Hi,

Is there a full GPU renderer avaiable for Houdini (with realistic renders obviously), like furryball for maya ?

Do you think mantra will become soon full GPU accelerated (no just for 1- funbctions as it is so far)?

Thanks

Needsun
Is there a full GPU renderer avaiable for Houdini (with realistic renders obviously), like furryball for maya ?

No.

Do you think mantra will become soon full GPU accelerated (no just for 1- funbctions as it is so far)?

Not soon, most probably never (hopefully).

ps what is “1- funbctions”?

why do you say “hopefully”? you prefer to wait days to render a scene isntead of minutes? :shock:

One might make the case that CPU renderers tend to be more flexible and efficient at accurately processing dense scenes with heavy memory demands.

Chip manufacturer Intel has made such such arguments [a676.g.akamaitech.net] in some of its promotional materials.

http://www.3dworldmag.com/2011/01/07/pros-and-cons-of-gpu-accelerated-rendering/ [3dworldmag.com]

if you look the current speed up from mantra thanks to the better data handling it's realy hard for me to look to a gpu render at the moment.

mantra is a AAA render and it's solid a gpu render can crash an gives you troubles with problem's that you don't have in a cpu render. (we have got here octane render)

but mantra has some disadvantage against a gpu render.
1. you have to learn the render to use it correctly. ( it's not a toy)
2. to get a solid rendering in a short time (< 10 min final render quality). optimize your workflow
3. think the houdini way.

i have made just for fun a environment ligth linked to the compositing nodes and now i have got functions nearly like hdr light studio for free.

Needsun
why do you say “hopefully”? you prefer to wait days to render a scene isntead of minutes? :shock:

Start working in a real production (films) and then you'll appreciate what is like to have solid, focused company developing state-of-the-art, responsive, and predicable renderer for the actual market of its choice.

None of the existing GPU can stand this try to my knowledge. Since all you see is either static architectural design or moving spheres you don't even know how many features are not supported in GPU stuff, and how many bugs and limitations you encounter before you reach any decent picture, motion picture.

Show me multi-segment deformation motion blur on GPU for start.

Saying that, I really don't mind support for some GPU renderers out there, just not making Mantra the one. Octane seems to be pretty mature. Also recently announced redshift3d looks like the first film oriented GPU render engine. Judging by its documentation, it has a solid design and might be pretty good. We will see.

Octane is very, very nice. The only thing that is missing from it is Houdini integration. I'm working with Octane and the Cinema4D integration and it's a dream, at least for simpler scenes. Octane is offering a special developer program to develop inegrations together with them - maybe Sidefx could go on the initiative here?

I think people looking at GPU renderers also do it because not all of us want to use Houdini to render fx, volumes, etc…

Sometimes, when using the procedural abilities of Houdini you end up just wanting to create and animate stuff then send everything out to an engine like Octane to get your result quickly, with a nice result and 240 frames rendered at 5sec/frame with path tracing.

I takes 3 minutes to render at full HD with mantra, but 5 seconds on the other side is not something you can ignore. Octane can eat a lot of geometry and textures with current hardware.

My current way is to export everything using Alembic and use Octane within Maya. I hate to have to re-open Maya for rendering (or anything…), but until a better connection to Octane comes to life, it's still the best solution…

What would be helpful is to compare the results you are getting from the Octane compared to Mantra. We know the quality that is demanded from Mantra today and before we move resources over something like Octane integration we would need to know it meets those expectations.

Would you be able to use one of the default PyroFX shelf tools to create an effect then show us the results rendered in Mantra compared to the Octane? Just curious to see the results.

Robert

What would be more interesting than a full GPU-solution (at least until it is up to par with CPU-renders, it will probably take a while), I think, would be for Mantra to take advantage of cards/co-processors like the Xeon Phi.

Hi Robert,

I don't think Octane is able to render pyro volumes so it will be hard to compare that kind of features. But again, people are using Houdini for other things than effects. As you are democratizing Houdini's price, people look at it as an alternative to Max, Maya and C4d. Although I'd rather do only FX stuff, I also have a lot of projects not involving any pyro or water sim, and I still do them with Houdini just because of it's procedural and nodal approach to 3D.

When it comes to render polygon geometry with skylight and path trace, even though it's a tedious setup to get the data to Octane (which until now still doesn't support alembic), and even with limitations like no geometry motion blur, it' s still much faster for me to get my stuff rendered with Octane and a few additional passes with mantra will help me get velocity that I can add in comp.

