@racoonart: see attached.
4min 15sec. 6-core i7-5930K CPU @ 3.50GHz

Hi my name is Harry this is the second year I bought Houdini Indie.
I like Houdini and I respect so much SideFx, but I'm a freelancer and to be honesty to use mantra day by day is not good to make money.
Houdini with the procedural think is wonderful but in the end time is money…I read this thread and in the end I found this comparative video.
Maybe it not just etc but now I'll use Houdini in my day work because now I know how I'll delivery my work on the right time.


If the next year Houdini have a new mantra to use with Cuda or OpenCl it will be good.
But for me with only one processor in my machine to use Mantra is impossible.

@RobotHeadArt: I rather avoid comparing to redshift. I'm using the demo and it is reaaaally fast. I also did some comparisions in max recently, with a complicated interior scene and it performed very very well (careful, facebook link: Redshift vs Vray/Corona [www.facebook.com]. The thing is that you need a gpu renderfarm, which is not going to happen for us any time soon, so I tend to stay with cpu vs. cpu renderers. However, I have to admit that redshift is closer to a cpu renderer than any other gpu renderer. Others are not remotely as powerful as redshift is (out of core, irradiance cache, etc).

@Konstantin Magnus: I'm not sure what you mean, if I add a box and measure the height, it looks fine (see below). Is there some “scene unit” setting I don't know about?
Either way, incorrect scale is completely unacceptable in a production scene but technically this shouldn't matter in my case. All the lights etc should behave the same way.
Yes, giving all the walls thickness helps, but that's missing the point. The scene works fine in other renderers (like vray and corona), and I even cheated here in houdini (I expanded the ceiling to make it actually intersect, it didn't in max). So ideally it should work with Mantra too.

@cpb: Thanks for the effort! Unfortunately you completely miss the point here. Faking it old school with placing bounce lights is really not how you should have to work (it can make sense if you want to create a specific look, but not for trying to “make it look ok-ish in reasonable render time”). If I did the same thing in Vray I would definitely be below 1 min rendertime.
However, I find your scene interesting, there are some thing you did there that I need to have a closer look at. I assume you exported the scene as a .bgeo and reimported it?

@harryabreu: I don't think that video is giving a realistic comparison. I read some comments under the video which seem to agree . The purpose of this thread here is to provide scenes and tests to proof that there is indeed room for improvement. The test in the video above doesn't help with that (and is very biased imho).

racoonart
If I did the same thing in Vray I would definitely be below 1 min rendertime.

yep sorry, had the light samples up too high to check micropoly.
attached scene is 35sec with raytrace.

converting to bgeo is just cleaner.

Hi everybody,
so can we conclude that Mantra currently does not excel at GI or indirect lighting?
I think so.

But cpb showed that you can “fake” it quite nicely in Houdini with Mantra. The use of the “attenuation ramp multiplier” is quite clever and i have not seen this feature in any renderer that i used before.
I deleted all the fake bounce lights and used a GI light instead. It works but i am at 10minutes plus 3-4 for the photon map again.

HDR lights is something else where i think Houdini with mantra is great. I build my own HDR studio light setup. I can even render HDR 360ies in houdini with mantra (no need for hdr studio software). And i can finetune the hdrs in cops (no need to go to photoshop to desaturate or blur…

I could go on and on like this…

So obviously i would love for Mantra to be faster. And if i had to do classic interieur archviz i might use vray or if i had to render cars i might use deltagen or vred. All renderes have their stronger and weaker sides. And its not speed only that makes me choose a renderer. Its also scene setup time or features. And especially to all the indie users. SideFX is treating us so great! 200$ per year compared to 600$ for a year licence of Arnold.

For what i am doing currently, Mantra in Houdini is just the perfect combination. Its fast enough for me.


greetings

Olaf

@Olaf Finkbeiner: cpb's solution is clever but completely unacceptable as a “workaround” for a modern renderer.
The problem is not just speed. That would be bad enough but you cannot reach the same quality as with other renderers. The curtains in the Interior picture still look very bad and there will be light leaks from interpolated photons on smaller details.
I'm not arguing that SideFx is giving us great value for little money, especially with the Indie license - but that doesn't mean we should stop here. Mantra is, in my opinion, one of the weak parts of Houdini and I don't think anyone would disagree that improving it would be much appreciated. Again, this thread is here to help them identify problems, not to just rant without solid evidence.

