Setting up a small studio.
Using GPU CPU workstations for rendering Redshift 3d renders and Nuke comps.
Workstations are HPz840 dual E2690v3 daul 2080ti RTX 128GB ram
We were assuming that we would set up two slaves on each box , one with full CPU affinity and no GPU affinity and the opposite on the other. So nuke and redshift would run simultaneously on each machine.
When setting up the slaves it appears that deadline will not allow a slave to have no GPU infinity. Does it then matter if two slaves have the same GPU in their affinity?
What’s the best plan:
Slave1 - All CPU, GPU 0
Slave2 - No CPU, GPU 0,1
or
Slave1 - All CPU, GPU 0
Slave2 - No CPU, GPU 1
Also, would redshift need some CPU cores allocated to it as well?
Cheers,
Jem
Hi Jem,
If you’re running 2x slaves, you could assign one slave to a CPU pool then it would only pick up CPU (nuke) jobs. You should be able to un-check all the GPU affinity to stop the gpu (although you may have issues)
You’ll have to ensure you don’t run out of resources running this way, if Nuke’s using all the network, cpu and ram then redshift jobs may fail. Similarly if Nuke’s trying to use the GPU and it’s max’d out in RS you’ll struggle.
Don’t forget Redshift still uses the CPU for Cryptomatte etc so you could bottleneck you render time if you’re rendering nuke at the same time, you’ll probably want to reserve at least 1 core.
Hey Anthony,
That is what we tried to do and this seems like the logical answer, but deadline insists on us keeping one GPU on the CPU slave and one CPU on the GPU slave.
At the mo we are just doing what deadline wants… ! unfortunately the 840’s dont have power for another card…
We are not so worried about redshift using CPU as we are launching as a redshift job with dependencies. first job going to CPU machine…
Main thing is to make most of both CPU and GPU for both nuke and redshift…