why mantra renderer is so so so slow?

[Guest Swann]

You all propably know this topic from SESI forum. Now after this announcement I would like to ask the same.

Additionaly:

Why Houdini is so slow ? Why it doesn't make better use of multicore processors ? They are available for masses on the market from 2005, that's six years ago.

Why it doesn't have proper edge group support ? Why you can't create geometry straight from VEX ? Softimage ICE does have all those.

Why when I'm going to page in help like this or this it's blank ? If it doesn't contain any information why it even exists ?

Why after so many years Houdini still lacks good documentation. Going thru Softimage SDK documentation is painless, Houdini HDK on the other hand... OMG ! It's not that Softimage SDK is simpler it's all about how their documentation is written.

Why program that costs 7000$ doesn't have features that cheeper or free programs have ?

Why ? Why ? Why ?

The list is longer but I hope you all get what i mean and I don't have to write it all.

All I hear is that SESI is small group but isn't it like that small companies can react faster to changes than big ones ?

SESI, please, give us some hope that feature, hopefully close one, will adress all those problems that haunts Houdini users for so many years.

Please...

PS. No flame war guys. Normal talk please.

[~nature~]

My greatest wish for houdini is SESI could implement CUDA in houdini especially in the dynamic simulation or there's an option which can switch this solver into CUDA mode to speed up the calculation based on end user's needs.

Edited April 25, 2011 by ~nature~

[hopbin9]

Why Houdini is so slow?

If referring to Mantra, then I agree it is slow but none of the other packages come with a render engine that is as good. So I'd rather invest in a larger render farm then have to live with a render engine that produces poorer quality results but faster (V-Ray the exception).

Why it doesn't make better use of multicore processors ? They are available for masses on the market from 2005, that's six years ago.

I've never had a problem with Houdini using my 8 cores. Can you give an example of where it doesn't use multicore?

Why it doesn't have proper edge group support ? Why you can't create geometry straight from VEX ? Softimage ICE does have all those.

Edges in Houdini are not attributes. They don't exist only point attributes, and faces. So they would need to introduce a new type of attribute just for edges (which might be a good idea). I agree, I don't like working with edges in Houdini.

Why when I'm going to page in help like this or this it's blank ? If it doesn't contain any information why it even exists ?

Totally agree, I also find links to examples are often broken and the documentation often uses technical terms of things I've never heard of before which makes learning what the node does a mystery. They need to stop having developers write the documentation and hire some real tech writers.

Why program that costs 7000$ doesn't have features that cheeper or free programs have ?

All that matters is that Houdini has what "you" need to get the job done. If you run a studio you make a big investment in your tools, and you need to be sure you can get a return on that investment. I'm confident Houdini is the right tool in that case.

[Netvudu]

Ah...another one of these threads. I thought Houdini was relatively safe from them

Regarding the render engine, after trying most of the stuff around (still haven´t tried "yummy" Arnold), Mantra is probably the most advanced that comes with any package natively, and one of the renders with most quality around (I still hold my vote for Lightwave´s native render engine which keeps on giving me fast photorealist output after all these years, plus its Linear Workflow might be the best implementation around).

Frankly, speed might be an issue for a render engine, but it´s definitely secondary compared to quality, flexibility and robustness. Stuff like the one you are showing in those youtube videos is nothing but a nice tech demo, and far from production-proven (besides I see no photorealism there). For every time speed is an issue, there are ten times when your engine not being able to render something properly is the real issue. Mantra won´t abandon you in dire times the way many others do.

I do agree with recent documentation lacking a bit. I always thought Houdini help cards were brilliant, but lately it´s been neglected a bit, or at least it looks like. If it wasn´t because many other learning materials are available, it might get a bit dramatic, because there´s been much development, but not so many explanation and a lot of obscure nodes with mysterious but important options are starting to show around.

Also, SESI is creating very nice webinars and similar initiatives, but less video tuts which isn´t that good. They seem to understand we do have to invest in purchasing tutorials.

Finally, the argument on 7000$ apps vs free apps is way too bland. It´s simply a different business model each one with its own strong and weak points. Blender still has to prove it´s a proffesional product as opposed to a good box of tricks that you sometimes resort for a very specific situation. I still fail too see many good final products done with Blender as the main workhorse, and as you can read in your own links, the developer for that cool quick renderer, said he went in a different direction because he was too limited with Blender´s native one. Meaning if you are a Blender user right now, you have to deal with a weak render engine and hope for the best in the future...much better having the good stuff already there.