Here are the next examples. Typical Houdini work, a standard pyro explosion from the shelf (with some minor adjustments). This time I didn't try to disable all the approximations and caches in vray to match Mantra because that just wouldn't make sense. They are here to make rendering faster and as long as it looks as good as without I don't think you should disable them - it's all about the result anyways. I first did the Houdini sim and render and then tried to match it in Vray. The defaults in Vray are way lower and render in a few seconds. I increased several settings to something I would call “usable for production”. Takes around a minute, wheres in Mantra it takes a lot longer and is still not good in my opinion. Settings in Mantra are mostly default, just Volume limit set to 1. The lower smoke twirls are very noisy. If I increase stochastic samples or pixel samples, the render time will go up a lot.
Scene without the cache is attached. Just simulate to frame 45, doesn't take long.
The next scene will be a lower density volume with a bright light close-by and GI enabled. That's something that is practically impossible to render in Mantra in reasonable time (coming back to the “it's not just speed” argument). (edit: actually, I seem to be incapable of reproducing the issue I had a few days ago, may have been an error on my side. My tests render well with a reasonable color limit. Unfortunately, I don't have the original scene in question anymore. So, forget what I said )

(I've been eyeballing the emission, gradient and density values in Vray, so there are some visual differences)

Hey Racoonart,
Neil Dickson (from SideFX) asked you to do this and i am sure it is much appreciated. Thx also from me!


regards

Olaf

Olaf Finkbeiner
I deleted all the fake bounce lights and used a GI light instead. It works but i am at 10minutes plus 3-4 for the photon map again.

Hi Olaf, could you publish your scene perhaps?

racoonart
I'm not arguing that SideFx is giving us great value for little money, especially with the Indie license - but that doesn't mean we should stop here. Mantra is, in my opinion, one of the weak parts of Houdini and I don't think anyone would disagree that improving it would be much appreciated. Again, this thread is here to help them identify problems, not to just rant without solid evidence.

This bolded line discredits the entire thread. There is simply no argument against it except that SideFx cannot allocate cycles/resources to improving it.

I'm not a native speaker, so I may not be getting the essence of what you're saying. I don't see how this discredits this thread? It's just a harmless “Improvements are always welcome” and not attacking anyone or anything. I'm confused.

I hasitate of taking part in this thread, mostly for a reason aRtye pointed out (and such shitstorm happens to happen from time to time here), but just for a sake of clarification and to answer original poster which seems to have a point and keeps polite tone, I'd like to say few words if you don't mind.

If some of you is unhappy with Houdini's builtin renderer, you should do exactly the same thing you do in case of Softimage/Maya/Max/Modo/Lightwave/Cinema4d - buy a proper renderer for your needs. This should serve you better than rants. I'm not saying you should stop making constructive critique, just keep it in the right perspective, which seems to be a problem - at least from my perspective.

Why on earth you expect Houdini builtin renderer will fight against entire universe of standalone solutions from whom every single one is developed by a team of programmers probably equal to SESI's core team? Go ahead and test Mantra against renderers provided to you for free with your favorite package. This would be fair starting point to discuss through and additionally would help you to realize why Mantra is awesome enough to be used successfully by many for two and a half decades (15 years in my case).

Mantra is developed from particular perspective of VFX industry, and while techniques developed by some major renderers in other contexts seems to payoff in VFX recently, they still provide a questionable value for many of our day work. I understand this is not somethings freelance audience cares about, but just keep in mind, that techniques which make VRay particularly good for static interior scenes are most of the time no-go for paying customers of SESI. It would be great if Mantra was doing architectural renders as good as VRay, but if you find your self in a spot you can't render such shot, it's most probably your fault of choosing wrong tool for a job. Why? Because you have a plethora of renderers available doing this kind of work great - exactly like in case of Maya (and no, Arnold is not builtin renderer of Maya, because a single license isn't enough for a job Maya is marketed for, it's merely a tease) - and if SESI would decide to scarify major Mantra's functionality in expense of faster rendering, it would break its deal with core customers and forces them to leave.

Both interior scene and pyro scene should have to be rendered motion blurred, displaced, heavily textured and with changing topology to make a valid point in the discussion. Even then previous comment applies: you might not have a performance of best market options, but you will have a decent working solution without +50K$ investment for a smallest render farm applicable in practice - and Houdini core market are companies having at least such farms.

