Progressive Mode - Understanding Ray Bundle Size

[alexyork]

Hi folks,

Vlado has mentioned that in order to maximise DBR efficiency you need to give your nodes enough work to do, so the best way to do this is to increase the Ray Bundle Size value.

Two questions:

1) raise it to what? what's a safe boundary? 256? 512? 1024?
2) the default is 128 - if we're only using 1 local machine for rendering, are we better off lowering this? to 64? 32? 16?

Some guide for the values' top and bottom ends would be really helpful to understand this. The amount of time I have for experimentation is extremely limited these days.

Thanks,

[Dariusz Makowski (Dadal)]

Get Slave monitor software like ReFaMo.

Then start DBR and adjust values untill all nodes run at 100%

The lower the CPU ray bundle size the faster it updates
the bigger the CPU rays per pixel the more nodes will stay at 100% cpu

[alexyork]

Get Slave monitor software like ReFaMo.

Then start DBR and adjust values untill all nodes run at 100%

The lower the CPU ray bundle size the faster it updates
the bigger the CPU rays per pixel the more nodes will stay at 100% cpu

Thanks - that all makes sense, and it's what I've been doing for DBR renders, but what about for non-DBR renders? I feel like, logically, reducing that value should be better?

[Dariusz Makowski (Dadal)]

Well it really depends what you are doing with it I think. If you are doing shader/light tests I tend to put CPU ray bundle to something like 4 so I get as fast updates as my patience allowed me to wait... If I do more final render I let it increase as then I get more render power as each update of RT window is a sacrifice of render power.

[vlado]

It is very similar to the bucket size for the production renderer - it determines the size of the image chunk computed by each CPU thread. The same rules apply - smaller values might get you faster previews, but require more frequent communication between the master and the slaves. Larger values allow to utilize the CPUs on the render slaves better but may slow down the updates.

Best regards,
Vlado

[alexyork]

It is very similar to the bucket size for the production renderer - it determines the size of the image chunk computed by each CPU thread. The same rules apply - smaller values might get you faster previews, but require more frequent communication between the master and the slaves. Larger values allow to utilize the CPUs on the render slaves better but may slow down the updates.

Best regards,
Vlado

Thanks Vlado - so my query about using only the local machine still stands - are we better off using sub-128 values in this case? Like 64 or 32? Or is 128 a kind of "sweet spot" for local renders?