64 computer rendering farm?

[medeamajic]

I do not have a render far but I am curious if some one could give me bench marks or advise.

If I have one 3.2 GHZ Celeron systems with 512 MB of RAM and it renders one frame at 1900 X 1080 in 8 hours, with several lights, glass objects, chrome objects and figures with dynamic hair, how long would it take 64 computers to do this?

Let's say all 64 computers have a 3.2 Celeron and 512 of RAM. Does it work like this?

1 computer = 8 hours
2 computers = 4 hours
4 computers = 2 hours
8 computers = 1 hour
16 computers = .5 hours
32 computers = .25 hours
64 computers = .125


Would doubling the computers only increase the speed by about 35-45%?


I am wondering if the Grid with 25 nodes is enough for Carrara to be used at the big studios like Pixar?

[Sarissi]

I don't think 25 nodes is big enough for a studio like Pixar (owned by Disney now). Shrek was done in Maya Unlimited in Redhat Linux, 93 minutes. That is 167,400 frames at 30 frames per second, and that is what was released to Theaters.

Your bottleneck would be the Ethernet connection: 100 mBits/sec is too slow. You need Gigabit ethernet for every node, and a fast switch. For home use, 5 nodes on Gigabit and a Gigabit Router with Cat 6 cabling would do.

[AWBenson]

From personal experience, I have a 2.4 Celeron laptop with 786 mb of RAM. When it is networked with a 3.2 Athon (single-core and not the cheaper Duron) with 1 gb of RAM, my rendering speed is 3x that of the laptop alone.

For by the time my laptop finishes 1 render block, the other computer has completed two.

Alot goes into a computer besides processor and ram. If you don't use more then 512 mb of ram and all machines have at least that, then that isn't a big factor to speed (RAM does have speed but it and various speedy RAM can really cost you). Processors are the most common speed improvement, and appearently the speed of Celeron verus Pentium makes for a really big difference.

Also, Pentium's Hyperthreading tends to give Carrara a 15 - 20% speed boost from what I've read from others.

Dual-Core is good, but it does share computer resources with the other processor.

Network speed becomes really important especially on starting the renders.

[medeamajic]

Thanks for the info folks. I am hip to dual Xeons and Dual core but you can buy 64 Dell Celeron computers for cheap. I would like to think if the Grid could network 500 computers it would be enough for Pixar Studios even if you use dynamic hair and cloth with reflections on glass and chrome. It sounds like doubling the computer does give a big boost.

I thought I read ILM or Pixar used 1000 Intel based computers on their network. It would be great if Eovia could offer a 1000 node Grid and let ILM or Even Pixar use Carrara for free for one full year. Who knows ILM may use Carrara for some projects.

[AWBenson]

ILM uses their own software for the most part. Relatively recently, ILM and Lucas Films (games, etc.) merged their operations together into the same city-complex. Since they are both owned by George Lucas, this isn't a problem. They developed a network bridge between compositing programs, 3D software (of various types using a universal bridge between them, so all assets are shared) and other digital mediums.

All this is done so that resources and people can be readily available across different mediums (film verse games) and totally different projects. (For example, a lamp post in King Kong is readily available for use in a game or another movie/commerical)

This is truly, high-tech stuff and will probablly become a model structure other companies will thrive to achieve.

[Sarissi]

ILM's render farm is the largest on the planet: 1,000 nodes during work hours, and 4,000 during off hours, and it is rendering 24/7.

Renderman is not an actual render engine. It is a Specification, which is why you see 'Renderman compliant' in software descriptions.

A bit of history: Back when Star Wars episode 4 was in early production, ILM was working on both Computer Hardware (building computers from scratch) and software. Two companies split off from ILM: Pixar and Silicon Graphics, Inc. (SGI). The actual Renderman spec saw its birth at ILM, as well as the mainframes that would become SGI. All subsequent development of the Renderman Spec has been by Pixar.

[Thomas]

(PRMan/Renderman IS a render engine, which is based on programming shaders. Other software work like renderman, using the same shader description. They are like "clones" of Renderman.
But you are true, it's also a technical specification
You can buy Renderman here : https://renderman.pixar.com/store/

[Sarissi]

George Lucas and his companies build their own Digital Motion Picture Cameras, which they invented, as well as the fully digital production environment that they used for Star Wars episode 3.

[medeamajic]

sarissi,

Sony worked with George Lucas to build the Digital HD camera he uses.

Silicon Graphics has been around long before Star Wars Episode 4. Alias Wave Front had Power Animator long Before Maya. I looked into getting and SGI Oynx or Indigo back in 1995. Back then SGI did not use any Intel based computers. Their systems used the MIPS CPU and Sun Micro Systems used the Spark CPU way back then. All the systems ran of a Linux/Unix type OS. SGI called their IRIX I think. Digital Inc. had a DEC Alpha Chip that was faster than any of them. All the CPUs were 64 bit.

[medeamajic]

I made a mistake about the Star Wars dates. I confused it with the newer Star Wars series instead of the older ones. So I guess they could have owned SGI at one time.

[Sarissi]

This was before SGI existed, or Pixar. These were what the groups that split off from ILM became. Reason: they were taking too much time away from SW episode 4 production. Those in depth cable tv specials explain a lot, as do the dvd special features.