Once buttom line is established (comparison to builtins), you may start thinking of an impact of programmable engine in your pipeline. Does it provide any value to you? Because if it doesn't, you should notice that programmability - the same as flexibility in general - doesn't come without a price! The more flexible renderer is, the slower it will crack common cases. This is a truth of life applicable for anything. The same goes for After Effect and Nuke, Final Cut and Avid, list goes on.

If Vray can render an animated and heavily textured scene - for whom irradiance caches or light caches aren't at any use - significantly faster than Mantra, this is a valid starting point of the discussion, specially if PRMan, 3delight or Arnold comes with the same results. Add to the bucket custom shaders, power of packed primitives, and we can start talking. If you're a freelance artist not carrying about those things, again, you should consider buying Redshift, it's out there, costs pennies, and you already did it for Maya or Max - not blaming ADSK for lack of fast light transport cached ray tracer.

From my experience, whenever I evaluate new renderer for a job (and I do it frequently for living), they are either not providing functionality customer looks for, or costs dozens of thousand of dollars and still comes with some limitations. I've recently advised Redshift for a constant-topology-light-geometry-show which seemed to suit it well, just to find out, that art director can't get the look he likes without lots of VEX code I know how to provide him in a day or two. Go figure!

I'm not saying Mantra doesn't need enhancements, this would be ridiculous for any renderer, I'm saying while it might be a slowest movie production renderer out there (which is not true btw, but let's say without supervision it doesn't play fast), it's still the most flexible (right after 3delight) and the cheapest one - specially teamed with Houdini for assembling and lighting. So with that in mind and considering any other builtin renderer can't even stand close:

Mantra is, in my opinion, one of the weak parts of Houdini

I totally disagree.

and I don't think anyone would disagree that improving it would be much appreciated.

while still supporting that specially from a freelance perspective (whom I would advice to use Redshift instead)

Szymon.

ps few words about Embree as a classical myth of CGI. Embree is fast compared to what? Really love this side effects of corpo marketing (the same was happening with OpenSubdiv btw). Any major renderer these days has already developed ray tracing core with acceleration structure at least as fast as Embree. Otherwise it's bummer with no future. The reason why PRMan or VRay, for example, has recently switched to Embree isn't mere performance, but maintenance. If Intel promises to support its development tightly synced with hardware advances this sounds like a great deal for software manufacturers forced to support high performance code on news hardware. Acceleration structure holds like 30% of mature renderer performance (unlike toy renderers which mostly shoots rays). Embree has to support all primitive types and advanced features of the mature renderer to be a good target to investigate, and even then making the mature Mantra code Embree dependent might slower it down for some time. Bottom line is, don't think that SESI team or Arnold team can't get the performance of Embree. They can do even more than that. At some point it's simply wise enough to push this efforts towards others focusing on your main expertise.

Similarly the same goes for TBB, which happens to be de facto current standard in VFX industry for multithreading architectures. There are actually much faster solutions out there. Just not supported by major hardware manufacturer - which doesn't look well for maintenance.

I wanted to say thanks to all who are contributing to this post.

I am learning much from this…especially your last post symek.

I think I have made clear what I wanted this thread to be and what I do (or rather do not) expect SideFX to do. I will therefore not repeat myself. If someone wants to provide scenes that they consider to be actual productions scenes, please go ahead. I'd like to do better but I'm just a beginner in Houdini.
I also welcome any improvements on the scenes I've uploaded. I appreciate your comment symek, I think you're right about a lot of things, including me having a different perspective (more of a freelancer than a big studio). I have a feeling that we reached a point in the thread where things usually tend to heat up and get ugly pretty quickly, so I'll rather stop right here and not participate in the discussion.

hi racoonart, I think you've made a good point, afaik SESI is listening carefully, and I'm pretty sure we will see some changes soon (note that Mantra was first production path tracer commercially available (I was, to be honest, pretty much against that ) - back in 2007).

I will just say that on a show I was previously part of we had 2x better render times (mantra) than current one (vray), with comparable look and complexity. The difference is one had senior artists setting up everything and using cheats wherever possible, while the current one is vray with basic setup + GI. GI in Mantra is not currently its best selling point. I'm quite sure SideFX is looking into this for upcoming release as major Mantra developments were in quite a few releases back. Also LOPs [forums.odforce.net], if its true they are Lighting operators and it moves Houdini+Mantra into Katana territory, the future is bright (GI added).

