Is there any reason to switch to Karma? Is unbelievably slow

   10559   62   9
User Avatar
Member
42 posts
Joined: March 2009
Offline
I've been playing around with all the awesome new features SideFx put into 18.5. But I can't grasp one thing - why did they bother with Karma?

Sure I understand it's in beta and not yet finished and yes watching your IPR converge is cool but I can't even draw a render region and it takes forever to converge to 100%. I was under the impression that this was supposed to be cool new technology. Well I'm finishing the last entagma tutorial on new features. Rendered in karma. 30 min per frame!!! Rendered the same thing in Mantra with diffuse limits set to 10 and 6x6 aliasing. 3 min per frame! 10x faster in Mantra. And honestly it looks better rendered in Mantra!

And yes I do understand that it's in beta but in addition to speed issues it's completely unusable. I get it to crash pretty much every few minutes.

Could SideFX not have invested all this time into mantra instead? What was wrong with mantra besides it being a bit slower? Definitely karma so far seems to have made the speed issue much worse.
Edited by LukeP - Oct. 25, 2020 23:43:00
User Avatar
Member
176 posts
Joined: Aug. 2013
Offline
Hi Luke.

Not that I can offer any advise but have you played with the convergence mode set to distributed yet. Have you found that any better/more controllable than the convergence mode set to variance.

I am trying pixel samples set to 32 and progressively increasing the diffuse samples to see when the grain is less noticeable (I am trying it on an interior which is the worst case scenario).

Best

Mark
User Avatar
Member
37 posts
Joined: Sept. 2018
Offline
Could SideFX not have invested all this time into mantra instead? What was wrong with mantra besides it being a bit slower? Definitely karma so far seems to have made the speed issue much worse.

I believe the main answer is USD and using the opportunity to create a new renderer (almost) from scratch.
Mantra simply can't understand USD and the devs have said there are certain things in Mantra that hold it back.

But I can definitely understand your frustration. I love working in Solaris (when it works) because of the great light handles, physic-based placement, and Karma is really really responsive which is nice. But I also haven't found a project yet where I could use it.
I'm sure a big reason for that is simply my own missing knowledge to properly utilize USD. For example, one of my constant issues is that a scene with packed primitives in Mantra will be 12GB while in Karma it will reach over 50GB in my RAM. I guess I could use USD Instances but I haven't figured out how to have the same flexibility in scattering as in SOPS.
The other thing is that I never really get rid of is fireflies in Karma. But this could again be because I know what settings to tweak in Mantra compared to no knowledge in Karma.
There is stuff that I like about Karma that Mantra can't do, like the Random Walk SSS, which looks incredible.

I honestly didn't expect there to be a learning curve like that as I remember using Redshift for a day and pretty much getting everything to work as I wanted.

I guess we do need to give Karma some more time to grow although I am unsure how “beta” it actually is and what parts are “working as intended”. SideFX said it will stay a “beta” until it has feature parity with Mantra (at least all features that are possible).

But USD is here to stay and so is Karma. It's unfortunate that we don't have a native production-ready renderer for USD in Houdini right now, especially because we all know how great Mantra can be. But hey, we will get Karma GPU at some point, and I'm sure Karma will become Mantras younger and hipper sister at some point.
User Avatar
Member
187 posts
Joined: April 2018
Offline
Karma is certainly much improved in 18.5, but I too am surprised by how slow it can be. I have a sculpted portrait that was previously set up in Mantra. A pretty clean image takes about 30 minutes with Mantra, maybe 20-30 with Arnold in Maya. Karma gave me an estimate of THREE hours. That's an outlier as simpler scenes render pretty reasonably in Karma, although noise is definitely a problem compared to Mantra. Still, something's clearly not quite right. I chalk it up to the beta status of Karma. I have no doubt it will be much more impressive in Houdini 19.
http://youtube.com/brianhanke [youtube.com]
http://twitter.com/brianhanke [twitter.com]
https://www.instagram.com/brianrhanke [www.instagram.com]
http://facebook.com/brianrhanke [facebook.com]
User Avatar
Member
176 posts
Joined: Aug. 2013
Offline
Hi.
I know denoising can be a swear word to some but I am trying the Intel Denoiser to speed up renders. It can use N, Albedo and GI output. The problem is I cannot see any options for Albedo and GI from the render Vars. Does anyone know how to make them (and plug them in to use)? best Mark
User Avatar
Member
7 posts
Joined: Sept. 2020
Offline
SideFX should have just integrated VRay as their goto renderer and tossed Karma and Mantra.
User Avatar
Member
176 posts
Joined: Aug. 2013
Offline
Hi. Although not fully setup for Solaris right now. 3Delight is blazingly fast. Defiantly the fastest CPU render I have used. It is still in Beta for Houdini right now though. Best
User Avatar
Member
716 posts
Joined: Jan. 2018
Offline
Another +1 for 3Delight if you're looking for alternatives. It is in beta, but it's solid as hell and works great in both the OBJ and Solaris contexts.

