PDA

View Full Version : Dynamite Discussions



Mattoo
05-18-2007, 07:18 PM
Does anyone know of any discussion groups based around the plugin Dynamite?

http://www.cantarcan.com/v11/html/main.html

I know there's a few threads here, but it's not what I'd call focussed discussion.

The manual is somewhat lite and so far I've spent days getting good results but am very aware I could spend three times as long before I get what I'm actually after.
Was just wondering if there was some more info out there, I'm more than happy to share my notes.

Many thanks in advance.

ece3k
05-30-2007, 10:40 AM
I have been trying to get info myself...even a basic "how-to" on dynamite. I have been getting some crashing during renders with it (using demo version in 9.2). Did you find anything?

Mattoo
05-30-2007, 05:25 PM
I guess there isn't much out there.

I've mostly been using the fluid side of it. I haven't had crashing at all, in renders or otherwise. The only problem I've had is trying to get high resolution simulations to calculate. It just gives up above a certain point.

ece3k
05-30-2007, 06:02 PM
I just want to make a simple torch to burn in a static location and can't figure out the steps. Can you go through what you do so that I can see what I am missing that may be leading to the crashing? Here is what I am doing now...making a null (torch_null) and a small sphere named "torch_emitter"...I use the dynamite draw plug in on the null in the object properties panel...changes it into a box, I add the dynamite plug in on the pixel process box, double click it and get the panel. From there I add the null, select the ball (all under the fluiids tab), give it some surface and particle direction info, hit calculate and it does (open GL looks like smoke...) then I hit f9 to test render and crash...it hangs at 25% of the scene. Any ideas?
Thanks,
Erik

Mattoo
05-30-2007, 06:36 PM
No not really, it sounds like a very simple setup. Could you post me the scene?

ece3k
05-31-2007, 08:26 AM
I'll send it with no textures because of size issues. It is posted here:
www.people.virginia.edu/ece3k

It is in a folder called BattleBridge. I want to make flaming torches on the bridge (you can see where they go.)
Thanks,
Erik

Mattoo
05-31-2007, 08:35 AM
I'll send it with no textures because of size issues. It is posted here:
www.people.virginia.edu/ece3k

It is in a folder called BattleBridge. I want to make flaming torches on the bridge (you can see where they go.)
Thanks,
Erik

The link doesn't appear to work. I get a 404 error.

ece3k
05-31-2007, 09:01 AM
its:
www.people.virginia.edu/~ece3k

Mattoo
05-31-2007, 09:54 AM
The scene appears to be using hypervoxels. Have you tried turning that off? Perhaps Hypervoxels and Dynamite are conflicting?
Either way the scene appears to run just fine with Dynamite on my machine.

Some stuff didn't load through, so again maybe that is conflicting in your scene. I don't have RealFlow...so that didn't load for instance.

I didn't bother messing with the shader much, just chucked on some quick settings to test.

sp.lwo is the only object I added to the scene - it's just the simple ball object that comes with the Dynamite demo scenes.

ece3k
05-31-2007, 12:52 PM
Yeah,
I changed it to hypervoxels after giving up (temporarily) on dynamite. I had added a simple sphere as the sp.lwo object just like in ball....but couldn't get it.
I'll try your return scene.
Thanks,
Erik

Phil
06-03-2007, 06:43 AM
I've had Dynamite generate cache files over 12 GB in size for some simulations. That's usually an indication that you didn't optimise the structure. What you can do is stragetically place small volumes in the larger surrounding volume to get the localised details you are after. This would be easier if you could use a gradient to drive the cell size, based on distance from an object/camera, but Dynamite doesn't allow for this at the moment.

I've generally worked around this by staging 4 or 5 separate volumes, with increasing cell size as you go further from the source. You can then calculate to separate cache files and save a lot of pain.

A big issue with Dynamite is that using the LW textures tab puts really hard edges on your volume. That sucks, and could potentially be a reason why HyperVoxels are never fluffy enough. I've therefore never used that tab.

