed is dependent on Smoothing being ticked ON. If you have it off, you will notice a slower performance even with as much as 500k polygons. Make sure you set Smoothing ON in surface editor for all your surfaces. I had a scene here with 48million polygons and with Smoothing ON, it was still way faster than a 1 million poly object with Smoothing OFF.

:)

I would be interested to know which video card/system specs you have - because in my case 8 trees (generated in Cinema4d 8 with xfrog 4) with each tree at 440.000 vertices/260.000 polys, even a single tree lags in Modeler, whether I turn on smoothing or not. In Blender I can duplicate the tree and create a small grove of them without a real slowdown (10 trees). And even while orbiting/panning/zooming at 1920*1200 the view, less than 9% CPU time is used. Even at 30 trees Blender's viewport is still not lagging, and relatively smooth. Selecting vertices and polys is still instant.

Selecting even a couple of vertices in Modeler becomes a pain, and is very slow for this one tree. No such problems in either Blender (working with the entire grove of trees), or even the aging Cinema4d v8 (with either one tree or 15 trees or more).

This is one of the main reasons why I left Modeler as a modeling tool (well, for the most part- I do still use LWcad sometimes). It is hard to justify NOT moving to another modeling app if the viewport's just so dead slow.

My system specs: [email protected], 12gb ram, ati 5870 1gb.

Rayek
07-25-2011, 03:53 PM
Oops, did not see your signature there - it seems we have just about identical specs, but for the video card. Yours is an old nvidia card. Maybe Modeler's opengl works better in geforce cards? Then again, before this card I had a 285gtx - and the performance issues were the same.

Myagi
07-25-2011, 04:27 PM
In Modeler there is VBO on/off toggle in Display Options > GL tab.

But OpenGL can be multi-threaded or not multi-threaded. I am not sure whether LightWave uses multi-threading for transfering data to gfx card. I think so it's not, because on dual-core cpu in Task Manager while spinning viewport I see CPU usage 50%.

Isn't the problem rather that it's constantly pushing a bunch of stuff over the bus to begin with, instead of sticking the primitives in an index buffer (in gfx mem) and doing one draw call?

Not that I have any idea what Softimage and Blender do, but in the posted videos there was nothing that couldn't be done that way, since everything hade the same surface/material.

(Or two draw calls for shaded+wireframe modes)

Cageman
07-25-2011, 04:47 PM
Oops, did not see your signature there - it seems we have just about identical specs, but for the video card. Yours is an old nvidia card. Maybe Modeler's opengl works better in geforce cards? Then again, before this card I had a 285gtx - and the performance issues were the same.

Sorry... I should have been clear that the tests I did was in Layout... that said, Modeler should be able to handle similar data in pretty much the same speed, since, when it comes to tumbling/orbiting something, it doesn't need to send data to the GFX card all the time, unlike when you select something or actually deform something. But, to my knowledge, Modeler and Layout should have the same OpenGL optimizations?

In any case...

Moving an object in Modeler is not the same thing as moving an object in Layout though. That said, deforming something in Layout is similar to moving an object in Modeler. When you do this stuff, LW will crawl, even on a low number of polygons.

Cageman
07-25-2011, 05:24 PM
I would be interested to know which video card/system specs you have - because in my case 8 trees (generated in Cinema4d 8 with xfrog 4) with each tree at 440.000 vertices/260.000 polys, even a single tree lags in Modeler, whether I turn on smoothing or not.

Just to be clear here... are you talking strictly orbiting the object as in not trying to deform it in any way (moving verts or whatever)? (we should have established, at this point, that LWs deformations/editing speed is very, very slow).

Also, does this tree consist of several layers that you view at the same time, or is it a single layer object?

If orbiting this object and if it is a single layer object, and Modeler is as slow as you describe, there is certanly something fishy going on.

So...continue on the speed of orbiting objects in Modeler...

Which version of LW are you using? I am testing LW10.1 Modeler now, and I have 3 layers visible at the same time, with the combined number of polygons reaching 2.654.208, and in Smooth Shade view, it very fast to tumble around the objects. I do sense a slight slowdown compared to not having any objects at all, but it is a tiny slowdown for sure.

Worth mentioning is that Modeler seems to be as fast with Smoothing turned on in Surface Editor as without smoothing enabled. So, Smoothing seems to only help Layout. And that it does with an order of magnitude.

netstile123
07-25-2011, 05:42 PM
I tried the same thing with 10.0 with the same performance as softimage. I don't see the meaning of the test?


