Use this file to discover all available pages before exploring further.
ErsatzTV delegates all transcoding to FFmpeg, and FFmpeg can offload the heavy lifting of video encoding to a GPU or dedicated media-processing unit. Enabling hardware acceleration dramatically reduces CPU usage, allowing a single host to transcode more simultaneous streams at higher resolutions. ErsatzTV Legacy exposes seven hardware acceleration backends through the HardwareAccelerationKind enum; the right choice depends on your GPU and operating system.
You must use the correct Docker image variant for your GPU type. Using a CPU-only image on an NVIDIA host will silently fall back to software transcoding even when NVENC is selected in your FFmpeg profile. Match the image tag to your hardware.
Open Settings → FFmpeg Profiles, create or edit a profile, and choose the appropriate Hardware Acceleration option from the dropdown. See the FFmpeg Profiles page for the full set of profile fields.
NVENC (NVIDIA)
QSV (Intel)
VAAPI (Linux)
AMF (AMD)
VideoToolbox (macOS)
RKMPP / V4L2M2M
NVENC uses NVIDIA’s dedicated video encoding hardware present on GeForce (GTX 750+), Quadro, and Tesla GPUs. NVDEC handles hardware-accelerated decoding on the same card.Requirements:
NVIDIA driver 418.30 or newer
Docker: NVIDIA Container Toolkit installed on the host
Set VideoPreset to llhq (low-latency high quality) or llhp (low-latency high performance) for H.264 and HEVC.
NVENC has a hardware session limit (typically 3–5 concurrent encodes on consumer GPUs). Exceeding this causes a fallback error; use an Unlimited Encode Sessions patched driver or a Quadro/Tesla card for more simultaneous channels.
PadMode can be set to HardwareIfPossible to keep the padding filter on the GPU.
Supported formats with NVENC:
VideoFormat
BitDepth
Notes
H264
8-bit
Widely supported
H264
10-bit
Falls back to libx264 (software)
Hevc
8-bit or 10-bit
Requires Maxwell Gen2+ GPU
Intel Quick Sync Video (QSV) is built into Intel CPUs with integrated graphics from Sandy Bridge onwards. It covers both encoding (hardware encoder) and decoding (hardware decoder).Requirements:
Intel CPU with iGPU (6th gen Skylake or newer recommended for best quality)
intel-media-driver (iHD) or libva-intel-driver (i965) installed in the container
/dev/dri/renderD128 (or equivalent) accessible inside the container
HardwareAcceleration = Qsv
Profile fields for QSV:
Field
Recommended value
VaapiDriver
iHD for Skylake+; i965 for Haswell and older
VaapiDevice
/dev/dri/renderD128
QsvExtraHardwareFrames
64 (default); increase to 128 if you see surface allocation errors
VideoPreset
veryfast
Tuning QsvExtraHardwareFrames:QSV allocates a pool of hardware surfaces for the filter graph. Complex pipelines (scaling + deinterlace + tonemap) require more surfaces than simple passthrough. If FFmpeg logs contain messages like not enough surfaces or Failed to allocate frame, increase QsvExtraHardwareFrames from the default 64 to 128 or 256.
VA-API (Video Acceleration API) is the standard Linux hardware video acceleration interface supported by Intel, AMD (Mesa RadeonSI), and certain NVIDIA configurations with the Nouveau driver.Requirements:
libva and the appropriate backend driver installed on the host
/dev/dri/renderD128 accessible inside the container
Select the correct driver for your GPU (see table below)
VAAPI driver selection:
VaapiDriver value
Target hardware
Notes
Default
System default
Let the system decide
iHD
Intel Skylake (6th gen) and newer
Uses Intel Media Driver
i965
Intel Haswell (4th gen) and older
Uses legacy VA-API driver
RadeonSI
AMD GPUs (GCN+)
Mesa RadeonSI — stable for encoding
Nouveau
NVIDIA (open driver)
Limited encode support
Verify VAAPI is working:
# On the Docker hostvainfo --display drm --device /dev/dri/renderD128
The output should list VAEntrypointEncSlice for at least one profile (e.g. VAProfileH264High).