https://www.3delight.com [www.3delight.com]

I kinda wish that SideFX would have bought/licensed their technology instead of pursuing building a new renderer from scratch. I can only imagine just how powerful 3Delight would be if it had the advantage of being a Houdini native renderer.
>>Kays
User Avatar
Member
884 posts
Joined: March 2009
Online
Midphase
I kinda wish that SideFX would have bought/licensed their technology instead of pursuing building a new renderer from scratch.

Disclaimer: I'm not affiliated with SideFX nor do I have any insight into the company beyond what's publicly available.

That said, from my point of view it's good to have core technology inhouse, even though getting there may be a long, expensive and (for end customers) sometimes annoying procedure.
The “let's just buy that tech and get on with our lives” leads to where Autodesk and Adobe are right now. You have a melting pot of mindsets and backgrounds and legacy tech under one roof, diluting your own competence, and with every quarterly numbers report you fear another department that was previously purchased will get the axe.

I agree that Karma in its current form is limited, but I would advise to allow some more iterations.
Martin Winkler
money man at Alarmstart Germany
User Avatar
Member
716 posts
Joined: Jan. 2018
Offline
protozoan
I agree that Karma in its current form is limited, but I would advise to allow some more iterations.

Autodesk and Adobe always seem to get maligned but they are market leaders with impressive products which a ton of creatives use and love.

I'm not saying that SideFX should not develop in-house and just buy off-the-shelf technology, but there are some instances where it would be a better use of their resources, and a new render engine is a top contender due to just how complex and time consuming creating one from scratch can be (as we are witnessing).

The industry moves too fast to allow for “more iterations”, do you really believe that any professional is sitting around waiting for Karma to mature? Especially when other industry-standard engines like V-Ray, Arnold and Renderman are production ready today.

I have praised SideFX left and right and been a strong public promoter of Houdini, but I also believe in criticism when it's justified, and IMHO Karma has been mishandled from the beginning and I'm not quite sure how many “iterations” it will take for SideFX to correct course.

In the meantime, I am happy that there are other options available.
Edited by Midphase - Oct. 27, 2020 12:10:29
>>Kays
User Avatar
Member
32 posts
Joined: Jan. 2018
Offline
I don't like USD at all.
User Avatar
Member
218 posts
Joined: Jan. 2008
Offline
sysyfu
I don't like USD at all.
You dont like USD or Solaris?
What do you think about ifd?
User Avatar
Member
716 posts
Joined: Jan. 2018
Offline
What is IFD?
>>Kays
User Avatar
Member
531 posts
Joined: June 2006
Offline
Midphase
What is IFD?

IFD is the file format for mantra.
https://www.sidefx.com/docs/houdini/render/ifd.html [www.sidefx.com]

As USD is for Karma.
User Avatar
Member
4365 posts
Joined: Sept. 2011
Offline
sysyfu
I don't like USD at all.

Is it your least favorite open source, layered, extensible, universal standard interchange format that's renderer agnostic? What's your favorite one, USD?
User Avatar
Member
716 posts
Joined: Jan. 2018
Offline
The intentional obtuseness of some of the members of this forum is rather aggravating and verges on outright trolling.
>>Kays
User Avatar
Member
218 posts
Joined: Jan. 2008
Offline
Midphase
The intentional obtuseness of some of the members of this forum is rather aggravating and verges on outright trolling.
Sorry, but the reason for asking is not trolling. Im interested in why people think communicating with the renderer is better done with checkboxes, takes and expressions then nodes.

Simplicity? Sure, but that is not a USD problem, its a interface/solaris problem and I can promise you that mantra can be very complicated as well.

A honest question, In what way are you forced to interact with USD at all? Isn`t houdini pretty much doing all conversions for you on the fly with the scene import LOP?

