#Kubernetes Setup with nvidia gpu

1 messages · Page 1 of 1 (latest)

digital mesa
#

Hi all,
I was wondering if anyone has been able to deploy ersatztv using kubernetes and using an nvidia gpu. I think transcoding is working correctly but when it comes to tone mapping for hdr content everything keeps going to cpu. Pretty sure I had this working correctly on the same machine previously using Ubuntu server and Linux install option. Now running same machine as a node in my homelab k8s cluster using talos Linux.

silver osprey
#

I think HDR with Nvidia requires vulkan

zenith geyser
#

I can use HDR features in k8s on Jellyfin but with ETV I know it's different

soft loomBOT
#

@digital mesa
Automoderator:
directly mentioning devs is not allowed - please open a #1006572433039433830 thread to receive help

digital mesa
digital mesa
#

I'm quite out of my depth here but running it in standalone docker seems to have the same issue. anything I can do get this working?

#

Native Linux install:

Pipeline: TonemapCudaFilter + libplacebo + vulkan=vk@nv + hwaccel vulkan + EncoderH264Nvenc
Fully on the GPU, no CPU involvement

Docker:

Pipeline: CudaSoftwareFallbackUploadFilter + ScaleCudaFilter + TonemapFilter + EncoderH264Nvenc
CUDA handles decode/encode on GPU but tonemapping falls back to CPU, causing high CPU utilization

silver osprey