At 03:59 PM 5/19/2006, philip wrote:
Thanks guys. Your answers were exactly what I was looking for. Micro$oft
would be the obvious and easy solution, but I'm very inclined to push the
linux route just for the satisfaction of knowing that I was one of the
people who took a traditionally windows project and used linux instead.
This is fine, if you're spending your own money, or if you can show that
the costs will be lower.
Considering that the 3D drawing packages pretty much have the ability to
run rendering farms out of the box (on Windows), you'd have to trade off
the cost of the multiunit license and the incremental number of processors
you'll need to overcome the performance hit from windows against the cost
of the labor to do the linux based cluster. (a quick google shows that
3dstudio Max 8 comes with an unlimited processor license, so all you'll
need to deal with is the Windows licensing. But hey, didn't your PCs come
with WinXP already on them?)
For that matter, there are many companies out there offering "render farm"
services for 3D Studio, etc. It IS something that lends itself to a
service bureau type business model.
The other reasoning is as was mentioned, the processing power of linux over
windows seems more suited for this type of work anyway. I also agree that
the opeteron route would make more sense.
At the core, the actual computations are being done on the same processor,
so the raw performance would be mostly the same. The question is how
efficient is the rendering process itself (that is, how many compute cycles
are spent on "rendering" vs how many cycles are spent on "OS
overhead") I'd be real surprised if the commercial products spent less
than 80% of their processor cycles on the raw rendering computations. It's
a compute intensive task, not relying on OS services, doesn't have much
I/O, and there's nothing else going on on the PC. So, the advantage from
Linux over Microsoft is going to be in that 20% of OS overhead. Even if
Linux is twice as fast as Windows (unlikely) you've only got a 10% speedup.
If it takes you one person year's work to port the rendering application
over (about $250K, fully burdened), you've got to have a truly huge
rendering farm to make this worthwhile. 1000 processors at $2K/each
As for max/autodesk, they're not for linux. I know that much. I'm more
interested in a method of deconstructing a max file into it's basic form of
being a state of geometry and settings, and then finding a linux engine that
can perform the rendering task, so that I could use something along the
lines of drqueue to simply submit the job to the cluster.
Is the file format published?
Do the consumers of your rendering process depend on peculiar features of
3DS max that are a pain to render?
Jim
_______________________________________________
Beowulf mailing list, [email protected]
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
