I answer my own question .

On pg. 6 of the HQueue Cloud guide:

By default, it is not possible to create a custom Houdini setup using your own HDK plugins and a specific Houdini version. If you are an Annual Upgrade Plan customers and you are interested in running your own instance using a custom Houdini setup, then please contact your account manager or email us at support@sidefx.com to discuss what is possible .

So it seems that it's possible.. just a bit more work.

hi all..

Just curious.. if i have custom HDK nodes referenced by my render or fluid simulation is it possible to use HQueue?

Assuming it is, suppose those custom nodes make use of CUDA or OpenCL?

thanks!

hi all..

If i have code which runs well standalone with CUDA, and i want to run it at comparable speeds from within Houdini, are there any reasons why i might want to consider running it with VEX if porting is less important than raw speed? It seems like a straightforward question, but i'm wondering whether there are any hidden costs to data marshalling and the like which might favor VEX.

I'm being lazy i guess.. on the off-chance that maybe there is something that isn't obvious.

thanks

Thanks for the heads up skk.

From voplib.h i get (after subbing in a macro and simplifying a bit):

float
vop_perlinNoiseVF(vector pos; int turb; float amp, rough, atten)
{
float nval = 0;
for (int i = 0; i < turb; i++, pos *= 2.0, amp *= rough)
nval += amp * (float(noise(pos)));
nval = (float(pow(nval, atten)));
return nval;
}

so, at least in this case, the roughness determines the amplitude for each octave. The attenuation is an exponent applied to the summed result. I'm not sure how the initial frequency of the noise is determined, but it's likely that the pos *= 2.0 is doubling it for each band.

I've found these to be helpful too:

- Ken Perlin's “Making Noise” 1999. Quite inspiring. Turbulence is described as the result of summing the absolute amplitude of “fractal” bands i.e.
noise(p) = 1/2( |noise(2p)| ) + 1/4( |noise(4p)| ) + …
which is what vop_perlinNoiseVF(turb >= 2, rough = 1/2) is doing, with the additional attenuation applied.

- Eberly et al “Texturing and Modelling: a Procedural Approach” 2002. Still seems to be one of the best applied books out there.

- Lagae et al “State of the Art in Procedural Noise Functions” 2010. Quite interesting.

I'm sure this is old hat for most, but hope it helps anyone else dipping into Noise..

The Houdini help for Gas Up Res defines turbulence as the number of noise octaves.

From the Houdini help for Fog, roughness is used to multiply the noise frequencies in each octave. A turbulence of 3 with roughness of 0.5 results in the following octaves:

octave frequency
1 base
2 base * 0.5
3 base * 0.25

So roughness is equivalent to lacunarity, with an assumption that the bands are true octaves (frequencies are related by a power of 2). The Gas Up Res assumes that incrementing turbulence doubles the resolution, which is consistent.

Attenuation is TBD.. it seems likely that it's related to gain. Rather dry stuff, but i find that it helps to understand what is going on .

Thanks Solitude. That is the method i used with the Turbulence VOP.

I suspect some artists are simply tweaking until satisfied with the look, but if there were standard definitions of these terms relative to the basic parameters then it would make it easier/quicker to reproduce or improve a noise effect developed in another package.

Maybe i'm missing an explanation somewhere in the help system or a relevant paper or even a Rosetta Stone to map effects across packages :roll: .

BTW i can get a form of undulation by key-framing parameters in the various noise VOPs, and the Displace Along Normal VOP help states that the displacement is anti-aliased.

hi all..

I'm trying to determine how to use the Noise VOPs to simulate a 3D fluid surface which undulates over time.

As i understand it, the 3D Perlin noise function takes 3 floats as input and outputs a float based on the following settings:
- a first “octave” or “band” with parameters:
- amplitude: the maximum absolute value the noise can attain
- frequency: the number of cycles per unit length
- optionally additional octaves, each derived from its predecessor by:
- lacunarity or persistence: frequency multiplier between octaves
- gain: amplitude multiplier between octaves
I think this means that i pipe the position into, say, the Turbulence VOP and output the noise through a Displace Along Normal VOP.

The Turbulence VOP, with type set to 3D Perlin Noise, has additional parameters which are not described in the VOP help:
- offset: ?
- roughness: ?
- attenuation: ?
- turbulence: ?
If someone can explain what these do, or point me to a reference it would be appreciated.

The fluid surface should undulate over time. The Perlin Noise wikipedia entry states that there is a 4D variant which takes time as a parameter. I notice that Worley and Voronoi generate 4D noise, but it doesn't look like the 4th dimension is time. Is there a good way to get the undulation?

