PDA

View Full Version : OpenGL Speeeed



toby
06-12-2006, 12:40 AM
It's been supercharged!

I tested a subpatch model in LW8 and 9, with a total of about 730k polys, in 9 it flies while in 8 it barely moves.

:newtek: = :king:

wavk
06-12-2006, 02:04 AM
is this in the latest rc2?

mlon

toby
06-12-2006, 02:41 AM
Yes, that's what I have

hrgiger
06-12-2006, 10:04 AM
Yes, the OGL in Layout is much improved. It's fairly dramatic.

Unfortunately, no improvement in modeler yet but it's supposed to come later in a 9.x release as I understand it.

G-Man
06-13-2006, 08:49 AM
Hopefully it won't matter as they integrate the modeling tools into Layout as planned. It will all just be FAST.

animotion
06-13-2006, 09:59 AM
Now if they could only get the bones sped up to a descent speed.

Bones really throw a monkey into the wrench. ummm, wrench into the monkey.
uhhh, Gums up the works?
You know what I mean.

Animotion

hrgiger
06-13-2006, 10:30 AM
Yeah, I rigged a character last night and still found bones to be immensely slow.

adrencg
06-13-2006, 01:38 PM
Yeah, I rigged a character last night and still found bones to be immensely slow.

That's a real bummer about the bones. The new OpenGL speed I'm hearing about sounds great...LW9 will be awesome for those projects where I need to spin a model around in Layout.

What good is the new OpenGL speed if:

a. It's only for non-subpatch polygons(correct?)

b. Bone deformations are still dog-slow.

Captain Obvious
06-13-2006, 01:43 PM
a. It's only for non-subpatch polygons(correct?)
In Layout, subpatch polygons and real polygons have the same exact performance, as long as the total number of polys is exactly the same. Check the statistics display (w), and you'll see the total polygon count. If it says 10,000 polygons, it doesn't matter if it's 10,000 real polygons or 1000 subpatches that are subdivided a bunch of times.

wavk
06-14-2006, 11:23 AM
i havent tested it yet but will soon.
btw. if opengl speed is 'upgraded' why isnt it mentioned in the rc2 update list?


mlon

toby
06-14-2006, 11:54 AM
Maybe it was there before rc2.

ericsmith
06-14-2006, 12:46 PM
That's a real bummer about the bones. The new OpenGL speed I'm hearing about sounds great...LW9 will be awesome for those projects where I need to spin a model around in Layout.

What good is the new OpenGL speed if:

a. It's only for non-subpatch polygons(correct?)

b. Bone deformations are still dog-slow.

A couple of points:

1. In my tests, rigged characters are faster than 8.0 or 8.5. It's not nearly as dramatic an improvement as the base OpenGL, but I'm seeing around 20-25% faster interaction.

2. In the bigger picture, the openGL speedup is taking more advantage of video card OpenGL acceleration than it used to. Something like bone deformation (or modeling tasks) still have to perform data manipulation on all those vertices, so it's more CPU dependant. Other apps share this same issue. In XSI, for example, you can spin 500,000 polys very quickly. But try to use a magnet tool on that mesh and it's a very different story. I'm sure it will be the same when openGL improvements are brought to modeler. We'll be able to manipulate the viewports much quicker, but anytime you actually change the model, it has to pass that through the CPU and then on to the video card. You won't be able to dragnet a 500,000 poly model smoothly in real time.

I'm not saying that there's no room for improvement for LW rig interaction speed, but there's a lot of factors in play. LW isn't the only app that will slow down with a complex rig. At Siggraph last year, I was talking with one of the Disney animators that was showing off the rigs they used for Chicken little. I asked about real-time feedback, and it wasn't even close. They were getting around 6 fps with their rig. This was in Maya.

The big problem that Newtek needs to fix is where multiple rigs really bogs things down. From what I've been told, they know what the problem is, and intend to fix it. Hopefully we'll see that one resolved over the 9.x cycle.

Eric

hrgiger
06-14-2006, 01:10 PM
Well, I currently only have one character in my scene righ tnow and it's still slow. I can neither rotate a bone or scrub the timeline in anywhere near realtime, in fact, I'm getting about 3fps. It's pretty painful. I don't have the newest processor out there, but I could hope for more from 2.4 GHZ. I haven't noticed a 20-25% increase in bone speed from 8.5 to 9. So unfortunately, I'm forced to work with a low bounding box threshold to get any kind of feedback from the interface, even with my subpatch display set to 0.

I have much hope and much expectation during the 9.x cycle for making character animation more productive in Lightwave.

ericsmith
06-14-2006, 03:24 PM
That's not how it should be. If you're willing to post the scene and character model, I'll take a look and see if I can find anything to help.

Also, do you have graph editor open? I've had severe slowdowns when GE is open, and for some reason every channel in the character gets dumped into it.

