Oscar's rendering thread

[Oscar J]

A new Indigo v4 shot. 12 mpix (4k * 3,2k) resolution, render time 25 minutes on my Radeon 7970. The model doesn't quite hold up to that res, but anyway.

[zeitmeister]

Fantastic.
How fast is the render refresh response when you add or change a material or rotate the camera?


Cheers, zeiti

[Oscar J]

I think it's best if one of the devs chip in and answer that kind of questions.

[zeitmeister]

Hm?
An example: indigos sample refresh is a lot faster here in my case when I use cpu; that means I see the picture earlier. When I use GPU (Titan Black), I wait a lot longer to see the image.


Cheers, zeiti

[lycium]

I think Oscar deferred to us because I spammed him to death explaining it recently

In that Indigo 4 preview video, you can see the super snappy feedback from direct OpenGL integration, which in my opinion is one of the major things people expect from GPU renderers - that Octane-style super high FPS preview. This rendering mode is tuned for low latency, but GPUs are fastest when tuned for high throughput, so the "normal" rendering mode has higher latency than you would see in that video - we blast all the GPUs with relatively long running kernels to get the converged image done as quickly as possible.

I've done some work to have both modes working side by side, with autodetection for the primary / display GPU and auto-switching between fast preview and normal (high throughput, multi GPU) rendering mode. So you can expect to have that in Indigo 4 for sure, but it's not in the current closed beta versions.

Another issue is that of kernel rebuilds: we're going to have to do a considerable amount of work optimising Indigo to do as few of these as possible, important when we have direct integration with modelling apps.

[bubs]

A new Indigo v4 shot. 12 mpix (4k * 3,2k) resolution, render time 25 minutes on my Radeon 7970. The model doesn't quite hold up to that res, but anyway.

Sweet!

[bubs]

when we have direct integration with modelling apps

I like the sound of that!

[Polinalkrimizei]

How did I miss the preview video?!
I didn't find it on the front page... (or elsewhere)

But you're making me all excited!

[lycium]

https://vimeo.com/113329185 The video I was referring to

[suvakas]

wowow...nice !!

Suv

[thesquirell]

This is the future.

I was wondering, maybe I even asked this before, but in GPU rendering, is it possible to use GPU farm?

[Oscar J]

Thanks guys!

Sure it is squirrel, why wouldn't it be?

[Oscar J]

A little carbon test. About a minute of rendering.

[thesquirell]

Seems unreal to me, the speed boost is just insane. Imagine a farm of GPU's, literally killing it. For RT visualization there's Unreal. For renders, there's GPU, and this is the future. You guys at Indigo are creating history. If you manage to get this working, with realism we have now, slightly improve support and make-up, this is going to be huge.

[Oscar J]

I agree, the speed is impressive. I have a single GPU from 2012, but 4-6x 980Ti or Titan X rigs are not unusual in the business. Would chew their way through 4k animations without breaking much sweat.

Speed is not everything though. As you say, there's a lot of work left do to in other areas.