Thinkbox Deadline for single machine?

[art3mis]

Hi

Saw this product recommended by another Houdini user. Trying to understand what advantage, if any , it provides if you only have a single machine to work and render on?

http://docs.thinkboxsoftware.com/products/deadline/9.0/1_User Manual/manual/app-houdini.html#app-houdini-faq-ref-label

[ikoon]

My newbie knowledge:

- If rendering with Redshift, with Deadline you can set the GPU affinity (one frame : one gpu). It is faster then (one frame : multi gpu). With Redshift 4 GPUs are just 3.5x single GPU performance. https://www.redshift3d.com/forums/viewthread/12149/
- Probably you may benefit from running multiple instances of Houdini on one machine, if your network is single threaded. You can distribute ROPs with Deadline. You may run out or RAM, so this is probably just for some special cases.

Also, It would be great to split OpenCL calculations like this. Houdini uses only one gpu. But I have investigated and the Deadline is not able to distribute OpenCL. Please somebody correct me if I am wrong.

[LaidlawFX]

10 hours ago, art3mis said:

Trying to understand what advantage, if any , it provides if you only have a single machine to work and render on?

IMO, it really comes down to how much rendering you are doing and how much time you have to babysit. If you're only rendering a few short sequences in a day, and if you are only working in Houdini I would say use Hqueue as it is free, or just run the nodes in a series. However, if you are going to use your one machine 24/7 to render from multiple programs, with multiple hardware needs per the computation type. Any farm would be better and reduce your overall workload and gain you higher efficiency. i.e. all the time you spend queuing up the next render. You can run that math in excel to see how much time/cost it saves, by taking a previous project and counting all the layers, programs, and time per frame it cost. You can even find the time between jobs it took to run between the last frame written to the next frame written, either with your previous render farm, or how much time it took you to press render on the next sequence. This time allocation can even go into how late you had to stay up to be your own render wrangler, testing for failed frames. No render farm will save you if you end up just rendering a bad sequence, but all the other allocation issues you won't have to worry about. 

IMO, Overall deadline is a good farm. I enjoyed working with it and I would recommend it. But there are hundreds of farms out there. We had another thread going about this.