Hi nice Houdini Forum people,
please find attached the scene mod from cpb.
I deleted all the fake bounce lights and the ambient. The rest of the lights are like with cpb scene.
I also changed the curtain a bitz but that was not realy necessary to get rid of the gi blotches.

I don´t like the GI workflow (maybe i understood something wrong)
What i do:
i have to mantra render nodes one for photonmap generation one to render
the difference is the diffuse limit. when generating the photonmap i turn it up (in this case to 10)
On the light i turn auto generation on and off (not sure here but i did it anyway)

I did not check if other values would be faster i just wanted a clean photon map.
The map without prefilter gave a cleaner result.

Also i saved the photonmaps to disc and they are huge! I have a fast m2 ssd!
they are: and its 5-6 gigbytes
I could not find a way to do this without saveing the photon map. I mean the 1 diffuse limit for rendering and 10 for photonmap generation. Maybe i could do this with light quality settings?

I think this can be optimised further but i did not do it here:
http://www.sidefx.com/docs/houdini/props/mantra [www.sidefx.com]

Photon target
Houdini vm_photontarget
IFD light:photontarget
When sending photons from this light source, this is the category expression to determine which objects will receive photons.

Photon weight
Houdini vm_photonweight
IFD light:photonweight
Controls the proportion of photons that should be generated by this light relative to other lights in the scene. By default, all lights have an equal weight of 1.
Photon modifier

Houdini vm_photonmodifier
IFD object:photonmodifier
Affects the way that photons are traced through the scene. When a photon hits a surface the parameter affects it depending on the selected option. This is only used if the scene has a light (such as the GI Light) that uses photon mapping.

None
Photon is not modified, bouncing as usual (this is the default).
Pass Through
Photon passes through the surface, not being stored.
Block
Photon is stopped, ending the path trace for that photon.


Ahoi

Olaf

Hi everybody,
here a 4k rendering of the scene.

For interiors, just use PBR, Area lights to fill (unchecked Normalized area and with Physical falloff). Help path trace with photons using GI Light. I usually do not go more than 1 mil photons and sometimes change Filter to 5-10, that is all I change from default settings. Use Environment light with HDR and if you have windows opening use Portals. It is, of course slower than Corona, but not significantly. Corona has better GI so it gives more pleasant results, but that is because it is optimized for arch vis work, it can not do half of the stuff Mantra can.

This image was rendered in 12 min, with default sampling setting and PixelSample at 4x4, Diffuse limit 3 and Color limit 5. On laptop with i7 4860HQ. Photon count was just 10000.

Hi @SreckoM, thank you for the tips!

I worked from @cpb's file and removed all the other lights except for the Sun.
Then I tried to somewhat follow your suggestions:

1. Created a new Mantra ROP
   * Set to PBR
   * Override: Half-res
   * Pixel Samples: 4x4
   * Diffuse Limit: 3
   * Color Limit: 5
   
2. I skipped creating an Area Light for now.
   * I wanted to see how the outside light goes in the room.
   
3. Added a GI Light
   * Photon Count: 10,000
   * Filter Samples: 8
   
4. Created an Env Light
   * I did not use and HDR for now.
   * I got the light color and intensity values from the lightcolor shader.
   * Deleted env Object (because I'm already using an Env Light)
   
5. Added Portal Objects
   * I used this for the enironment light.
   * It was cool seeing the effect of the portal on how it really helped focus the lighting and lower the noise.
   

Here were the stages of lighting, and the attached images below:

1. Sun
   
2. Sun + Env Light
   
3. Sun + Env Light + GI Light
   
4. Sun + Env Light + GI Light + Portal
   

So far its looking good, only two issues I would like to address.

1. Noise:
   In playing around with it, I was only able to fix the noise by basically increasing Ray Samples.
   Are there any other parameters I could set to lower noise?
   
2. Area lights:
   Should I have added some Area Lights inside to help with the noise? How did you get such a smooth looking render @SreckoM, I would love to see you file if possible
   
3. Curtains:
   The odd looking indirect lighting from the curtains seems to be coming from the low and filtered Photon Count. I could increase it. But I was just wondering if there are other factors that is affecting this?
   

By the way, great and informative thread!

Regards,
Jeff

Hi Galagast,
my suggestion
set the Quality of the GI Light higher. Maybe even to 10.
reduce the diffuse limit to: 1
more photons would also not add much time.

regards

Olaf