I have several clients with 8x GPU boxes, a lot of them use it in two different configurations
1x Slave for all 8x cards (GPUBOX_MAIN)
4x Slaves with 2x cards each (GPUBOX_CARDS_nn)
This is done for a couple of reasons, either to share the node up between more users, or for Redshift as it doesn’t scale across the cards

The 5 slaves would look like this
GPUBOX_MAIN
GPUBOX_CARDS_01
GPUBOX_CARDS_23
GPUBOX_CARDS_45
GPUBOX_CARDS_67

I find several clients having the same problem, they want an easy way to switch between the two modes of operation,

for example, rendering on 1x slave and 8 cards using VrayDBR to be interruptible by a 3dsMax / Vray job to be rendered across 4 slaves with 2x GPU’s each.
Or an Octane job using 1x slave and 8 cards, then a RedShift job using 4x slaves with cards each

I can’t find a way for Deadline to flip between the two modes. I tried to submit the DBR job as interruptible but when the Vray job follows it isn’t interrupting the job as it’s running on a different slave.

The only way to change them is to manually disable the group or main slave.

Is there a possible workaround to this? Would I need to create an ‘event’ that checked whether the ‘master’ slave was running or the other slaves? Or am I missing a much easier solution to this?

Thanks
Ant

For GPU flexibility, we built into the GPU affinity as well as the number of GPUs per task. Needing to switch is definitely difficult with a certain number of Slaves per GPU.

Historically, the design of the multi-Slave feature was to support things like different NUMA nodes or GPU vs CPU jobs.

I think right now, if users aren’t making use of the GPUs per task feature in the supported app (Maya, C4D, etc) we need to find out why and fix it for them, and then create a blog post or documentation page… Ant, have you tried this with clients? Were there issues?

docs.thinkboxsoftware.com/produ … u-affinity
docs.thinkboxsoftware.com/produ … u-affinity
docs.thinkboxsoftware.com/produ … es-rollout