Set up hardware encoding
For: household admins with an Intel, AMD or NVIDIA GPU on a Linux host. You’ll get: the GPU passed to Loomarr, so it can play more channels at once, including 4K HDR.
It’s optional. Without a GPU, Loomarr encodes in software: everything works, you just run fewer channels at once. This only applies when Loomarr does the streaming (the default); on the Tunarr backend, Tunarr’s own transcode settings apply. Docker Desktop on macOS can’t pass the Mac GPU through, so there Loomarr always encodes in software.
Which GPUs do what, including HDR tone mapping, is in the hardware reference.
Intel and AMD
AMD support is best-effort: we don’t test AMD GPUs before a release.
VAAPI, QSV and Vulkan all reach the GPU through /dev/dri. Pass it in:
PLAYOUT_RENDER_DEVICE=/dev/dri docker compose -f docker/compose.yaml --profile sqlite up -dLeaving it unset is fine: the container starts normally on a host with no GPU.
More than one GPU
Loomarr probes the render node /dev/dri/renderD128 by default. With more than one GPU, such as an
Intel Arc beside the CPU’s integrated graphics, or an Arc beside an NVIDIA card, the one you want is
often renderD129 or higher. A single-GPU host can skip this section.
-
Find the node that can encode. It reports
VAEntrypointEncSlice…entrypoints:Terminal window ls -l /dev/dri/by-path/Terminal window vainfo --display drm --device /dev/dri/renderD129 -
Point Loomarr at it:
Terminal window PLAYOUT_RENDER_NODE=/dev/dri/renderD129 \PLAYOUT_RENDER_DEVICE=/dev/dri \docker compose -f docker/compose.yaml --profile sqlite up -d
NVIDIA
Loomarr needs NVIDIA driver 610 or newer on the host. Check with nvidia-smi. An older driver
still runs Loomarr, but without NVENC, so playout encodes on the CPU. The 580 branch that many
distributions and cloud images ship is too old. NVIDIA’s repository names newer branches without a
number, as nvidia-open or nvidia-headless-open.
Install the NVIDIA container toolkit on the host, then add the NVIDIA overlay:
docker compose -f docker/compose.yaml -f docker/compose.nvidia.yaml --profile sqlite up -dThe overlay asks for capabilities: [gpu, video]. Keep both: with only gpu, the container sees
the card but every NVENC trial fails.
Check it worked
At boot, Loomarr trial-encodes with every encoder it finds and keeps the ones that produce output.
Leave PLAYOUT_ENCODER empty so that measurement stands. The boot log names every encoder family
that passed:
docker logs loomarr 2>&1 | grep -i 'encoder\|capability'If the encoder you expect is missing, check the device or overlay above; on a multi-GPU host, the
render node. On NVIDIA, a message naming “the minimum required Nvidia driver” means the host
driver is too old. scripts/playout-diag.sh gives a fuller read-only snapshot: ffmpeg processes per
channel, GPU state, and whether each airing is direct-playing or transcoding.
Choose an HDR tone curve
An HDR film on an SDR channel is tone-mapped. The default curve, Hable, is the cheapest on every GPU. To change it, set HDR tone curve in Settings → Playback; streams that start afterwards use the new curve. The reference lists what each curve costs on each GPU.
Channels at once
Loomarr sets how many channels can transcode at once from what its boot trial measured. Leave
PLAYOUT_MAX_CHANNELS at 0 (automatic). Set a lower number only if complex content needs more
headroom; it can never raise the limit above what the trial proved.