Building my first render node (vray swarm)

[jasiu.mk]

Hello
I'm planning to build first render node in my small company (5 people). Currently we are working on 3ds max and Vray (1.5), but we want to switch to SketchUp or Rhino with Vray.
So the render node would be build on probably the newest AMD Threadripper 2990WX (32 cores) CPU, but I'm also considering GPU rendering (honestly don't know how to compare the two).
I would be very glad to get anserws to some of my concers:

1. How Vray SWARM works in terms of queuing task? If 3 people would try to render at once with render node will someone have to wait for the first one rendering to finish or SWARM will distribute CPU power of node among all the task at once?

2. How much RAM does my render node need? How should I calculate it considering we are working with projects that take less than 16 GB on average? If I throw 3 projects that take 16GB of RAM does it mean that station should have at least 46GB on render node to render them at once?
What is more important for render node, ECC memory support or pure RAM speed?

3. GPU rendering - does it scale 100% with every card I add to it? If I throw 6 GPUS instead of 3 should my render speed cut in half?

[Peter.Chaushev]

Hello, Jasiu,

1. Currently, V-Ray Swarm does not automatically redistribute the load when multiple users submit jobs to the same machines. First one to engage a node keeps it for their rendering process. There is, however, a way to limit the usage by adjusting (even during rendering) the Swarm Goal slider to a certain value (e.g. 35% of all available machines) or using tags for fast setup of different node combinations (e.g. "3rd_floor", "2nd_shift", "render_farm").
2. Each machine engaged in assisting the render process has to load the whole scene and all of its assets. Therefore, memory usage equals that of the local machine. Keep in mind, however, that certain phases of the process are not being distributed (Light Cache calculation, denoising).
3. Regarding GPU rendering, take a look at https://www.chaosgroup.com/blog/guide-to-gpu-rendering

Keep in mind rendering does not scale at 100% with additional machines - assets take time to be distributed and loaded on additional render node machines, LC not distributable (but can be cached as file and loaded), GPU memory not combined.
For hardware advice and comparison, I strongly recommend exploring the user test results at https://www.chaosgroup.com/vray/benchmark
You may also find interesting the following V-Ray for Rhino video. Swarm functionality is practically identical between V-Ray for SketchUp and Rhino:
https://docs.chaosgroup.com/display/...arm+QuickStart

Kind regards,
Peter