Open Funtime60 opened 6 months ago
Simple way to do this is to run 2 Node Docker containers and assign a GPU to each with the Docker container settings, that way they're kept independent. Doing it with a single Node is a bit trickier, the 'Set Video Encoder' flow plugin should auto select the GPU with least utilisation but does not work in all cases.
The issue is that this was in response to an issue on a native Windows node. I don't know if you can passthrough on docker desktop but even if you can it's not a good solution.
As for the auto GPU selection on a single node, I don't recall it ever touching the second GPU, but I'm probably wrong.
It still doesn't help with health checks.
On Wed, Feb 28, 2024, 6:24 PM HaveAGitGat @.***> wrote:
Simple way to do this is to run 2 Node Docker containers and assign a GPU to each with the Docker container settings, that way they're kept independent. Doing it with a single Node is a bit trickier, the 'Set Video Encoder' flow plugin should auto select the GPU with least utilisation but does not work in all cases.
— Reply to this email directly, view it on GitHub https://github.com/HaveAGitGat/Tdarr/issues/944#issuecomment-1970210153, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF3U3U32DR6PQ2PNEL5YN3TYV2BMBAVCNFSM6AAAAABBQ76A5GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZQGIYTAMJVGM . You are receiving this because you authored the thread.Message ID: @.***>
You could try using the Custom Arguments
plugin and then put -hwaccel_device 0 -gpu 0
for one node and -hwaccel_device 1 -gpu 1
.
You can check the node details using Check Flow Variable
and checking {{{args.deps.configVars.config.nodeID}}}
or {{{args.deps.configVars.config.nodeName}}}
Is your feature request related to a problem? Please describe. While I see that multi gpu support was added for classic plugins, it either isn't available or isn't obvious with the new flows.
Describe the solution you'd like I would like the ability to set the ffmpeg GPU selection parameter based the node. That way I could just run two nodes on my dual GPU machine and set one GPU to each node.
Describe alternatives you've considered I haven't considered any other options as it's late at night and I cannot think on demand.
Additional context Potentially this could be expanded to allow for more variables to be stored/used per nodes.