I had some problems moving a single object. But no problems moving the view

XswampyX
07-25-2011, 06:07 PM
Try running this.... http://www.gremedy.com/

It's an OpenGl profiler. I think the saying is "If you can't say anything nice, then don't say anything at all." So I won't comment.

http://i465.photobucket.com/albums/rr16/xXswampyXx/Progiler.jpg

Try running Modeler and then 'something else'.

Sensei
07-25-2011, 06:24 PM
hello?! In vimeo video there was 11 million polygons..

but who made this video didn't specify his computer specs, nor software version, 32/64 etc. That's unfair.

speismonqui
07-25-2011, 06:57 PM
to be fair modeller code is old...try that with Core and you'll see much better performance.

Great! can you get me Core?
:D

Intuition
07-25-2011, 06:58 PM
to be fair modeller code is old...try that with Core and you'll see much better performance.

Aw Cress, your hurtin me man. ;D

.... old wounds re-opened.

;)

Naw but seriously. The correct statement would end here "Softimage handles".

Dexter2999
07-25-2011, 07:11 PM
hello?! In vimeo video there was 11 million polygons..

but who made this video didn't specify his computer specs, nor software version, 32/64 etc. That's unfair.

That was the first thing that came to my mind.

For all I know the video could have been made to show off his system, not the software. (Watched it on my phone so pardon me if I missed something.)

3D Kiwi
07-25-2011, 09:20 PM
Its been well known for a long time that XSI handles huge objects and scenes well. In fact that is the sole reason we changed to XSI from Lightwave. Cant remember if it was version 7 or 9 we were on but the scene I was working got to the point that Lightwave choked. Not just viewport performance but it wouldnt render it either. Took the scene via .lwo and mdds into XSI 5 and it was flawless. One thing i have found that XSI dosnt like is huge amounts of objects. If you had 1000 spheres merged into a single object XSI just laughs at it. But If had 1000 single spheres then its a different story. (this has improved over time tho) There was a paper around ages ago were they tested Max, maya and XSI for this kind of thing and XSI beat them hands down. I would say the playing field is a bit more even now and when Modeler gets some work it should improve as well you would hope.

AmigaNewTek
07-26-2011, 02:09 AM
In my opinion, modeler (but layout too) don't use any acceleration at the moment. Don't forget that current generation of processors have some math units MMX, MMX2, SSE (1, 2, 3 4) 3DNOW units and so on...
Rather that simple turn on the compiler option to use this math coprocessors, the code shuld be optimized for it. To not talk about GPU. Gpu today can be used for general computing tasks.

colkai
07-26-2011, 04:18 AM
Worth mentioning is that Modeler seems to be as fast with Smoothing turned on in Surface Editor as without smoothing enabled. So, Smoothing seems to only help Layout. And that it does with an order of magnitude.

So are you saying that an object with no smoothing is slower than an object with smoothing and an angle of zero? (or say 0.5)? Or does the smoothing angle need to "kick in" to notice the speed improvement?

AmigaNewTek
07-26-2011, 04:54 AM
This is why i often become frustated with Lightwave. The software and the technology is great, but there are some minor issues that prevent me to have a good learning curve. I was trying to create a waterfall scene with high poly number, but LW was too slow, both in modeler and in layout. Simulating a particle waterfall is complicated too. Calculations are done by one processor only (@ 25% utilization). 3 Cores are unused on my quad system.

Sensei
07-26-2011, 05:36 AM
You should blame cpu manufacturers, that they started adding parallel cores, instead of increasing clock. Dynamics is especially b*****ch to make multi-threaded, you can't start making frame 2, if frame 1 is not finished, like in rendering, or splitting image to regions and each core is working on it's own region of image.

clagman
07-26-2011, 06:16 AM
Well, it's not so easy to increase the clock is it. Limits of manufacturing and materials and all that. I think the paradigm has shifted to the worse in some ways. Back in the day, large vector computers would eat up nastran simulations because they made the CPUs wider. Although we are sort of making a comeback to this with GPU computing the truth is that the rather limited vector components they put in the CPUs could be improved on drastically.

