Isotropix Forums

GPU rendering for final frames

GPU rendering for final frames

Unread postby bjoern » Thu Mar 08, 2018 1:03 pm

I know this was discussed before. And from what I understand, isotropix is working on accelerating the viewport via GPU raytracing (refering to last siggraph and the Nvida link). But please hear me out and why this is important and also technically realistic.

Short version:
-The indie/freelance Market is the future and biggest market, I made be wrong. I know for a fact that it is for Maya (60%). Rendering in Clarisse (with CPU) is from the money side simply not realistic for such an individual, or at least not optimal since the rise of GPU rendering (proof in the long version)
- Let the artist make the decision with what he wants to render. If he/she goes GPU, well then they have to optimize the scene for RAM. It's not like this is anything new to us ;) . Not giving Artist GPU because you cannot render everything you want with it is the same as babysitting an adult, hehe

Long version
- Ones https://elara.io/ hits the market the landscape of VFX is gonna experience a tectonic plate shift. Slowly but surly! More Individuals will do there own shops or gang up. On that note, I hear folks always saying GPU's are expensive... Which makes little sense to me. Buying every 1-2 years 6-8 GPU's for your own render server is not expensive. ...you don't always need the top of the line model just pick the generation before the newest one and you are good. CPU rendering for an individual is only realistic when utilizing the cloud. But then you need to keep an eye on your render cost. Which will be long term more costly then rendering on your own little GPU farm. And This is not a guesstimation. I have former students who are in exactly this spot. They did there own thing, pipeline is Maya+Redshift. They never used a cloud so far. GPU is locally cheaper. But doesn't Elera and local GPU server contradict it self? A little. But it mostly will have a big effect because of Nuke I think. Artist need it. Its hard to find talent for fusion if you need to scale up. And clients sometimes request the work to be done in nuke for handoff's. Of course you could go all out and to also Katana + Arnold etc... But I doubt individuals want to invest time into that (mainly katana).

- On Optimization... I see games where I'm sometimes not sure anymore if it is rendered or ingame. But I'm 100% sure that if you would take these game assets and set them up in clarisse with some modifications you can definitely render them on the GPU as final frames and they will hold on the big screen. And to be honest, ever since this short, now 3years old (redshift): https://www.youtube.com/watch?v=WWN-z6zjDaw Didn't the argument, "that that you cannot render impressive images on the GPU", received its final knockout!? ;-)
Yes for sure you can max out the GPU Ram and make scenes that will never render. But who cares?? If an artist is not mindful and is dumping assets and textures like a mad man into a scene, then of course it is not gonna render. Again. Who cares!? Let it be up to the artist to optimize his scene :) And let him decide how he approaches his rendering. And if one does not want to optimize his scene, well then go CPU and spend the $$$.

All this will be probably only a temporary situation. But I wanted to get it of my chest, hehe. Ones PRman and Arnold can render final frames on the GPU https://blogs.nvidia.com/blog/2018/03/02/media-entertainment-gtc-2018/. Everybody else has to catch-up. I just loved to see isotropix to be first rather then to catch up.
Last edited by bjoern on Thu Mar 08, 2018 3:16 pm, edited 1 time in total.
bjoern
 
Posts: 236
Joined: Thu May 11, 2017 1:43 pm

Re: GPU rendering for final frames

Unread postby bjoern » Thu Mar 08, 2018 2:20 pm

A great example on how to setup a shot while managing complexity (Redshift). https://vimeo.com/243841207
If I showed this to a non-Artist on the street, I doubt they would know its CG. And from what I could research, this kid is still a student. Imagine a seasoned veteran handle such a shot and tweaks it. Detail detail detail...! :-)
bjoern
 
Posts: 236
Joined: Thu May 11, 2017 1:43 pm

Re: GPU rendering for final frames

Unread postby xilofoton » Fri Mar 09, 2018 1:15 am

bjoern wrote:-The indie/freelance Market is the future and biggest market, I made be wrong. I know for a fact that it is for Maya (60%). Rendering in Clarisse (with CPU) is from the money side simply not realistic for such an individual, or at least not optimal since the rise of GPU rendering (proof in the long version)


Another proof of concept for indie/freelance market is the story of Octane render (now it's available in Unity3D) + Render Token.
I asked earlier some questions from Jules Urbach on Reddit about Clarisse+Octane:
https://www.reddit.com/r/IAmA/comments/72ds8k/im_jules_urbach_the_ceo_and_cofounder_of_otoy_we/dnhs6su/
He linked a video, it's worth to watch, and it is 5 years old!

However I can barely imagine how Dneg could render the extremely detailed VFX for 2049 with the limitations of the GPU...
Maybe for the brute force rendering of the rain (they used droplets with proper shading), but for the 600 billion poly cityscape...

With Houdini you can separate the simulation with multiple partially overlapping boxes - clusters - and each can run on different machine. (kinda buckets in space). If similar is possible with GPU rendering in a network and each side of a box is a portal light thingy which can communicate with other render nodes, it might be a winner... but needs a quite fast fibernet I think :)
VFX & Cinematic Scientific Visualization Artist, Clarisse Specialist
https://artstation.com/scivfx
https://scivfx.wordpress.com/
xilofoton
 
Posts: 240
Joined: Tue Jul 21, 2015 2:29 pm

Re: GPU rendering for final frames

Unread postby bjoern » Fri Mar 09, 2018 11:12 am

However I can barely imagine how Dneg could render the extremely detailed VFX for 2049 with the limitations of the GPU...
Maybe for the brute force rendering of the rain (they used droplets with proper shading), but for the 600 billion poly cityscape...


