Working Faster? How Do You Guys Do It?

[Atom]

Hi All,

 

I have been using Houdini for almost a year now and I still do not have anything worth showing.

 

This is my biggest setback with using Houdini. It is so damn slow. I can easily out think the computer and reduce it to a crawl with a simulation and still not get anything worthy of output for a daily. And then there is Mantra rendering which is very slow as well.

 

How do you guys do it?

 

I have multiple systems AMD 4 core, AMD 6 core, Intel 8 core with ram ranging from 10Gb to 24GB across the machines. I have an nVidia 660GTX and a 730GT so I can leverage some OpenCL acceleration but it is hard for me to detect any significant performance benefit from OpenCL. Sure I can use monitors to detect that my CPU load has gone down and my GPU load has gone up but the overall performance is not really faster.

 

So is it just my hardware? I do have access to fast intel zeon computers, at work, but when I run Apprentice on them they don't seem to be any faster then my home machines as far making Houdini work better.

 

Sometimes I read/watch people's work, who produce amazing results with Houdini, who say..."Oh yeah, I did that on my laptop in the hotel room while waiting to go to dinner."

 

How is that possible?

 

I feel like, after a year, I have a good grasp of most general systems but noodling up clever node arrangements still seems like high science/art.

 

I guess I don't know what is typically expected from a Houdini User/Master. In a typical day working at a studio are you expected to produce results in an hour, a day, a week, a month? I am currently working in an advertising agency environment which can be fast paced at times. Houdini does not seem to be a good fit for that environment.

 

 

[Skybar]

It is very different working on a single workstation, or at a studio where you have a farm at your disposal. Simulations and renders do take time, that is true for everything not just Houdini. 

[3__]

how to do it: Fear of deadlines... real ones.

I'd suggest fully committing to using the software for your ads, resist the temptation to fall back to you normally use on the job.

You'll end up cheating/hacking your way through just like with any other program. After a while you'll be throwing stuff together in no-time.

[iamyog]

Invest some time learning VEX

while it may appears longer to type rather than just laying down nodes, it is usually quicker for everything that Houdini will cook (and H do cook a lot) It will be very noticeable for anything heavy in your scene.

 

Simulations and renders though... read David above, or master Mantra and each solver to know where you can cut time off.

 

A test shown on vimeo is also completely different than any show you will work on : No client/agency changing mind, no notes to address, "no need" to keep a tidy hip file for a handover, etc etc.

Production environment and personal environment are not to be compared based only on a video and its description. It could also be that you are watching Houdini gourous' channels and speed is also what they make them gourou

[Annon]

how to do it: Fear of deadlines... real ones.

 

About right

[Diego A Grimaldi]

I've been using H for about 2 years now and I agree with what Anthony said: "A test shown on vimeo is also completely different than any show you will work on : No client/agency changing mind, no notes to address, "no need" to keep a tidy hip file for a handover, etc etc."
Plus all productions are very different. I used Houdini on both film and commercial and the amount of testing and time that you have on features is so different than the time you get to spend on commercials. So you have to think differently and work differently. At times you'll be able to show more of your personal works while working on a kickass short film that you worked on for a couple of months than if you work for a year on a big feature film.

[Atom]

Thanks everybody!

[kleer001]

The #1 thing that improves my workflow speed is a pile of good preferences. This allows me to get up and running faster than from scratch. I have dozens of presets all around SOP-land. There's lots of options for setting/saving preferences in the Gear menu for any node.

 

The #2 is working simpler. But that's not always an option. Fewer nodes (and notes and comments) can make it easier to understand when you come back to a network in a few years.

 

I'm also faster when working on lower rez, smaller slice, or shorter time versions of my final effect. Up rezzing volumes is probblemattic, using a small part of a sim is a good balance of speed/detail but you can run into problems with other parts of the sim, and sometimes you can't tell a lot of what's going to happen in 3 frames, but in the end there's always trade offs.

 

The lovely Houdini library "qLib" has some great processes speeder-uppers. http://qlab.github.io/qLib/

[sebkaine]

i am not a houdini expert as i only use it for less than 2 year. but i find that

-  visualising precisly your target ... know what you want from H before going inside the beast is crucial

- keep an atomic logic, you first work on the small atomic lego pieces of your fx and you polish them

- then you scale the beast by duplicating / randomising your atomic / simple atomic lego

- keep it manual when it's faster, if you have to delete face , just delete them by hand and don't build a 30 nodes vop / sop to do that.