Lewis
07-26-2011, 07:27 AM
1. Your openGL modeler settings are set wrong :). Go to GL tab in modeler, switch shading mode to MTS (instead GSLS you are using now) and turn on VBO. That will turntable FLAT gray polys (textured and transparent is anther story) pretty fast at 1.7million polys.
2. Just don't manipulate (select/deselect/move/array/drag....) polys and it'll be fine. as soon as you start draging them modeled slows down to crawl speed :(.

jasonwestmas
07-26-2011, 09:26 AM
I think the point that was made earlier is that the slow downs in LW in general have less to do with OGL rendering algorithms and more to do with deformation and editing code. Sooo, tumbling an object as it's being rendered in OGL is not going to be much different (depending on your OGL settings) from other apps till you try to edit the object(s) or start keyframing them etc.

Rayek
07-26-2011, 08:57 PM
I think the point that was made earlier is that the slow downs in LW in general have less to do with OGL rendering algorithms and more to do with deformation and editing code. Sooo, tumbling an object as it's being rendered in OGL is not going to be much different (depending on your OGL settings) from other apps till you try to edit the object(s) or start keyframing them etc.

Well, I beg to differ: I have tried every setting and combination of GL settings, as suggested in this thread, in Modeler, and the opengl performance does not really change one bit. The cloned balls lag enormously, and orbit smooth as a hot knife through butter in Blender's opengl; even old Cinema4d v8.5 is about three times faster and workable.

I tried using the opengl tool previously posted in this thread, and performance is not really measurable (less than 0.1 fps) when I test LW to see what the problem is. It is kinda odd other people in this thread state they get good frame rates at over a million of verts - in my case at a 1.17 million polys the lagging when trying to simply orbit or pan is absolutely horrible, and unusable.

It's very frustrating, because I'd like to use lwcad more on higher poly count scenes, but for some reason on my system Lightwave modeler (v9.6) does not perform well at all in the opengl department.

Does anyone have a clue what's going on with LW on my system (ati5870 1gb, [email protected], 12gb)? Is v10 that much better? I would really appreciate the help.

Sensei
07-26-2011, 09:00 PM
Replace gfx card..

Rayek
07-26-2011, 10:15 PM
With what exactly? I already had a 285gtx, which performed much worse than the 5870. And I switched a 480gtx for the 5870 because of the dismal opengl performance of that card. The 5870 works just fine in other apps, just LW modeler seems to lag behind a lot.

Cageman
07-27-2011, 12:30 AM
With what exactly? I already had a 285gtx, which performed much worse than the 5870. And I switched a 480gtx for the 5870 because of the dismal opengl performance of that card. The 5870 works just fine in other apps, just LW modeler seems to lag behind a lot.

You have a faulty system... end of story.

Lewis
07-27-2011, 12:38 AM
Can you post screen grab of your modeler GL tab/settings ?

AmigaNewTek
07-27-2011, 03:48 AM
With what exactly? I already had a 285gtx, which performed much worse than the 5870. And I switched a 480gtx for the 5870 because of the dismal opengl performance of that card. The 5870 works just fine in other apps, just LW modeler seems to lag behind a lot.

It's not a matter of graphics card here. You can achieve nice result optimizing the pipeline with a Matrox Parhelia too and other old cards. Yes, you can have extra speed using new high end card (gtx 480) (5870) etc, due to GDDR5 memory, but if the software is not optimized to use GPU or multithreading, you can't reach real need speed.

Using GPU there are already solution than can fire real time ray tracing (1600x1200 @ 26 FPS) fully illumitated fully shaded and with more than 21 M polys using 3 GPU. I can't tell more about his due an NDA.

dualboot
07-27-2011, 08:58 AM
Lightwave's OpenGL is ancient, there is no way of saying it different. It is slower than any other OpenGL or DirectX implementation on the 3D applications market. It is even worse on newer versions of Windows (Vista and 7). From my expirience You will gain alot of viewport speed(~50%) by going back to Windows XP...sad but true. Lightwave does not know how to handle Aero environment included in Windows 7 and Windows Vista. Handling objects over 50 000 polys is nightmare, and most of the time i split object into multiple layers to gain some speed...off course this is not always a possibility and then You need to wait a second or more depending on an object for the command to actually do something that You can see :)

My main expectation and hope for LW11 or LW 10.5(if there will be one) is to have rewritten OpenGL since this one has been outdated for ~5 years.

-EsHrA-
07-27-2011, 10:10 AM
lw opengl is very slow compared to blender and cinema. period.
lw's old outdated code is the culprit here.

Rayek
07-27-2011, 12:38 PM
Thanks, EsHrA and dualboot. I suspected as much - it cannot be caused by a 'faulty system', otherwise I would have experienced issues in all the other 3d applications (and games) I use, and my comp runs dead solid: only one or two crashes in two years.