I'm glad you bringing this up. Because it should have really no influence for isotropix to implement GPU rendering or not!
Big studios have from my experience not a good habit of being efficient when it comes to building assets or being mindful with resources. It is often the opposite! And its not the artists fault, time constrain, mass production etc...
Chatted with an old coworker recently... VFX production story time again about exactly that, unefficient workflows... Assets get completely overly subdivided with gigantic textures for no reason and so on (Major recent Blockbuster production, Big Oscar winning studio).
But hey, they can render on the CPU anyway. So they are covered, "check", case closed! Now let's focus on everybody else,
who does optimize there scenes and is not proud of 4h hours render-times.
The forest scene up top is a good example. You could blow that one up easily, with lots of Hires speedtree models, lots of sub-dived assets until the you only see wire frame and lots of displacement. Or you do it right, and then you render it on the GPU, in a fraction of the time!
bjoern
 
Posts: 236
Joined: Thu May 11, 2017 1:43 pm

Re: GPU rendering for final frames

Unread postby mdkai » Sun Mar 11, 2018 1:10 pm

I'm a bit more pragmatic about this idea... Technically it's a bit more than just add it..
Beside the fact you need to maintain two code branches which will affect the production cycle of the entire product I doubt that this is making sense considering that clarisse tries to be clean and as flexible as possible.
Im not a developer but from my experience it's barely worth the effort to implement such a feature, which would affect the entire feature list like instancing, image textures, nodes etc. as well as time and resources that would impact the regular development.. Since the whole marketing strategie behind clarisse is to be able to manage massive scene.. Adding GPU or hybrid solution will have them to take this back an say "only in xyz mode".

I would not make that mistake.. There are plenty of GPU engines so why try to get into their domain if you already rule in your own.

Aside GPU and CPU aov passes are not really for mixing since they use different math processors.. Same rule applies already for cpu brands

And your examples are as promising and impressive as they are, having one thing missing.. A student got way different levels of time then any artist running shots or assets on a budget with client. :roll:

Overall I would also like to see that as an extra for sure.. But I do not think that at the moment isotropix will invest so much into an extra render when they would focus on more important features and rather open they API to support other renderengines easily :)
mdkai
 
Posts: 116
Joined: Tue Oct 07, 2014 8:24 pm

Re: GPU rendering for final frames

Unread postby bjoern » Mon Mar 19, 2018 11:53 am

Beside the fact you need to maintain two code branches which will affect the production cycle of the entire product I doubt that this is making sense considering that clarisse tries to be clean and as flexible as possible.

Arnold, Vray, have todo the same.
bjoern
 
Posts: 236
Joined: Thu May 11, 2017 1:43 pm

Re: GPU rendering for final frames

Unread postby mdkai » Mon Mar 19, 2018 3:44 pm

And did you check on what state they are and how long it took so far ?
Arnolds GPU branch is not officialy there, if it ever shows up..
And Vrays GPU render took years(!) to be at the current state, which is still not 100% identical to the CPU part..

So again, not that i wouldn't like to see that some GPU power is harnessed.
I think they can spend time better and also following not the same route all others are on and rather find a better, unique usage for the GPU.
mdkai
 
Posts: 116
Joined: Tue Oct 07, 2014 8:24 pm

Re: GPU rendering for final frames

Unread postby bjoern » Tue Mar 20, 2018 11:50 am

I understand what you are saying. But in regards to small businesses, I would disagree on the importans of GPU development. Any render that wont be capable of using the GPU in the next 1-2 years will become to these company's completely irrelevant. Regardless of the other features that software might bring to the table. Most of them do it already, firewithoutsmoke for example. First Arnold then they moved to Redshift. And there are allot more.
I personally "believe" who ever masters the GPU best, and is orig. a CPU render (Arnold, PRman, Vray etc...) will take the lead in the global market very quickly, for all sizes of businesses. But it's just my personal opinion :)
bjoern
 
Posts: 236
Joined: Thu May 11, 2017 1:43 pm

Re: GPU rendering for final frames

Unread postby mdkai » Wed Mar 21, 2018 12:34 am

I agree 100%
We recently started a small branch, that is working with a few redshift/octane licences and it works like a charm for them. It is also for an area that is used to fast turnover dates, sometimes almost same day.. and not using the most complex scenes, it is quite the opposite .. small, simple, but very slick.

The other area is far more complex and requires hair , fur, stereo, support for ACES and complex AOVs and masks .. GPU simply can not provide the same experience in terms of complexity and quality (but speed and quality).
And that's fine and legit.

Thats why i think clarisse , as a tool for really large scenes and complex structures will no fit inside a GPU core branch for rendering.
And i really am happy to hear that they plan to use it for the viewport acceleration which i think is much smarter as that is the most they can do to utilize the extra power.

Also, constant development and marketing is also an important part .. just look at Zbrush vs Mudbox .. One is very active, doing webinars per week, gives free feature rich updates for years , the other is doing not much anymore and gives updates free of any feature at all (they do compatibility updates every year ...).. a shame really.

And i want a swarm client to add more machines to the main local machine ;D
mdkai
 
Posts: 116
Joined: Tue Oct 07, 2014 8:24 pm

Re: GPU rendering for final frames

Unread postby xilofoton » Tue Mar 27, 2018 12:12 pm

VFX & Cinematic Scientific Visualization Artist, Clarisse Specialist
https://artstation.com/scivfx
https://scivfx.wordpress.com/
xilofoton
 
Posts: 240
Joined: Tue Jul 21, 2015 2:29 pm

Next

Return to Feature Requests Archives