Eric

hrgiger
06-14-2006, 04:09 PM
Eric, I'll email you the files if you'd like but I don't want to share them en masse at this point. What's your email?

Actually, I spoke too soon. I'm getting 15fps in Lightwave 8.5(almost workable) and only 3fps in Lightwave 9. And no, the graph editor is not open.

ericsmith
06-14-2006, 04:48 PM
Send it to support (at) stillwaterpictures.com

I did just remember something. Setting the character to display subdivision 0 significantly slows things down compared to subdivision 1 or above. I've already reported this to Newtek, and am hoping it gets fixed before release. But interestingly, the same rigged character plays back faster at subD 1 in 9 compared to subD 0 in 8.0 (which is actually faster than 8.5). Try unsubDing your character and see what kind of speed you get from that.

Eric

hrgiger
06-14-2006, 05:02 PM
Thanks Eric.

Unsubpatching it only raised the fps to 5 in LW 9. It remains around 15fps for LW 8.5.
I'm sending you the files. Forgive the crudeness, it's a first pass animation.

ericsmith
06-14-2006, 06:17 PM
I emailed you back, but to keep others in the loop on this issue, there were a few things that didn't get unsubpatched - the eyes and beltloops. Once you unsubD them, you should get at least the same speed in 9 as 8.5. For me, it was a little faster (23-25fps in 9, 20-22fps in 8.5) The big difference, however, was when I loaded a hi-poly set into 9. With over 600,000 polys total, I was still getting 12fps.

This is on a single processor AMD 2.6ghz FX55 and nVidia 5900FX.

Eric

hrgiger
06-14-2006, 06:43 PM
Yeah, I knew about the eyeballs and such but didn't think it would make a huge difference concerning the number of overall polygons. I guess it does.

hrgiger
06-14-2006, 06:50 PM
Ok Eric, I went back and unsubpatched the eyeballs and beltloops. Loaded the scene again in both 8.5 and 9. 9 is now much improved with around 13fps but is still not as good as 8.5 which is still holding around 15fps.

Your machine is a little faster then mine but I would think that the consistency would be between 8.5 and 9 but now we have two different users experiencing different results for each version as far as which version seems to handle bones faster.

Well, the one thing this might be showing is that 8.5 still handles subpatch better then 9 since I could still get 15fps with the eyeballs and belt loops still in subdivision.

ericsmith
06-14-2006, 07:29 PM
Yeah, I think the jury is still out on this one. I've done some more tests, and found varying results based on lots of little factors.

As far as 8.5 handling subpatch objects better, it's only at level 0 that 9 really bogs down. As soon as you compare 8.5 to 9 at a subpatch of 1 or more, there's much less difference. And as you've discovered, even a few polys that are subpatched causes the problem. It definitely seems like some weird code anamoly.

Eric

colkai
06-15-2006, 03:04 AM
Guys,
Please keep this going with any other observations you may get.
The gem about level 0 vs level 1 is very interesting.
The only thing I could add to that, based on my exporting test rigs to Motionbuilder is that Level 0 is the pure cage (quads) whilst level 1 is the cage effectively tripled, which suggests it is the switch from quads to tri's that speeds matters up for LW9. I wonder, if you had a stand in model non sub-D which was just the tri-d cage, would it be as fast, or faster, than level 0 in 9?

animotion
06-15-2006, 11:54 PM
Wow rc8 is twice as slow as 8.5.
I'm getting 32 fps in 8.5 and 14 fps in rc8 at subpatch level 0.

Level one is also slower in rc8.

hrgiger
06-16-2006, 03:37 AM
Animotion, I got even worse results with LW9 (By the way, I'm not sure the NDA allows you to specify a build specific even though we are allowed to talk about 9-I could be wrong). However, once I unsubpatched my model, I got close to similar results in feedback. Is your model subpatched in modeler?

scott tygett
07-04-2006, 02:51 AM
I have a question which may belong in its own thread:

If I recall the news release correctly, OpenGL had been made avaliable for translation only, and I'm pretty sure it excluded morphing.

But morphing with LightWave is translational, so is somebody going to come up with an LScript or something so that we can all morph in OpenGL? I should add that I'm not that clear on what OpenGL is about, though I suspect it goes straight into the chip and so is much faster.

Take all of the object points and make a null out of each, and then make the morphing into position changes, then, er, uh, something about the object polygons attached to the vertices being copied, yeah, that sounds do-able, like cloning, oh, but cloning isn't OpenGL, okay, so the first step is some sort of Worley plug-in proxy-maker, yeah, ... hmm?

Jarno
07-04-2006, 03:10 AM
Morphing is shown just fine with OpenGL, but it is not as accelerated as meshes which do not change shape. That's because morphing changes the shape of the mesh, and therefore has to be recreated in OpenGL (and reshaped in LW as well) every frame.

---JvdL---