Separate names with a comma.
Discussion in 'Mac Pro' started by sparkie7, Jun 18, 2009.
Like for faster 3D rendering and graphics and the like?
Whats involved if this is possible.. TIA
To be brief, no.
Many 3D applications have this functionality built in. Like the 999 free render nodes in LightWave3D's ScreamerNet and the 30 or so ScreamerNet alternatives that are available both for free and commercially - like LightNET, Spider, RenderFarmer, and Butterfly NET. Some applications have limited nodes available while yet others have none at all. Some like Maya I believe charge the big bucks per node
So for 3D rendering anyway it all just depends on the software. Search google or yahoo for "renderfarm", "renderfarm setup", or "renderfarm tutorial". And yes it's TONS faster!!! (depending on the number and speed of the machines you're farming out the renders to)
It's also available in applications like Adobe AfterFX and maybe Motion or Shake too. At least the PC only software I use for it has that built in. - I use eyeon's Digital Fusion (now just called Fusion).
To be one letter less brief, yes.
Why the heck "no"? Ever heard of a render farm?
thanks guys. thats the word I couldn't remember. Thats what happens when you get to bed at 4am and sleep deprived
It is all application dependent. Some, like Pro tools, have distributed cluster rendering (Compressor is a great example of this). Other tools like Cinema 4D have Net Render Clients. You would need to consult the individual program's documentation on how to set this up. The poster above me suggested some good terms to toss into Google or Bing that should point you in the right direction
I am actually considering taking some Power Mac G5's that I am retiring from one of the labs I administer and setting them up as a render farm.
As a heads up, there is no reason to target Mac Pros specifically as machines for a render farm. If you already have a few setup on a network at an office, sure... but if you are starting from scratch you would probably get more bang for your buck building a farm of zombie computers that are optimized for that sort of work. It would be cheaper and you'd ultimately get better performance. The one thing about researching render farms that has always stuck with me is that it isn't the core clock speed that is so important but rather the number of cores you are distributing the work across. The more "brains" you have working on the project, the better the system distributes the work. Obviously taking a dozen 486's isn't going to beat out a pair of Mac Pros... but modern "bargain" workhorse/server chips are probably your best bet.
If I am completely wrong, other posters feel free to correct me
No, you're right. You can even run the farm nodes headless if you wish to. (headless is a renderfarmer's or beowulf term for a machine without display capabilities.)
EDIT: Keep in mind however that different CPUs may produce different seed values for various random algorithms. This can affect procedural texture patterns as well as procedural music/tone generation and may cause the final plates or sound clips not to match up. You should try to keep intel to intel and PPC to PPC as well as AMD to AMD.
I had thought you must be able to. With a renderfarm you are basically just talking "herding them" to each machine over the network so you probably would only need to attach peripherals to it if it dies / kernal panics and can't be resuscitated through Terminal.
Tesselator -- can you recommend any Flash video encoding application that would work in a renderfarm environment? My unfortunate experience with the encoder in CS3 (haven't upgraded yet to CS4) is that it only utilizes one core at a time
I edited my post above - have a look.
But no I don't know very much about flash. Also the apps like AfterFX (Fusion in my case) I don't think use multiple nodes for the encoding process. Just the image processing of the frames in the frame sequence(s). At least I've not seen it myself.
Which is basically yes, or if you write tend to write your own software on the Mac, or tend to use a high end Linux production video crunching app (and have a million dollars in hardware going through your business every year.)
No if you are a casual surfer and cannot reset the clock on your VCR.
Edit: as said before, the Mac tended not to be a targeted platform. So if you like cheap PC servers and Linux, you'll likely find a lot more renderfarm solutions.
XGrid would work.
Yes. GridMathematica is easy. Have seen a config with it running across 5 mac pros and 2 XServes. That's 56 kernels on 7 8-core machines. And you can even add windows and unix boxes as well....
Thank you Tesselator. I hope this post was helpful/insightful to the OP. It certainly was for me. Thank you all.
Thanks for all the useful info guys. Pity Mac OSX / SL doesn't allow for such a setup to be applied across all apps. ie 2/3/4 machines to act as one. maybe one day
also some of the apps in final cut studio can be set up with qmaster to do distributed processing for rendering/compressing.
Once software becomes optimized for large numbers of cores (with asynchronous threads) through OpenCL, it will be possible to abstract the threads to other physical machines. The main limitation is and will be the fact that the communication between two machines is drastically slower than communication internally. Software that works in clusters of threads is and will always be suited for server/render farms.
How about via the (2) SATA connections on the motherboard direct to another MP's SATA connections. Or via direct PCI to PCI expansion slots with a specialised card for this
The aught nines don't have extra SATA, but what card do you have in mind?
Most large clusters use Infiniband (http://en.wikipedia.org/wiki/InfiniBand) or something similar.
The limitation still remains even with the fastest connections available aren't anywhere near the speed between processors on the same machine. If I remember correctly, cores and processors can communicate with each other with a bandwidth in excess of a couple Tb/s. Even PCIe 2.0 can't compete with this speed.
Terabytes/sec? Doubt it..
Plus it's just the transmission of data that needs to be processed/result..nothing else really.
I think the small "b" means bits so terabits. I know 10,000-BASETX FO is pretty easy to set up and it's getting cheap now.
On GridMathematica I tend to manage this by having each core do complicated things locally and have the smallest possible exchange of information. That keeps the efficiency up. E.g, with Monte Carlo simulation each core just works out a few moments of the massive samples computed locally and sends that small packet back to a master. Get about 80% efficiency over only 1G ethernet if you can work that way. Not all calcs work like that, of course. I looked at faster comms and it was stupidly expensive, at least this time last year.
A bit off of topic but......
Unless you got thousands of dollars of multiple Mac Pro's...
I've heard that you can use a PS3 with linux and Mental Ray to build you own personal render farm...