My 2 cents...

[lukeiamyourfather]

My greatest wish for houdini is SESI could implement CUDA in houdini especially in the dynamic simulation or there's an option which can switch this solver into CUDA mode to speed up the calculation based on end user's needs.

Did you say CUDA but actually meant to say OpenCL?

As for the OP, there are many things that Houdini offers that Blender (and other tools) do not. The grass is always greener...

[symek]

Big chunk of your problems, Swann, comes from a fact that you use Houdini, instead of modo, and you acctually love it. You like to waste your time recreating modeling toolset you get in modo for free, or waiting for mantra to finish render, which would take modo a fraction of this time - for a very reasonable money compared to SESI's price list. Aren't you a luxology client, lost in the woods? There is nothing wrong with using other software.

AFAIK no one with production expierience could possible say that mantra doesn't payback or can't handle its duties. As always there are cases it works better and worse.

Mantra is not the fastest renderer on the market, but it is pretty amazing blend of technology, cleverness and flexibility (which always comes with speed penalty, what people tend to forget).

For example, Mantra is currently the only programmable path tracer on the market. You can do things normally reserved for reyes engines in fully physical environment. I susspect you don't consider this as an advantage, because you simply don't need it. Many do.

[zarti]

i do not use modo , i do not have problems with hou price , i like the flexibility and control which mantra gives , i like how easily i can do network rendering and Co .

also , i wdnt be afraid from mantra if it could run faster .

.cheers

[Macha]

I can see 3 ways to make a 3d soft faster:

- By providing a supportive environment and by adding tools and features. Houdini is quite good at that. Look at rops, how great are they! Then we have daily bugfixes. We can have multiple sessions open and copy paste nodes between them. Can write our own shaders and make our own tools. We can dive into various nodes and investigate/modify.

- By applying newer and better algorithms. I think this is overlooked by some people. Nevermind multithreading and cores and all these things. Real, magnitudes-of-order-faster progress is often achieved by better algorithms. I don't know how good SESI it at doing that, I have no idea, but implementing the right thing at the right place can make things very fast.

- Multithreading, etc. From what I understand this is not a trivial matter. You can not easily rewrite something to be parallel. Some algorithms don't even benefit from it, others are positively impossible. Even in the best case you do not get an extra 100% in speed for each processor or core you add. Perhaps there is something quite fundamental about Houdini that makes this especially difficult and risky and rather than rewrite the impossible a new context will be added. Who knows.

As for the docs. I find them quite good actually. There's only a few obscure nodes missing, some of them I can guess the meaning anyway. Or not? Well, HDK, that's like flipping through a space-shuttle manual shredded with a telephone book. I feel quite stupid whenever I dare near it.

Perhaps what Swann is asking for is some detailed explanations of these issues. It's frustrating to wander about in the dark. I'm quite a rambling wanderer myself.

Edited April 26, 2011 by Macha

[ganzo]

I actually think houdini has been getting really fast. Mantra mpr and pbr are getting faster plus it is a really flexible and robust engine. Also, rigid bodies, distributed fluid sims, huge data sets for particles...you cant go wrong with this arsenal for vfx. I do recognize that modeling and animation does not come close to what you can get with modo and maya...but thats no problem....houdini is one of those packages where you can import anything and it works...contrary to other packages...all I know is that I am a happy houdini+maya user. Cant go wrong with that combination.

But documentation I agree...although it has been improved...learning houdini is quite a task...if it were not for odforce users posting their .hip files with experiments and people like peter quint putting all these video tutorials out there...a lot of people would still be struggling to climb the learning curve.

Edited April 25, 2011 by ganzo

[hopbin9]

Well, I'm completely pist that Houdini doesn't work on my iPad. I mean, for $7,000 there should be a frigging app for the iPad!!!

[~nature~]

Did you say CUDA but actually meant to say OpenCL?

No, CUDA is different from OpenCL, Industrial Light+ Magic, I believe, is the first studio to employ CUDA ( http://en.wikipedia.org/wiki/CUDA )in the film visual effects production , such as Harry Potter, Transformer, Avatar, The Last Airbender to speed up simulation(fluid+RBD) significantly.

Recently rendering engine like vray and mentalray have also implemented CUDA. I wonder if it is possible mantra could participant in the parallel computing world in the near future.

