News for hackers. Stuff that matters.
Re:How close are we to truly photorealistic, real-time games? by monochromatic on Friday February 10, 2012 @03:27PM (#3577335)
> 288 hours PER FRAME

So it took close to a year to render each second on-screen. I do not believe this.

Re:How close are we to truly photorealistic, real-time games? by VikingCoder on Friday February 10, 2012 @03:51PM (#3577422)
288 cpu hours.

Divide that by a large number of CPUs.

Most rendering is very parallelizable.

Re:How close are we to truly photorealistic, real-time games? by astrodust on Friday February 10, 2012 @06:55PM (#3578042)
Yeah, that's a day on a 12-core system.
Re:How close are we to truly photorealistic, real-time games? by sliverstorm on Friday February 10, 2012 @04:18PM (#3577513)
I believe that number is per node, e.g. they have a cluster of 10 nodes that rendered for 24 hours producing 1 frame per day. Much like the concept of "man-hours" in a project.

[1] http://www.hollywoodreporter.com/news/transformers-dark-moon...

Re:How close are we to truly photorealistic, real-time games? by astrodust on Friday February 10, 2012 @06:56PM (#3578048)
There's only 24 frames in a second for production-caliber film. They also throw an obscene number of cores at the problem. On a 12 core system you can render that out in a day. If you have 24 of those systems in a rack you can pump out one second of film time per day.

That means in a month you can deliver 30 seconds of footage. If you have a five minute sequence, get ten racks like that and you can turn it around inside a month.

Also, since when is 288 hours even close to a year?

Re:How close are we to truly photorealistic, real-time games? by Jimmie on Sunday February 12, 2012 @12:25PM (#3582325)
288 hours per frame, 24 frames per second, 24 hours per day = 288 days per second of film time.

Pretty close but the assumption that you'd be using a single CPU is wrong.

Re:How close are we to truly photorealistic, real-time games? by noblethrasher on Friday February 10, 2012 @03:58PM (#3577452)
Rendering is one of those embarassingly parallelizable problems we've been hearing so much about.