For now I can live with constructing pieces with LWcad, and import those in other modeling environments for further editing and bringing everything together.

Cageman
07-27-2011, 04:28 PM
Thanks, EsHrA and dualboot. I suspected as much - it cannot be caused by a 'faulty system', otherwise I would have experienced issues in all the other 3d applications (and games) I use, and my comp runs dead solid: only one or two crashes in two years.

Well, you seem to have some extremely slow behaviour related to LW that none of us experience to the same level you do. So, something IS fishy with your system or your settings.

Can you post a screenshot of your settings regarding LW Modeler OpenGL?

Sensei
07-27-2011, 04:44 PM
Different apps might use different functions of graphics card. If there is some slow one in specific driver, and one app will use it, there will be slow performance in that app. But it won't affect other app, not using this function.

Lewis
07-27-2011, 05:15 PM
In the meantime here is screen grab of how modeler should be set for Tumbling speed (just tumbling, editing of million of polys is another story and you can't do anything to speed it up :().

Also be sure you are tumbling in - Texture, Texture Wireframe or Wireframe modes. If you use some other modes you might get massive slowdowns since they weren't accelerated with "VBO" during 9.x period.

jasonwestmas
07-27-2011, 05:48 PM
Also, make sure ogl color space settings are linear too in modeler. Just in case you are using LW10+.

Rayek
07-27-2011, 06:32 PM
Silly, silly me: I changed every setting in the GL prefs, but I am used to working in flat shade mode. I changed to texture mode, and voila: orbiting, panning and zooming is now smooth at 1.2 million polys. Even three times as much is still okayish.

Selection is, however, a no-go. Modeler slows down to a crawl and becomes unresponsive. So I will still have to import all the parts into another 3d app to combine.

Thanks, guys. And sorry for my stupid mistake - should have known better (working in LW for quite some years now). We don't know what we don't know :-)

Lewis
07-28-2011, 12:41 AM
Good now we solved this, actually I agree with you about selecting and working with actual polys modeler is crawling with hi-polycount but tumbling is much better than selecting when in "proper" mode ;).

cheers

jasonwestmas
07-28-2011, 09:34 AM
Yes, we who 'sorta' like the ogl rendering know that selecting and editing masses of polys needs a lot of work.

-EsHrA-
07-28-2011, 10:05 AM
but still..

lw needs a better implementation.
its still one of the slower bunch.

Rayek
07-28-2011, 10:34 AM
The viewport performance and quality of real-time rendering/previewing is one area that got a lot of attention by devs in other 3d apps the last couple of years- didn't Core have some great improvements in that regard as well?

jasonwestmas
07-28-2011, 10:38 AM
The viewport performance and quality of real-time rendering/previewing is one area that got a lot of attention by devs in other 3d apps the last couple of years- didn't Core have some great improvements in that regard as well?

NT is going to rip out more of the guts to put any of that OGL tech in there I imagine. I have to say the ogl and cg rendering in core looked really really nice.

Cageman
07-28-2011, 04:42 PM
The viewport performance and quality of real-time rendering/previewing is one area that got a lot of attention by devs in other 3d apps the last couple of years- didn't Core have some great improvements in that regard as well?

First, nice to see that you got OGL speed working. And yes, there are TONS of improvements needed for manipulating the data, but at least you got orbiting speed up to what LW is capable of. :)

To answer your question; YES!. CORE has some extremely nice and fast OGL capabilities, such as realtime refraction, spherical harmonics ligthing, occlusion, bumpmaps... well... you name it. Some, of which I still havn't seen in other apps...yet. It was demoed at last siggraph and you can get glimpses of it in the LW10.0 presentation video. Since then though, plans has changed. And, as Lewis has mentioned many, many times, CORE is able to manipulate components on objects with millions of polygons. Obviously, it slows down somewhat, but still extremely interactive. Modeler and Layout are lightyears away from the performance that CORE has regarding this.

But NewTek has some very fine tech in their hands. I think it is a matter of time before we start to see this stuff gradually popping up in future versions of LW.

NinoK
07-29-2011, 12:37 PM
But NewTek has some very fine tech in their hands. I think it is a matter of time before we start to see this stuff gradually popping up in future versions of LW.

Man, i just hope by the time we see it, that fine tech doesn't turn into what modeler is now, because after all, fine tech is all relative.