The same is true in the softimage real time simulation.

http://www.nvidia.com/object/io_1273059136418.html

Edited April 26, 2011 by ~nature~

[LaidlawFX]

This set looks like it has been taken, but to define one thing first and mix in some more answers, especially from some on that has a long wish list.

Use the right tool for the job. Houdini can do a lot, but it is not a panacea. Don't use your hammer when you need a saws-all. If you want to do UVing of Organic modeling, use UV-Layout, if you want to do 3-D painting use Modo. Most software, and especially render engines are specialized, or have specialties. If you want to use 3rd-part render engine, Houdini can easily help you out.

Beyond that;

1. Just asks if you have a specific(or even general) question, to the forums, or sidefx help, or make a request for them to fix something. The software is pretty big so create a list of the blank help pages you find and e-mail them in. These guys fix stuff daily, nobody does that in a 3-D package. It can def get frustrating when your by yourself working, I know I've learned a bunch of answers by asking my co-worker or asking on the forum that would have taken me hours to figure out, mostly those dumb little thing.

2. The documentation has some stuff lacking, but it hand down beats down most other packages. They have certainly fixed the mantra rop documentation from when I wrote up my own, and it is better worded than mine. You don't see that between versions in any other 3-D package.

3. Multi-core, Side-FX knows this and is gona make some changes in version 12, I think they've gotten the picture(though, I don't think it hurts to keep on asking). Prisms-Houdini is one of, if not the oldest 3-D software on the market, it's not like it's fresh technology with out a history for them to create it clean and new and sparkly fresh. The software has a good fundamental design, because of this, that a lot of other people keep on mirroring from scratch.

4. The cost of the software is proportional to the user market for it, Houdini is one of the lower seed groups with 3ds Max being the highest in the world, maya actually doesn't come close to 3ds Max from what I understand correctly. Remarkably for that small market share, they are in nearly every major vfx and animation house. Modo and Nuke aren't that strong. There is also several affordable purchasing plans per a project to use. I use escape everyday at work, so you can go that route. Also you get what you pay for... Though it wouldn't hurt for them to make some more tutorials, educate more users, to buy more software, to lower the price, and technically lower the amount of the money you make using the software, which I am all for. I like working normal hours, the sun is a marvelous thing. In the medium run it wouldn't be bad to see the full package at 3 grand.

5. Put down the time/money for the sidefx tutorials, cmiVFX, peter quin, the houdini books. I am 100% a new school houdini user from version 9 and there are a bunch of resource to get you around in it. Old school users didn't have anything like that. The specialized stuff does have it's problems to learn, I had to learn shading and rendering on my own, but if you are resourceful, or just ask online, you can look at a mental ray, renderman, or v-ray book and understand the terms that are happening in mantra when the document lacks, with the added benefit you learned more of the other render engines to make yourself more marketable. Most are really the same with one or two algorithm being faster than another and then being copyrighted to give the engine their edge. Otherwise Mantra would be exactly like Renderman. Most Old-school people are also pretty generous with their knowledge, since the only way they learned is from a mentor, so a majority will pass the info along. Some just use the threads more than others, and always get paid more when working than teaching, the schools need to fix that part.

[3__]

You all propably know this topic from SESI forum. Now after this announcement I would like to ask the same.

Additionaly:

Why Houdini is so slow ? Why it doesn't make better use of multicore processors ? They are available for masses on the market from 2005, that's six years ago.

...

I'm with you on most points; the software is too slow (& not just Houdini), but...

If you run into a specific problem then isolate it and post a hipfile.

There are many here who love fixing problem hipfiles (or like to try anyway) and you will usually get a few good responses, or links if the problem was previously addressed.

[Stalkerx777]

From my experience, users, which say that mantra is so so slow, actually, never had real production experience in mantra. I know it, several years ago, i thought the same. Most of dissatisfied users, have tried to render chrome/glass balls, or simple models in white room, with a little, or, in most cases, without any knowledge of mantra. They get long render times and very noisy image. And then they go to forums..... The problem is, they can't(or don't want) to understand, that mantra - is production render, aiming towards vfx production. Ability to render huge amount of geometry with dof,displacement and motion blur, and great flexibility over all - is the most valuable features.

As SYmek says: "Mantra is currently the only programmable path tracer on the market". I know one guy, he is programmer.And he sad, that it's pretty easy to write simple raytracer, based on CUDA.(like in video in 1st post). But can it be used in production? Off course no,for numerous of reasons !