It's also worth considering turning off your OpenGL displays for each volume before calculating it. This has a big impact on the calculation time and couldn't hurt stability. You'll also want to check out the other options in the config dialog.

Finally, Dynamite really can struggle when you drive it with a PFX effect. There are issues in the raymarcher that surface on seemingly random frames where the render really starts to crawl, with render times of those frames being 10 times those on the surrounding frames.

Threading can also be an issue, particularly if you have raytraced lights affecting your volume. Only the first thread will render cleanly...the others will have corruption which is random and impossible to avoid. Duplicate your key lights that are affecting the volume, turn off raytraced shadows on them and exclude them from affecting anything in the scene. Use these to light your volume and you'll see a marked improvement in performance, and no threading corruption.

The threading issue is known to Can and he's apparently in contact with NewTek about finding a fix for it.

Mattoo
06-03-2007, 12:27 PM
There's some great pointers there Phill. So far I haven't hit a single one. But I've only used it for the fluids function so far, it sounds like a lot of the problems you've had are particle related. Would you say that was the case?

It's actually encouraging to hear it will let you generate large cache files. I had feared I was pushing my luck with the high resolution simulation, but it bombed at the 400mb mark, which isn't even close. It might give it another go if 12Gb is possible. And sadly my scene can't be broken up into smaller volumes. It's actually a fairly localised effect I want, but I just want a lot of detail in the fluid. I understand I could use a 3D texture for that, but I quite fancied the fidelity of genuine fluid detail.

Do you still hear from Can? Do you know if he's still on the case? It's been 11 months since the last release. :(

Phil
06-04-2007, 01:58 AM
I here from him periodically, but he does not always reply to mail (from anyone), so it can be a difficult exercise communicating with him. If I knew his nationality, I would have a crack at emailing him in his native language :D Anyway, he's spent a lot of time trying to debug that threading issue, for example. The fix apparently needs SDK assistance, so we'll see if NewTek and Can are able to work something out.

Contrary to his web site, he has a stable Mac version (which I received after a couple of emails requesting it, but it's not tremendously useful due to the general performance issues of LW under Rosetta). Happily, he is planning a UB version as well :)

Sarford
06-04-2007, 04:21 AM
Nice to hear this Phil. I had so much as given up on this plug-in since his website hasn't been updated since the introduction. I also had send him an e-mail wich didn't get any answer. And an UB version is good news eventhough I don't have an Intel mac yet.

Mattoo
06-04-2007, 04:35 AM
I only had need to speak to him once, and that was back in the pre-release days, I was asking for World Coordinates for textures on particle based volumes and it is indeed in there now, although I'm sure I was not the only one in need of that ;)

I guess I'm just surprised there hasn't been any incremental releases.

I think he's from Turkey - just going off the name though.

Phil
06-04-2007, 05:45 AM
Damn. My typos in English keep getting worse and worse. I obviously need to take a refresher course in my native language :( Bum.

My wishlist remains :

- split cache files (i.e. split after every GB to keep things portable and less painful to chuck around the network)
- arbitrary multiple light support (this is also true for HV3....I get rather irritated by the '1,2 or all' limitation that we've had for years)
- calculation using multiple threads
- gradient support for subdividing the containing volume
- surface staining, such as smoke damage, via a shader and intersection handling. That would be really cute and allow for all kinds of interesting effects.

He had suggested that the lighting request would be in the next update, but that's not surfaced yet :/

Mattoo
06-05-2007, 01:07 AM
I asked a programmer I worked with once about that '1,2 or all' lighting and he suggested there might be a rendering optimisation with 2 lights or less and I'd find that using 3 lights would not be a linear speed impact, but far worse.
I've never got around to testing it, I never think of it when I've got free time.

Surface staining would be pretty sweet.

The biggest improvement for me would be if the Simulation Speed wouldn't effect the actual simulation but would merely be a speed function. I can't tell you how many times I've got a simulation just right, but it's a bit too fast, so I slow it down via Simulation Speed and the whole shape and dynamic changes the next time I run it.