Hi,
I'm looking to set up a workflow where I work off my Macbook Pro and then network render with Octane using a windows renderslave. My original plan had been to get an eGPU for the MacBook Pro and use that in conjunction with the 4 GPUS in my renderslave using network render as seen in this tutorial https://www.youtube.com/watch?v=Vvf_-toAOU8 to have a total of 5 GPUS rendering in IPR and etc
However, I was wondering if the eGPU is actually necessary to do this? For instance would it be doable to have a setup where you have a MacBook Pro with the correct CUDA drivers installed etc, no actual eGPU, and then network render using a Renderslave that also has Octane?
Thanks for your time and help!
Network render without eGPU
Moderators: ChrisHekman, aoktar
As far as I can tell from my own attempts under 10.12.6, no, this won't work. I think what's happening is that the OS won't load the CUDA and/or NVIDIA driver (not sure which is the actual requirement, maybe both) unless an appropriate GPU is detected on boot/log-in, and Octane won't start a render without the driver, even when multiple network slave GPUs are already "bound" in the Octane Network Preferences window. An Octane scene can be opened and edited, but any attempt to render results in the error, "CINEMA 4D Studio: There is no CUDA device which is supported by Octane Render." Maybe things are different with more recent OS updates, but that seems unlikely.
It would be pretty sweet, though, and I'd love to be proven wrong. Who knows, maybe Ahmet can do some of his magic and make it happen? It would be wonderful if it was just a matter of hardware check in the code that runs before a render which could be toggled off.
Edit: I was thinking about this some more on my drive home last night. I suspect that Octane needs some of the actual code in the CUDA drivers, but not long ago in the 4.0 release cycle, OTOY uncoupled part of that code from what's in the distributed NVIDIA drivers. You see it now as a separate download option in part of the installation process. I wonder if the new, directly supplied by OTOY code should be enough to run the renderer, at least as far as coordinating the job with remote GPUs?
There would undoubtably be a bit more of responsiveness hit to rendering this way, and you'd definitely want at least gigabit ethernet, but this could be a real game-changer for a lot of people and businesses. Already, network rendering with Octane 4 has taken up some of the load we normally manage with a much more complicated Thinkbox Deadline system running across three OSes, local workstations, local render machines, and cloud render nodes.
It would be pretty sweet, though, and I'd love to be proven wrong. Who knows, maybe Ahmet can do some of his magic and make it happen? It would be wonderful if it was just a matter of hardware check in the code that runs before a render which could be toggled off.
Edit: I was thinking about this some more on my drive home last night. I suspect that Octane needs some of the actual code in the CUDA drivers, but not long ago in the 4.0 release cycle, OTOY uncoupled part of that code from what's in the distributed NVIDIA drivers. You see it now as a separate download option in part of the installation process. I wonder if the new, directly supplied by OTOY code should be enough to run the renderer, at least as far as coordinating the job with remote GPUs?
There would undoubtably be a bit more of responsiveness hit to rendering this way, and you'd definitely want at least gigabit ethernet, but this could be a real game-changer for a lot of people and businesses. Already, network rendering with Octane 4 has taken up some of the load we normally manage with a much more complicated Thinkbox Deadline system running across three OSes, local workstations, local render machines, and cloud render nodes.
Animation Technical Director - Washington DC