[Tig] 8k IMAX scans... 16k next

[Bob Friesenhahn]

On Thu, 24 Jul 2008, Rob Lingelbach wrote:
>> used (for free) in future productions.
>
> Bob, something that also hasn't ceased to surprise me, in addition 
> to your comments, is that instead of developing parallel processing 
> to its fullest, still the basic layout of compositing workflow 
> involves a render farm, where the machines are mainly 
> single-processor, and instead of doing multi-threaded multiprocessor 
> rendering, you're basically just putting a bunch of commodity 
> machines connected via ethernet and letting the tasks be allotted by 
> some scheduling daemon.  It was like, 30 years ago, that parallel 
> processing was

Yes.  There are well-entrenched companies in the business (e.g. 
"Rhythm & Hues", "Industrial Light & Magic", "Pixar") who have turned 
this into an art form and are able to develop in-house technologies 
and resources to apply to new jobs.  At least these companies are 
building something with a view to the future and are able to advance 
the technology.  Sometimes they even contribute to the external world 
(e.g. OpenEXR & CinePaint).

There are other companies/productions who set up shop, purchase a 
bunch of new hardware, license a bunch of software, and hire people to 
do the work.  These companies are not building anything other than the 
production they are working on at the moment. They are slamming out 
the production the same way they would an on-site set.  In the end 
everything is dismantled or lost.  It is a sort of "Groundhog Day" 
scenario for those involved.

> idle.  There is a very good article in American Scientist (my 
> favorite magazine) back 9 months ago or so that talks about how 
> finally, multi-core CPUs are bringing the threading nature of 
> programming into reality, many decades after its conception.

This article is good and accurate.

> (why haven't they been thought out in the last 30 years? because of 
> the cheap availability of commodity hardware) where programmers will 
> have to understand how to stand in the same place at the same time, 
> so to speak, and relegate portions of programs, in a way different 
> than used in today's clusters.

Yes.  In times past I have developed heavily threaded software from 
scratch which works great on multicore/multiple CPUs.  Unfortunately, 
large legacy applications like GraphicsMagick are very difficult to 
adapt to take advantage of multiple CPUs to accellerate individual 
tasks even though the benefit is quite high.  Multitasking is much 
easier to accomplish than accellerating one task via mutithreading.

Bob
======================================
Bob Friesenhahn
bfriesen at simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/