AMD Advanced Media Framework (AMF) provides hardware-accelerated encoding on AMD Radeon GPUs (GCN architecture and newer) via the AMF SDK. On Linux, AMF uses VAAPI under the hood; on Windows it uses DirectX.Requirements:
AMD GPU with VCE/VCN hardware encoder
HardwareAcceleration = Amf
On Linux: same device mapping as VAAPI (/dev/dri/renderD128)
Notes:
AMF H.264 and HEVC encoding is supported.
The AMF encoder on Linux is accessed through the h264_amf / hevc_amf FFmpeg encoders, which require the Radeon Software or ROCm drivers.
Quality presets are not exposed through ErsatzTV’s VideoPreset field for AMF; leave VideoPreset unset.
Apple VideoToolbox is the hardware video codec framework on macOS and iOS. It provides access to the hardware media engines present in Apple Silicon and Intel Mac GPUs.Requirements:
macOS 10.8 or newer
HardwareAcceleration = VideoToolbox
No additional driver installation is needed — VideoToolbox is part of macOS
Notes:
Supported output formats: H264 and Hevc (both 8-bit and 10-bit on Apple Silicon).
ErsatzTV is typically run on macOS without Docker when using VideoToolbox; the Docker VAAPI compose file does not apply.
VideoPreset is not meaningful for VideoToolbox — leave it unset.
Apple Silicon M1/M2/M3 chips contain multiple hardware encode/decode engines; simultaneous stream limits are much higher than NVENC on consumer NVIDIA GPUs.
These two backends target resource-constrained single-board computers.RKMPP — Rockchip Media Process PlatformUsed on Rockchip SBCs such as those with the RK3399, RK3568, and RK3588 SoCs.
Set HardwareAcceleration = Rkmpp
Requires the Rockchip kernel MPP driver and the librockchip-mpp library inside the container
The /dev/mpp_service or /dev/video* devices must be mapped into the container
Supports H.264 and HEVC encoding on compatible chips
V4L2M2M — Video4Linux2 Memory-to-MemoryUsed primarily on Raspberry Pi (BCM2835/BCM2711 hardware encoder) and other embedded Linux platforms that expose a V4L2 codec device.
Set HardwareAcceleration = V4l2m2m
The /dev/video10 (or similar) codec device must be accessible inside the container
Raspberry Pi OS includes the required bcm2835-codec driver by default
Supported formats depend on the SoC; Raspberry Pi 4 supports H.264 encode/decode
Both RKMPP and V4L2M2M have fewer quality controls than NVENC or QSV. VideoPreset and AllowBFrames have no effect on these backends. Stick to 720p or lower resolutions for reliable real-time transcoding on single-board computers.
When HardwareAcceleration is set to None, ErsatzTV uses FFmpeg’s software (libx264 / libx265) encoders entirely on the CPU. This is the safest and most compatible option but consumes significantly more CPU resources. Use ThreadCount to limit the number of CPU threads if ErsatzTV is sharing the host with other workloads (set to 0 for automatic).
A single modern CPU core can sustain roughly 2–4 simultaneous software transcodes at 720p H.264. For more channels or 1080p output, hardware acceleration is strongly recommended.
FFmpeg exits immediately with 'No such file or directory' for /dev/dri/renderD128
The render device is not mapped into the container. Add the devices section to your docker-compose.yml as shown in the VAAPI and QSV tabs above.
NVENC error: 'OpenEncodeSessionEx failed: out of memory'
You have exceeded the hardware session limit on a consumer NVIDIA GPU. Reduce the number of simultaneously active channels, or apply the nvidia-patch to remove the session cap.
QSV: 'not enough surfaces' in FFmpeg log
Increase QsvExtraHardwareFrames in the FFmpeg profile to 128 or 256.
VAAPI produces green or corrupted output
Verify that the correct VaapiDriver is selected for your GPU generation. Intel Skylake+ requires iHD; older Intel hardware requires i965. Mixing them causes colour space errors.
FFmpeg Profiles
Full reference for all FFmpeg profile fields.
Streaming Modes
Choose between TS, HLS Segmenter, and HLS Direct output.