Here is a wip example of the same frame rendered with mantra and Octane. Mantra (on the left) is around 3 minutes, Octane took 17 seconds.
Hardware : intel i7-3930k, geforce GTX 590.

Bruno

i would prefer a Mantra on a Xeon Phi and the porting time should be small because the most part of the optimisation you have done already.
the best thing is to get in contact with intel ask them for a testdrive and invest 1-2 days for the porting. with phi you don't have to write opencl or cuda code!

or there is also the possability to integrate something like that:
http://mycryengine.com/index.php?conid=59 [mycryengine.com]

I recently purchased Octane Renderer - and am very impressed.

One of the advantages of an unbiased renderer is that it takes far less time to setup lighting. You don't have to move this parameter up because you moved that one down. It's far simpler.

With 2 Titans in my box, and the ability to add 4-8 more using an external chassis - the speed improvement is really massive.

I'm looking at Houdini now as a possible alternative to 3ds/Maya and Lightwave (love those daily Houdini builds and the almost immediate fixes I've read about - an alternate universe from the reality at AutoDesk or Newtek).

Not having a plug-in for Octane is one thing holding me back, so I'd like to add a +1 to any effort to get this implemented.

I know that Maxwell renderer is in the process of been ported and you have the “fire” interactive mode (I think it's hardware accelerated)

It seems that Maxwell is available now for Houdini 11 & 12:

http://www.maxwellrender.com/index.php/products/maxwell_render_suite/plugins/#/houdini [maxwellrender.com]

However, it looks like it's CPU based - and they brag on not requiring “expensive GPU's”:

http://support.nextlimit.com/display/maxwelldocs/Maxwell+Fire+-+Fast+Interactive+Rendering [support.nextlimit.com]

I own RealFlow and they have added some GPU support to part of it, so hopefully this will change.

I can certainly attest that GPU's are less expensive than CPU's - and a whole lot easier to setup than a large render farm. And with 128GB RAM on my dual xeon, the memory per thread available for rendering is 4GB - 50% less than available with each Titan's 6GB.

I'm sure the next geometry shrink will provide even more GPU memory and GPU power.

Re

“i have made just for fun a environment ligth linked to the compositing nodes and now i have got functions nearly like hdr light studio for free.”

any chance you want to share that .hip file? I was thinking of setting up the same thing.

Thanks, Kevin

Looking very promising:
http://raytracey.blogspot.com [raytracey.blogspot.com]

Killer if they get Alembic i/o in there…

cpb
Looking very promising:
http://raytracey.blogspot.com [raytracey.blogspot.com]

Killer if they get Alembic i/o in there…

Brigade is a project OTOY (Octane) are working on for the game market.

Apparently a lot of the research ends up in Octane.

I'd guess the end-result will be real-time previews in Octane - and massive development $$ for OTOY as I think the game market is larger dollar-wise than the entire movie market.

Here's some info on Octane and Brigade:
http://www.fxguide.com/featured/octane-render-realtime-ray-tracing/ [fxguide.com]

and an even more recent article on all the major renderers (including Octane)

http://www.fxguide.com/featured/the-state-of-rendering-part-2/ [fxguide.com]

+1

houdiniWannabe
I recently purchased Octane Renderer - and am very impressed.

One of the advantages of an unbiased renderer is that it takes far less time to setup lighting. You don't have to move this parameter up because you moved that one down. It's far simpler.

With 2 Titans in my box, and the ability to add 4-8 more using an external chassis - the speed improvement is really massive.

I'm looking at Houdini now as a possible alternative to 3ds/Maya and Lightwave (love those daily Houdini builds and the almost immediate fixes I've read about - an alternate universe from the reality at AutoDesk or Newtek).

Not having a plug-in for Octane is one thing holding me back, so I'd like to add a +1 to any effort to get this implemented.

Robert,

I would rather prefer if resources are spent on making Mantra faster. For that example, which of course is not globally representative, of 3 min Vs 17 seconds IMO does not justify the speed gain. No real displacement, no real transformation and deformation motion blur, no volumes, no shading language, no subdivision surfaces …)

Sorry but IMO GPU renderers have still a long way to reach the quality and reliability (cross-platform & hw-agnostic) of a good offline renderer. If the GPU can assist the CPU render in a reliable way this would be already a big success (and you will still hit driver and cross-platform issues).

P