Czech National Team | Projekty CNT | Statistiky CNT
Distribuované výpočty CNT | SETI CNT | Einstein CNT | Asteroids
Fórum CNT | Chat CNT | Galerie CNT | BOINC

Finally, after months of work, testing, debugging and fixing we have Cycles as a supported engine on the farm. You'll get the power of Cycles through CPU rendering on our grid, with the supported feature set. To be able to upload Cycles sessions you should make sure you have the latest uploader version, which you can find at Blender bf-extensions SVN. Place it in your Blender directory, and in there to 2.62/scripts/addons/

While Cycles renderer allows to easily render with parts, make sure you don't do postprocess compositing with the node editor, as for that the old restrictions still exist.

Since Cycles works differently from Blender Internal we can chop up the workload in two ways, before we recombine it all into the final rendering. With Blender Internal, at best we could divide the work for one frame into smaller sections of that frame. These parts would then be stitched together into the final frame. Some features of Blender Internal renderer make it quite hard to do dividing into smaller parts, especially since our grid is not build on one CPU and OS architecture.

Apart from chopping a frame into physical parts we can also divide the workload of calculating the samples. Lets say we have a session we want to render at 100 samples. We can divide the workload over 4 work units each of which will result in a 25-sample rendering.

By ensuring that each part is rendered with a different seed we can ensure that we don't get 4 times the same result, but instead get what we want. These subsample results we can all add together and average out so that in the end we have a 100-sample rendition of the frame.

The nice thing is that we can do both ways of dividing the calculation work at the same time. So we can divide the frame into 4 parts and each part into 4 subsample frame work units. The stitcher will take care of placing the result parts into the end frame and average out as needed.

What this means is that we can now also do high-sample renders for still, since the workload can be divided over many clients. For a 10.000 sample still we could divide the workload into 100 workunits each covering 100 samples. For huge renderers we'll probably also render it in parts, just to make sure that workunits won't be too long.

When Cycles supports the GPU properly through OpenCL, also for AMD/ATi cards, then we'll investigate how easily we can roll that out on our farm. That said, it is obvious that CPU rendering will allow us to run much larger scenes, since GPU RAM is still a very scarce commodity.

As final note, the CPU-support is restricted to the "supported feature set", so bear that in mind while preparing and sending your sessions for rendering on our farm.

Enjoy Cycles!

I'm happy to say that this year I'll be representing Renderfarm.fi at both the Game Developers Conference (San Francisco, California) and South by Southwest (Austin, Texas). In case you're heading to either of these two great venues, please do seek me out! At GDC you can find me best at the Finnish section of the Nordic Pavilion (North Hall, booth 1824), while at SXSW you can find me best from the stand of the Finnish Mobile Association (stands 1123, 1125). In case you do get to either venue you can pick up your copy of the free and now classic Renderfarm.fi poster "Join the Blender Army", proven to also be hanging on the walls of great studios like the Blender Institute and Energia:

We'll be having a get together for Blender users at the Game Developers Conference on Thursday the 8th of March. Let's meet up at around 2PM at the Nordic Pavilion (North Hall, booth 1824) and move on from there if necessary. Similar meet up could be held at SXSW if enough Blenderheads show an interest. I know of at least two IRC #blender regulars who will be attending GDC with me, so the meeting's going to happen in any case. If you're there or even if you aren't but know of some Blenderhead who will - let them know!