Edited April 26, 2011 by Stalkerx777

[curveU]

When talking about the company strategy as being a small company, I believe every boss is working on math:

X = extra cost hiring some people (tech writer, developers, etc)
Y = extra benefit from X (more pieces software sold out, better credit in the industry, unique advantages over other competitors, etc)

if (X > Y)
  expanding the team;
else
  do not do it;

Of cause this is just a very simplified version. There are more factors affects the decision of the company owner, but one thing might be true: It's not likely that the company does everything requested by the customers.

[lukeiamyourfather]

No, CUDA is different from OpenCL, Industrial Light+ Magic, I believe, is the first studio to employ CUDA ( http://en.wikipedia.org/wiki/CUDA )in the film visual effects production , such as Harry Potter, Transformer, Avatar, The Last Airbender to speed up simulation(fluid+RBD) significantly.

Recently rendering engine like vray and mentalray have also implemented CUDA. I wonder if it is possible mantra could participant in the parallel computing world in the near future.

The same is true in the softimage real time simulation.

http://www.nvidia.com/object/io_1273059136418.html

I know what CUDA is, hence the winking. Just pointing out that there's more than horse in the race. Especially one that doesn't employ anti-competitive practices which only end up hurting the industry in the long run.

[zarti]

... mantra - is production render, ...

hi ,

what is a production render ?

[hopbin9]

hi ,

what is a production render ?

It is a term used to describe a renderer that can generate a sequence of images in a consistent way, without flickering, without antialiasing issues, without crashing, handle large volumes of data and manage memory effectively. It should also be able to render 4K sized images (if enough Ram). It should be able to produce shadows that are free of errors (i.e. holes, flickering, or ghosting). It should also be flexible enough to meet the needs of the production.

It basically means a renderer with really good reliability, consistent results and flexibility.

It has nothing to do with the features of the renderer (i.e. global illumination). For example, there are cartoon render engines that are classified as production renderers, because they are reliable enough to be used on a large scale film.

When does a render engine become classified as a "production render". It never really does. It's either marketing hype by the software developer, or it's the reputation it has in the industry. Pixar's RenderMan is the golden standard for production renderer. No other render engine has produced more seconds of 3D animation for feature film (that I know of). So you can express this as an equation, given X number of hours rendering by number of error produced by RenderMan. This becomes the ratio that other render engines need to match or surpass, and that's going to be very hard.

[LaidlawFX]

(or like to try anyway)

lol...

[exu]

From perspective of Maya user, also, mental ray, and Renderman (Prman, 3delight) user... and of course Houdini User...

Mantra it's not slow!

this assumption that mantra it's slow it's become most people that use MR and Maya or Vray and Max, didn't count your render times seriously!

I mean, they count frame times, with no AOV, without sum average time per render layer, and or passes! and of course without Motion Blur!

If they do that i pretty sure that MR it's slowest render than mantra...

Assume that you need render a realistic car, if you only count the render time in render view of maya, you don't count the translation, also if you use a FG, you don't count the number of times that you must render until flickering is gone or free of artifacts!

If you put some passes and didn't use a rasterizer (with has a lot of issues of anti-alias), your time frame will reach the roof!

Not mention that for some heavy scenes, you will need separate in layers just because the memory blows away!

The solution? render AO in a render layer, render another layer with some shader to do motion blur in post etc...

if you take the time of all those passes (i.e render layers), and sum it, you will get a really slow render....

Ok, if you use a Vray/max you probably don't need or have a render layer, but you need to wait much more time to render without flickering! or not ?

I'm sure that was a tricks and workarounds... but it's not the same? or you will go to Kaos group forum and post something like:

"why Vray it's slowest render on the market?"

So, just because those people knows the issues about those renders, and use a workaround to produce your images, this is not means that it's not exist!

why instead they just come here and said, "too slow", they don't ask "what i'm doing wrong'?

Sorry for my english...

but i'm surround by maya and max users and all the time it's happens when people starting using a new software...

Also i'm still working with maya, just because there's no many places here in brazil to only work with Houdini, i hope that it's change, soon!

PS: using AOVs in mantra and Renderman, give you a full control of your passes, including per light, per objects and so, with a minimum cost in your render times....

Not mention too that RSL and VEX also gives you a freedom to make some impossibles things easier than programming a shader in C++ like others renders like MR and Vray...

cheers!

Edited April 26, 2011 by exu