…And Im not trying to tell you what works best for you. I do understand people reacting to how complicated some things are in Houdini.
User Avatar
Member
716 posts
Joined: Jan. 2018
Offline
I wasn't necessarily referring to you, but to answer your question, Solaris/USD is a solution in search of a problem for many freelance artists who have adopted Houdini as their main DCC. Just like PDG, it's good that it exists and it's obviously a welcome tools for our Method Studios friends above. However for many stand-alone artists, we would prefer to remain in the OBJ context while still having access to the more up-to-date render engine that Karma offers. Note that pretty much all 3rd party render engines including Redshift, 3Delight, Octane, Arnold, V-Ray, Renderman work both as Hydra Delegates as well as OBJ context renderers with no apparent hit in performance, so the question remains as to why Karma is any different?

When Sysyfu said that he doesn't like USD at all, what he probably meant is that he doesn't like Solaris.

We can keep going in circles arguing back and forth over something that is not going to change anyone's mind – we might as well be arguing politics or religion!
>>Kays
User Avatar
Member
37 posts
Joined: Sept. 2018
Offline
Midphase
However for many stand-alone artists, we would prefer to remain in the OBJ context while still having access to the more up-to-date render engine that Karma offers.

That is similar to how I feel as well. USD and Solaris are adding a level of complexity to scene building and management that wasn't there before.
You can't really start your scene from scratch and build and tweak everything on the go. USD/Solaris expects you to have pre-built assets, take great care with naming and collecting them, and wrap your head around terminology that changes between SOPS and LOPS.

I don't really want to import an asset from SOPS and tweak it in LOPS with a Sop Modify. But I also don't want to jump between the Obj and Stage context all the time.

That being said. I do like working in Solaris and I love the general idea of USD. Dreaming of something like USD, a standard that every major software will implement, that is literally the whole scene, was unthinkable a few years back. Adding the camera of your 3D scene into your compositing software with ease because it understands USD will be amazing. Sure FBX and Alembic can do that right now, but you'll be working in USD anyway and won't need to export it to something else. And it will enable more render engine variety across all software that supports it.

Then there is the node-based layer approach of Solaris which is just incredible compared to the loose assembly of nodes in the Obj context. It's much clearer and much more like working in SOPS.
Need a different light setup for the second shot of the sequence? Just branch it like you would when creating different versions of the same model in SOPS. This is also SO MUCH easier and clearer than using Takes. For me, this way of working feels much more “Houdini” to me than the Obj Context.

Then there are the benefits of easier reusing/importing scenes. Load the old USD file, prune everything you don't need. Then realize you need something else. You won't have to reopen the scene file and copy-paste again. The whole scene is referenced in the file you've imported at the beginning.

These are some of the reasons why I am going to force myself to use Solaris more (I can't really for actual jobs right now as I am using Mantra as my main render engine and Karma simply isn't ready yet, but I dip my toes into Solaris every chance I get).
Maybe I will be able to adjust my workflow to better fit into USD but I do think that it will be worth it for the benefits we can see right now and in the future.
We also shouldn't forget that one or two years ago almost nobody used USD besides the creators of it, Pixar. So it's to be expected that it's geared towards these kinds of companies and workflows. It's not been that long since the wide variety of audiences that Houdini serves have come into contact with it. I do think it's critical to make us heard, to assure that Pixar and SideFX are thinking about Solo Artists as well while adding to and changing USD and Solaris but I'm personally very optimistic that the future of Solaris will be good for my work. Maybe it will never be as easy as mashing together stuff in the Obj context but maybe the benefits will outweigh this.

Not sure if I gave anyone a valuable new viewpoint on this or maybe what I'm saying will simply not fit into your workflow, but in my (very subjective) view I see this as a chance.
Edited by No_ha - Oct. 28, 2020 20:21:23
User Avatar
Member
68 posts
Joined: Nov. 2013
Offline
Midphase
However for many stand-alone artists, we would prefer to remain in the OBJ context while still having access to the more up-to-date render engine that Karma offers.

I agree both options are needed because the power of solaris/lops/usd/hydra isn't always going to offset the cognitive load. But this problem really does boil down to a full-featured scene import lop running behind the scenes, exactly like there has always been a scene-import-to-mantra step happening before (that few ever cared about).
  • Quick Links