I am preparing our studio for the 6 months working on the full feature movie. The main rendering engine will be Cycles.
We are using Debian 7.6 Linux at studio, and Blender is our main 3D software. For now I am buying workstations for 3d artists.
I also have about 10000$ in out budget for the rendering solution. For now I am planing to buy ten gtx 880 with 8GB on each card when they will be released, and use GPUbox as render manager.
But I also thinking about cloud rendering as another solution. Maybe it is much better way? Who has the experience in Cycles GPU cloud rendering? Maybe you can share some goals and cones?
GPUBox Artist is already a cloud rendering solution that can be deployed in a private network, it is not render manager
It separates the GPUs from the host layer, so the devices are seen by the application (i.e. Blender) as if they were mounted in the used workstation. That is the core idea of GPU cloud computing.
For 10000$ you can set up a pretty nice rig. And when it comes to Blender - GPU cloud rendering on Cycles works really well and unvails lots of new possibilites (such as sharing the GPUs), but you need a decent network to connect the nodes. 1 Gb/s is not enough if you do not want to lose the performance of your GPUs.
When it comes to cloud rendering without using your own GPUs, you can consider GPUBox Web Service (http://www.renegatt.com/webservice.php - it will be launched soon) which basically is renting GPUBox Artist license, deploying it on Amazon EC2 instances with GRIDs K520 or Teslas M2050 and using them on a virtual system in web browser.
However, in your case building a private cloud seems to be a better solution. If there are a few artist working on this project you could make a really good use of sharing the GPUs. It could enhance your workflow during the project, big time…
If you do not have problems with distance between your workstations and servers you can consider InfiniBand (even second-hand). The cabling can be pretty expensive, but all in all it is pretty cost-effective compared to 10Gb Ethernet and also very impressive performance-wise (even middle-end InfiniBand configurations can let you run on 20Gb/s).
I am looking for some data/links that confirm that you can build a GPU blender farm across multiple local machines. I know I can do this and go to CPU render farm, but with what release of blender did GPU farm become an option? Thanks.
Well, on Renegatt Software Website you can find a lot about building a private GPU cloud computing environment. That is what we spoke here in this topic.
