3 ms·

I'm not an expert in either cloud-computing or 3d rendering, so I'm not saying that I've got it all figured out; "easy-peasy". But I do think there are some way

I'm not an expert in either cloud-computing or 3d rendering, so I'm not saying that I've got it all figured out; "easy-peasy". But I do think there are some ways to transform the problem that could be more beneficial than the "traditional" render-farm approach that most of the industry seems tied to. Things like, progressive enhancement rendering -- either quality or resolution -- like wavelet compression for rendering. Massively parallelized rendering pipelines -- borrowing from the high performance password cracking on GPUs, and real-time game inspired rendering pipelines. Static scene analysis/dynamic compositing; rather than require the artist to think about layering and compositing, let the computer algorithmically break down a scene and maximize rendering speed. Perhaps it will eventually be determined to not be cost effective -- and I'm sure people much smarter than I are working on it -- but I think we will see some really interesting and novel changes in this space in the next 5 years after being relatively unchanged for the past 20. And just for the record, I think Clara.io is fantastic, not saying you guys are doing anything wrong, just talking. :)

bhouston 13y ago
Yeah, the issue is GPU/CPU power required for photorealistic rendering is really high right now and it is costly. If there is some way to bring it down, which is outside of my area, then you are right the economics will change and we can always do live updating progressive rendering for nearly nothing. This will happen of course at some point, but I think we are at least a few years away. For now we just have high quality viewport that is similar to game quality graphics, that is super cheap because it leverages the local GPU rather than cloud resources.
- Arelius 13y ago
  I think the problem is that the economics of offline rendering play out a bit differently. We've been doing offline rendering for a pretty long time now. And compute resources keep going up vs cost and time significantly, but the price of rendering is so small versus the price of artists. That your bottleneck is always* going to be on the artists, and improvements in performance will always* be spent with improving visual fidelity once you get past a certain speed. Turns out, if you want high-quality(All quality is relative) rendering that doesn't take much time at all, we can do it any modern game engine can output decent image in about 30 ms. And an even better one still quicker than a human can react. And there are tons of people which that is quite sufficient. Videos from source filmmaker show that: http://www.youtube.com/watch?v=WUhOnX8qt3I http://www.youtube.com/watch?v=WUhOnX8qt3I http://www.sourcefilmmaker.com/ http://www.sourcefilmmaker.com/ Also see UE4 infiltrator: http://www.youtube.com/watch?v=-bLOi3mo9NE http://www.youtube.com/watch?v=-bLOi3mo9NE * I use always lightly here, that is of course not true on certain time scales.
Arelius 13y ago
I'm not really trying to contradict you, so much as I'm a little humored by some of your suggestions, not because they are naive, but because they are common. > Things like, progressive enhancement rendering This is a very common option in a huge number of renderers. Particularly physically-based renderers It particularly useful when you want interactive feedback, less so for 'final' renders. > Massively parallelized rendering pipelines. Rendering is one of the original problems that generally fit into the "embarrassingly parallel" category, hence why render farms are even a things. There are very many existing offline renderers that take great advantage of GPU computation. Paired with progressive enhancement, certain renderers use this particularly well for interactive editing. But it's also used in very many a render farm. > Static scene analysis/dynamic compositing ... let the computer algorithmically break down a scene and maximize rendering speed This has been done since likely the day rendering was invented (not actually). The artist only thinks about layering and composting when the artist wants specific composting control in post. And there is no way you could do that as an automated process, just as you can't do the rest of the art as an automated process. > Perhaps it will eventually be determined to not be cost effective What won't be cost effective? Those techniques? They already are! Or cloud-computing for offline rendering? It already is as well! Some of the first clouds as a service were and are put to great use for offline rendering from the very start. And a 'render farm' is really just a 'private cloud' for rendering. > after being relatively unchanged for the past 20. This is confusing to me. Rendering is a field with one of the most progress every year. You only have to go to Siggraph once to see that as self-evident. Anyways, I'd love to talk more or answer any other questions you have. Rendering has been a passion of mine for over half my life. And anything that has the ability to transform how we do things is very interesting. Just as Clara.io very much is.

bhouston 13y ago

Yeah, the issue is GPU/CPU power required for photorealistic rendering is really high right now and it is costly. If there is some way to bring it down, which is outside of my area, then you are right the economics will change and we can always do live updating progressive rendering for nearly nothing. This will happen of course at some point, but I think we are at least a few years away. For now we just have high quality viewport that is similar to game quality graphics, that is super cheap because it leverages the local GPU rather than cloud resources.

Arelius 13y ago
I think the problem is that the economics of offline rendering play out a bit differently. We've been doing offline rendering for a pretty long time now. And compute resources keep going up vs cost and time significantly, but the price of rendering is so small versus the price of artists. That your bottleneck is always* going to be on the artists, and improvements in performance will always* be spent with improving visual fidelity once you get past a certain speed. Turns out, if you want high-quality(All quality is relative) rendering that doesn't take much time at all, we can do it any modern game engine can output decent image in about 30 ms. And an even better one still quicker than a human can react. And there are tons of people which that is quite sufficient. Videos from source filmmaker show that: http://www.youtube.com/watch?v=WUhOnX8qt3I http://www.youtube.com/watch?v=WUhOnX8qt3I http://www.sourcefilmmaker.com/ http://www.sourcefilmmaker.com/ Also see UE4 infiltrator: http://www.youtube.com/watch?v=-bLOi3mo9NE http://www.youtube.com/watch?v=-bLOi3mo9NE * I use always lightly here, that is of course not true on certain time scales.

Arelius 13y ago

I'm not really trying to contradict you, so much as I'm a little humored by some of your suggestions, not because they are naive, but because they are common. > Things like, progressive enhancement rendering This is a very common option in a huge number of renderers. Particularly physically-based renderers It particularly useful when you want interactive feedback, less so for 'final' renders. > Massively parallelized rendering pipelines. Rendering is one of the original problems that generally fit into the "embarrassingly parallel" category, hence why render farms are even a things. There are very many existing offline renderers that take great advantage of GPU computation. Paired with progressive enhancement, certain renderers use this particularly well for interactive editing. But it's also used in very many a render farm. > Static scene analysis/dynamic compositing ... let the computer algorithmically break down a scene and maximize rendering speed This has been done since likely the day rendering was invented (not actually). The artist only thinks about layering and composting when the artist wants specific composting control in post. And there is no way you could do that as an automated process, just as you can't do the rest of the art as an automated process. > Perhaps it will eventually be determined to not be cost effective What won't be cost effective? Those techniques? They already are! Or cloud-computing for offline rendering? It already is as well! Some of the first clouds as a service were and are put to great use for offline rendering from the very start. And a 'render farm' is really just a 'private cloud' for rendering. > after being relatively unchanged for the past 20. This is confusing to me. Rendering is a field with one of the most progress every year. You only have to go to Siggraph once to see that as self-evident. Anyways, I'd love to talk more or answer any other questions you have. Rendering has been a passion of mine for over half my life. And anything that has the ability to transform how we do things is very interesting. Just as Clara.io very much is.