- stay as low level as possible by knowing vex, it will help simplify your trees and code is imo more expressive than nodes ...

Edited June 12, 2015 by sebkaine

[davpe]

i would ask how long/whether are you doing 3d professionally? if your only experience is that one year learning houdini, dont worry about not being fast. i spent some 5 years self learning 3d before i even tried to make money out of it. and even then i was (from todays perspective) very very slow, very unexperienced and i didnt know much about how it actually works. it took me another maybe 2 years until I was able to do this job on a level that I dared to call professional. then it took me another 3 or 4 years before I noticed that I am actually working faster and more efficient than some other guys. and it will take me rest of my career learning to do things better and faster and it will never be enough. cheers ;-) D.

[willow wafflebeard]

-don't forget to work on camera perspective, if your not in games. so much can be cull out of the camera view
-don't preview your object at full geometry when you don't need to, you know H's viewport is funky.
-think what you need in comp
-previz it first
-show it to people when your stuck/content

[Atom]

@davep: I have been working professionally in animation for about 9 years. Mainly motion graphics. So I have a strong foundation in composition and movement. I used Blender and C4D prior to acquiring a Houdini Indie license. So I kind of know what I want to achieve. The procedural approach is nice, however. Previously I spent a lot of time writing Blender python code to achieve results that I can now get out of Houdini's "canned" nodes. Even though I can write code I prefer not to. I am no stranger to scripting or expressions. But the brute force, calculate my effect now, process in Houdini is what bogs me down. That and rendering. I am always looking for a faster way to render. In the motion graphics business, this is king. The quicker I produce output the more time I have to work on animation. In Houdini this drops me to the floor. Where as I can accept an overnight render for 10 seconds of coverage, if I make a mistake anywhere in the process I have lost an entire day.

 

I wish Houdini had a faster render system, even if it was old school scanline or something. Like Blender's Internal Render engine. That engine is dang fast. Sometimes all I need is some geometry in motion with lighting and shadow. Physically correct does not matter for art. I have been exploring Houdini's OpenGL rop. This engine has the speed I need but the material/lighting output is limited. Couple that with fact the GLSL shaders are not currently supported in H14 and it is a show stopper. 

 

I continually explore Mantra trying to "bottom" it out for speed yet maintain quality but it is still not fast enough.

Edited June 13, 2015 by Atom

[sebkaine]

i don't think mantra is slow, if you compare it with most commercial competitors, you have a huge versatility.

i agree that the shading workflow is painful and not that impressive.

but the problem is more related to a necessity of coding your own shading workflow ... Mantra in itself really shine !

 

the combo C4D / Thinking Particle / Arnold / After Effect really shine when you work under very very short deadlines you should give it a try.

because houdini in a commercial environment with very short deadlines, is kinda hardcore ... this is what i do now but i agree that sometimes you would love to have some basic OTL prebuild.

[Guest tar]

There is a push out there to use Unity for real-time rendering, but, there are quite a few of us who are pushing for greater OpenGL Rop functionality in Houdini too.

 

In terms of simulation performance, as you are coming from mo-graphics I'm not too sure you have had to deal with the amount of data that Houdini can generate - you probably need to brush up on what is happing with PCI buses, bits, data rates etc to understand where the bottlenecks in the computer are.

 

One elementary performance boost is to use Houdini on Windows or Linux before OsX. Preferable Linux is you can. 

[Guest mantragora]

I always come here and post my problem + I throw into topic couple sounding wise words, to make it look that I know WTF I'm asking about. Then I wait till Marty, Anim, Symek, Jason or rest of gurus come and solve it for me, by falling into trap of proving how good they know Houdini.

 

Works every time.

 

[davpe]

well, with 9 yrs experience you should have pretty much all you need. i have never worked in motion graphics field so i can understand something else than you under the terms like "working fast". my focus is mostly for film rendering so i can appreciate that houdini can handle very heavy rendering scenarios. anyway, i started learning houdini at home myself, but when i started my first real houdini project in the company, i could not do a thing. and after one month on this project i learned much more than at 6 months learning it at home (but it was a very tough month . As some of the other guys at the begining of this thread said, its very different to work at home or at the company environment. 

