Low Cpu usage while rendering with Distributed Rendering ON

[stefan_lopusny]

Hej all, have a strange issue surfacing recently.

Noticing severe cpu throttling when rendering with distributed on.
Monitoring the cpu threads as they hover around 20-40% performance for the whole duration of this render.

Once I disable the distributed on the same scene, the cpu is rendering at 100% all threads.

Running Vray 7.00.04 , 3ds max 2024, on Windows 10 & AMD 3990x

Thank you all for any thoughts!

[Morbid Angel]

Are you using progressive or bucket? Would be something to consider, the progressive probably won't be utilizing cpu 100% with dr. Also the amount of assets could affect your network, are they locally stored? That could also impact the load times for all the machines which would affect the performance.

[stefan_lopusny]

Hi, thanks for the thoughts!

Bucket rendering is being used and the render nodes seem to load quite fast and join the render.

The issue is that the local machine is rendering at very low cpu speed when the distributed is on, but at 100% when the distributed is off.

Quite strange behavior, first time seeing this :X

[Morbid Angel]

The distributed machines are pulling assets/textures from network? Can this be the bottleneck?

[stefan_lopusny]

This has never been a problem that we would know.
Yes they are pulling from network but there is very big difference. When DR is on, main machine is going relatively normally and all rest are hanging like crazy.
Not sure if this is 2 separate issues, but we did not have this before.
All the regular downtime like loading, network etc is considered.
Its actually lot faster to render low noise, 6k render on 1 machine than on DR with 10 nodes. with DR it became upwards 5 times slower.

[Morbid Angel]

I believe under DR you have sync missing assets options where you can sync the assets locally to each DR machine before rendering, it could potentially give you an idea if it is the network if you try that option and it renders with 100% load. I know you said it never happened before, but perhaps on this project you have something that loads the network too much. If you can monitor the server io during DR and see if its maxing out the throughput, it could help you trouble shoot this issue.
Could be a different issue though, but without knowing more I would only consider that to be the case. In logic of raytracing, rays must hit every part of the frame (what camera is seeing) so to render 1 bucket, the ray information must be synced between other buckets and if the DR machines are all reading textures or proxies for example at the same time and proxy is a large file or whatever, it may take a while that's why the cpu load drops. It could also be something else like subdivision, but I can't know for sure.