The Turbulence VOP doesn't seem to anti-alias the noise. If not, is there a good way to handle this?

There are a variety of other VOPs available, but i've focussed on Perlin because it seems to be suitable for a variety of tasks. A quick look on the www turns up Wavelet noise and newer variants of Perlin noise. Is there another noise generator that might be better suited?

thanks!

The DMM technology is a variant of tetrahedral FEM techniques optimized for speed. It is perhaps best known for its use in Lucas Arts for Star Wars: The Force Unleashed. The ACM paper “Real-time deformation and fracture in a game environment” is the best description i know of the technology. At the time i believe there were some issues with the sudden spike in polygon count whenever the player felt particularly destructive (which was often).

I like the example 8). Thanks!

Thanks johner. I think i understand the reasoning behind the decisions that you and Jeff have made and for what it's worth given my limited knowledge of Houdini and the problems it solves they seem valid to me. That said, I'm still not sure that Voronoi is a great fit for wood splinters, but perhaps it's nothing that a little motion blur/smoke/sawdust/increased poly count wouldn't cure .

AMD's Open Physics Initiative combines DMM2, Bullet, and OpenCL so it will be interesting to see how it fares. It seems like a promising approach for non-hero fragmentation.

Thanks for your replies.. they certainly help to scale the learning curve

Thanks! I will dig into this component further… it looks quite interesting.

Is it this one: http://forums.odforce.net/index.php?/topic/9119-voronoi-dynamic-location-based-fracture-wip/page__hl__johner%20voronoi__fromsearch__1 [forums.odforce.net] ? I had this flagged for further investigation, but the first post says his favorite feature is that it stays away from the Cookie SOP and volumes so i assumed it was significantly different from the Killzone approach, though they are both handling fractures. A volumetric approach seems like it might be better suited for materials with internal heterogeneity, like strong-grained wood, but if there is a simpler/better way i'm all ears .

In the Killzone 2 case study they mention a process (Waves of Destruction) where a volume is:

- an object is converted to a volume primitive
- cut up into voxel chunks using a “VEX volume shader”
- the chunks are converted back to polygons
- the original object is divided by transferring chunk id attributes from the internal voxel chunks
- a Foreach Sop is used to loop through all chunks and stitch them to the corresponding bits of original geometry

Are there any examples about which might help me to understand these steps? Use of a shader for a geometry operation seems unusual to me. Handling of the UV coordinates is also of interest, though it doesn't seem to be described in the article.

thanks!

Thanks.. it seems like it might be workable..

Thanks.. fluid flow sounds like an interesting approach.

A knot is actually the core of a branch rather than a distortion of the surrounding grain and, after following up on your suggestion, i found a paper by PJ Pellicane that models wood using laminar fluid flow around an elliptical obstruction. This would work ok for my needs i think. Alternatively a simple 3D procedural function of the type commonly used for procedural textures might do, too. I think it would be quicker.

These don't fully simulate knot formation, though. Perhaps there is a way to do this with an l-system simulating the growth of a tree? Probably overkill :roll: .

hi all..

I've tried prototyping wood grain as a series of grids stacked on each other, but handling knots is a problem. I tried using a sphere to represent a knot then using a Ray SOP to deform the grid with an offset.. but it wants to warp the entire grid around the sphere and i'm treating the grain as a series of planes which deform where they intersect the sphere's volume of influence. Does anyone know of a good way to handle this? Perhaps cloth.. but i suspect it would be slower and more complex.

thanks

Thanks Anthony.. regrettably i don't have any ROPs at present, but will keep it in mind. For this particular project a top-level HDA is used to hold the parameters. So far it works ok.

hi all..

I'm sure there is an easier way to go about doing this, but currently i have several objects which can be rendered via copy (typically for preview) or point instancing. Each object has a switch statement on it to control LOD (bounding box/polyreduce/untouched).

Is there a good way to set the LOD centrally such that the switches can read it (assuming the switch approach is best)? I would prefer that the settings be specifc to the project rather than global to Houdini. Perhaps it is possible to extend the Houdini GUI somehow using a Python HDA.. an icon for each LOD level would be nice 8) .

thanks!

The angle was set using a copy stamp parameter, which is not available when point instancing is used :roll:

hi..

I'm on Ubuntu 9.10 64-bit running Houdini 10.0.554. I am rendering an l-system using a stamp function to fetch a variable for the angle value. The variable affects the geometry in the OpenGL view but not in the Mantra render. Probably my ignorance for missing a setting somewhere.. does anyone have any ideas?

thanks!