NICE re/streamer pro

Guides

GPU transcoding (NVIDIA NVENC) and ffmpeg builds

With an NVIDIA card the panel can transcode channels on the GPU (NVENC) instead of the CPU – many more channels per server. It only needs the driver and an ffmpeg build with NVENC.

The ffmpeg builds

The panel keeps several ffmpeg builds in /nice/ffmpeg.d and uses the active one:

BuildGPUSystems
static-6.0.1no (CPU only)any – it is the standard one
static-gpuNVIDIA NVENCold systems (Ubuntu 16.04 …) with the NVIDIA driver and the CUDA 8 libraries installed
Jellyfin ffmpegNVIDIA NVENC, Intel QSV, AMD AMF, VAAPIglibc 2.28+: Ubuntu 20.04+, Debian 10+ – downloaded from the panel with one button

1Install the NVIDIA driver

Install the driver from your distribution and check that the card is seen:

nvidia-smi

It should list the card and the driver version. Newer ffmpeg builds need a recent driver.

2Choose a build with NVENC

Configuration → FFMPEG. Every build shows its versions, whether it is static or needs system libraries, and its GPU encoders (NVENC in green). A build that does not start on this system says why – for example the missing libraries.

Configuration → FFMPEG: Jellyfin build with NVENC/QSV/VAAPI encoders and the active static build

From the command line the same: /nice/restream.sh ffmpeg list, /nice/restream.sh ffmpeg <build>. The top of the section shows the NVIDIA card the panel found.

3Transcode a channel on the GPU

In the channel form: Stream Mode → [ffmpeg] dynamic or static, then in FFmpeg command:

Test runs the settings on the source for a few seconds before you save. A changed build or setting applies to each channel from its next start.

How much does the card take?

Tools → FFmpeg Stats shows every running ffmpeg with its CPU and memory; nvidia-smi shows the load of the card and its encoder sessions. GeForce cards limit the number of simultaneous NVENC sessions; Quadro/Tesla do not.