Vray Swarm on the network how it works

[Jarek88]

I am having a little problem understanding how does really Vray swarm work. I made it work on 5 pcs in the office and it supports our workflow. Time to time though I have few people coming and asking me that their computer is suddenly slowing down. Here I have no idea if it is that I use swarm on their PCs or their computer decided to do something out of ordinary. Is there a way to check it and prove to my colleges that this is their computer especially if I use their computers for 3 days and only then they come on the 3rd day. I do not want to make them believe that it is always rendering that it slows down their computer.

Secondly when I applied large fur material in the scene it seems to make other computers on the network stop working to that point they need to be restarted.

Thank you for your comments.

Best,
J

[Peter.Chaushev]

Hello, Jarek,
You can manage each swarm render node machine through the Swarm web interface. It gives controls over which network machines (with V-Ray Swarm installed) can be used for distributed rendering and even specify the CPU usage while rendering.
See https://docs.chaosgroup.com/display/VRAYSWARM/ and https://docs.chaosgroup.com/display/...arm+Networking more specifically for further info.

Regarding the V-Ray Fur slowdown, this depends on how the scene is set up and what is the computing power of each machine. When used truly excessively, VRayFur can indeed cause a strain on a the available resources. My suggestion is to reserve VRayFur usage mostly to parts of the scene which is in focus while substituting the fur/grass in the background with a textured material.

Let me know if you have any further quesitons!

Kind regards,
Peter

[ivan.slavchev]

It'll be best if you have some agreement with your colleagues on when you can use their workstations. An option might be to leave a couple of CPU cores out of SWARM so they'll not be engaged when rendering. To do that - set the number of threads to the number of CPU cores on the machine minus the number you want to leave free. E.g. - if the CPU has 8 cores and you set the number of threads to 6 -> two cores will not be used when rendering.

[Jarek88]

Thank you for all of your comments !

That is true the FUR surface was way to large. For the next project I will minimize it to the essential.

I did follow your advice and the problematic machines run now with 6 nodes and 75% of CPU usage. Testing and patients of my colleges to follow.

One thing I would like to note here or rather share my morning testing CPU + SWARM vs CPU+GPU Hybrid rendering. Well it turns out that in my case GPU+CPU on my machine was almost 50% faster and came down to 2:52:17 min where CPU+SWARM 45 nodes was 4:50:53 min. The conclusion is obviously to use just my machine with GPU+CPU Hybrid but there is one thing that I do not understand why GPU+CPU is not rendering TOON material (please see the picture attached). Secondly in my opinion CPU+SWARM renders more sharp image (any ideas how to get GPU+CPU Hybrid to the same quality or sharpness ? How to alter number of passes ?)

Thank you for all the help !

Best,
Jarek

[ivan.slavchev]

Using SWARM with so many nodes is not effective. I most cases you'll get faster render with 20 additional nodes than with 40. That's because the time and resources needed for the scene transfer, for dividing and assembling the image and governing the whole process when using so many nodes is too great and effectively results in a slower render. You can experiment how many nodes you can add in SWARM before the render times start to increase. It depends on the scene usually, if it's very heavy it can use a couple of more nodes compared to light scenes, but normally 20-25 nodes are the maximum per frame.

You should have in mind that the render nodes will use the same engine as your workstation - CPU or GPU. Hence - you should consider if they have powerful GPUs or CPUs and use the engine that benefits you most. Also - if you use 20 machines - pick the most powerful ones.
CPU+GPU means GPU engine, where the CPU simulates another GPU.

The Toon material is not supported in GPU - https://docs.chaosgroup.com/display/...orted+Features
The absence of Toon material is probably why you assume GPU output is blurrier than CPU. It should be the same altogether as the noise threshold shares the same value between engines.
Some minor differences between GPU and CPU, however, could be expected. It is recommended to stick to one engine throughout your work process.