What renderer do you use with Houdini and why?

in The Commons
I've contracted a very capable Houdini hacker to write an Iray-to-Houdini materials converter. He's made a first pass for Mantra, but says that Mantra is slow and outdated. In order to not give him any more bad directions (that end up costing me money) I'd like to start a conversation on what render(s) he should support.
So far, Mantra has been kind of ruled out, being both old and slow.
The new one, Karma, is in beta and not yet feature complete, e.g. volumetrics is not fully implemented.
Octane is of course very nice, but apparently there is no free option for Houdini, and it requires a persistent internet connection.
What are some other informed choices?
Comments
Mantra is one of the major Go-To-renderers in the movie/FX industry. Claiming it's "outdated" mainly shows that the "capable Houdini hacker" most likely is coming from the Indie Games world and isn't familiar with normal productions. Even though Mantra will eventually be replaced by Karma, you can be pretty sure it's going to be around for quite some time. Plus, its integration into Houdini is something to keep in mind if you are thinking about "easy to use" approaches.
The renderer of choice always comes down to your specific requirements. Like your introduction says, a conversion pipeline from one renderer to another is most likely what you will have to implement in any workflow. There is no "PBR" that would just simply define values and give you the same results in every render environment that "supports PBR", you will always only deal with directions.
I have written iray- and 3ddelight importers for Houdini (Mantra, Redshift, experimental Karma), but those cannot work 100% reliably, since iray IS NOT Mantra and you will have to tweak values anyway. The way light sources are handled are different, your scene scale may be different, adding in parametric materials may result in differences, converting data to linear color spaces will yield different render results (DAZ material mainly uses a decades-outdated 8-bit-per-color texture world that converts VERY badly to anything more modern than 1998, tongue in cheek).
3d Delight (the real deal, currently in hot development - not the decades old "ancient archaeology" version used in DAZ) is definitely something to look at, it's moving at fast speed, is very flexible and its Houdini integration is quite good.
Redshift, Octane, VRay and all the other "standards" out there are candidates, but if you want "many users" of whatever you are developing, Mantra would be my suggestion for at least five more years.
Marc Albrecht
I agree with Marc that Mantra is an exceptional renderer and very capable but I also agree that it is slower and it is not the way of the future. If you are forward-thinking, Karma is the future. Here is why: Karma will be Houdini's native USD render candidate and if you are not thinking about USD, then you won't be in step with the rest of the industry. Soon every VFX pipeline will revolve around it. Give it serious thought. 3Delight is also a good choice, though I would recommend its older sister, Renderman. Renderman will be a staple of USD for a long time, I think. Redshift is one of the fastest renderers and it will be USD capable soon, if not already. And then there is Arnold. I prefer Arnold because its standard surface material shader is looking to become a standard across many render formats, thus it will be much easier to convert the Arnold Standard Material to other popular renderers such as Octane and Redshift, both of which will be integrating it into their platforms. In the end, as Marc stated, it depends on your product goals and time target.
Sorry, could you please say what USD is?
Probably the best thing is to show you. What is USD?: https://graphics.pixar.com/usd/docs/index.html What will be the next powerhouse platform based on USD?: https://developer.nvidia.com/nvidia-omniverse-platform
Thanks. Will read into this "Universal Scene Description" (and not $).
edit:
Reminds me of Outliner from Maya.
Thanks for all the valuable input so far, particularly the points about USD.
Blender can already export to USD,USDC & USDA.
Not many apps import and render the format yet.
The NVIDIA Omniverse plans looks fairly forward thinking
in that trends seem to indicate that demand for ready to use content .in a standardized industry format, will increase.
People in the content markets should take heed.
The notion that people with established pipelines.
will come to your market install your application ,buy
content in your exotic proprietry format (.duf)and have to
load it into your application and then export to your software using some "bridge" plugin is folly IMHO.