 

regarding rendering, i dont think Mantra is a slow renderer (at least not slower than comparable ones). in fact i think it is maybe the best and most versatile renderer i have ever been using and is able to render things that would be hardly possible otherwise. however, as i said, motion graphic gesigner may have very different needs than artist rendering for feature film where accuracy is usually the main goal and render time is a secondary concern. anyway i think you can achieve very fast renders using mantra and micropolygon or raytrace engine. houdini is very robust system and maybe that prevents you from using is efficiently. have you seen the cmi vfx`s tutorial on motion graphics? as far as i remember rendering is addressed there as well and maybe it could help you find better workflows.

[amm]

 

 

I wish Houdini had a faster render system, even if it was old school scanline or something. Like Blender's Internal Render engine. That engine is dang fast. Sometimes all I need is some geometry in motion with lighting and shadow. Physically correct does not matter for art. I have been exploring Houdini's OpenGL rop. This engine has the speed I need but the material/lighting output is limited. Couple that with fact the GLSL shaders are not currently supported in H14 and it is a show stopper. 

 

I continually explore Mantra trying to "bottom" it out for speed yet maintain quality but it is still not fast enough.

 I'm also mainly on motion graphics. Workhorse is still Softimage ICE and 3delight. Found 3delight, always in REYES mode for such tasks, as sweet spot between speed and reality. While is nowhere close to speed of old scanline, when it comes to lot of shadow maps, it gives me a 'real' 3d motion blur at no cost (I guess you're using built in 2d motion blur with Blender Internal or C4d 'old' renderer). Plus, beautiful sampling, no big deal to get toon ink of 0.2 pixels or so, without any flicker. I'm getting literally perfect texture sampling too - which always was ... so so... with old scanline renderers. Plus, have some additional trickery in bag, like faking the subsurface or even shadows sometimes, with ray casting in ICE (should be perfectly doable in H), or by long shadow map bias over flat shading, so on.

That said, I think it definitively worth to explore Mantra micro poly (REYES) rendering. Don't believe it can defeat the 3Delight, but should be close. Didn't tried Mantra micro poly, yet, my Houdini and Mantra is still for 'luxury' tasks.

Other than that, I'm always trying to explain to 'relevant' people, how rendering is time consuming process. In almost 15 years, whatever I did to speed up rendering, always was taken as granted, still not enough fast. So, maybe it's better to negotiate a bit, just to get a nicer midpoint.

[Atom]

I know what you mean by people (clients, account managers, editors, directors, etc..) taking for granted that rendering is insignificant and just happens. "But we only want a text change. I know, but that requires re-rendering."

 

Mantra does have several modes, Micro Polygon, Raytrace and PBR but they all seems to take the same amount of time. They all look the same too. If I setup a scene and then switch between the modes I might get a different noise pattern but it takes just as long.

 

Is there a way/setting that will convert the Micropolygon to a faster render mode?

[iamyog]

It's probably the wrong way to approach mantra thinking a simple rendering engine switch will speed renders up.
I'm far from a mantra expert but a rendering engine is first a choice you make because of what you have to render. Then you tune it according to what you've got in your scene to achieve a quick and "error free" render.
You won't build the same shader whether you are rendering with micropolygon or PBR.

The same idea is explained on this page : http://www.sidefx.com/docs/houdini14.0/render/noise
One would say "I can I remove noise in my render". This would be the wrong approach.

The good one would be to break everything down (check direct/indirect lighting and samples) to identify what's causing the problem, then you tweak the corresponding value for that particular faulty thing only.
If you blindly push the global samples because "more is quality" you'll end up with insane render times and useless oversampled areas.

Every render engine requires you to go through the trial and error steps to understand what's efficient and what needs to be avoided.
Mantra probably has a stiffer learning curve than some others and it cannot be summarize as a "which checkbox should I disable to speed everything up ?"
Let's trust sidefx people to be kind enough to turn the checkbox off by default if it has such a negative impact on rendertimes.

Overall, Mantra is one of the numerous pandora box that needs to be opened in Houdini, it is indeed probably not the smallest.

 

TL;DR: Newcomers hate it, long time users love it.

[sebkaine]

if you use micropolygon render and occlusion light tinted by an environment you will have very fast renders.

 

what is expensive is when you pathtrace with indirect diffuse.

you have the ability to write photon map to compute your indirect diffuse and speeds things.

 

but whatever you do if you pick a pathtracer you have set your priority on quality not on speed, RIS / Clarisse / Arnold or Vray in Brute Force mode will all be slow.

if you need very fast renders in small shops with no farm redshift might be an interesting option.