The Wowza Streaming Engine (WSE) transcoder supports accelerated video encoding and decoding using NVIDIA graphics cards. This enables live-stream transcoding at greater scale and speed. The WSE transcoder also supports offloading transcoder video scaling to NVIDIA CUDA-based GPUs.
NVIDIA GPU-accelerated transcoding can be performed either through the legacy transcoding pipeline and NVIDIA’s native APIs, or via the MainConcept Easy Video API (EVA) framework introduced in Wowza Streaming Engine 4.9.7.
Note: When configuring NVIDIA GPUs using the legacy NVIDIA transcoding pipeline, we recommend using the NVCUVID decoder, NVENC encoder, and CUDA scaling for best performance. For the EVA transcoding pipeline, use the NVCUVID EVA decoding, CUDA EVA scaling, and NVENC EVA encoding implementations. NVIDIA acceleration workflows that mix the GPU and CPU implementation may degrade overall GPU performance.
NVIDIA GPU and driver support
Installing NVIDIA drivers for GPU-accelerated transcoding
Before using an NVIDIA GPU for transcoding, the appropriate NVIDIA display drivers must be installed your WSE server. This is required for GPU-accelerated decoding, scaling, and encoding to function correctly.
See NVIDIA's Driver Installation Guide, for OS-specific instructions. For convenience, we've included links to several of the most commonly used resources:
Notes:
- Hardware-accelerated transcoding with WSE is supported on Amazon EC2 instances launched from NVIDIA-provided AMIs, which include the required GPU drivers and kernel configurations.
- Accelerated encoding is not yet validated on virtual hardware environments such as VMware or Xen and may not be available or function as expected.
- Older NVIDIA graphics drivers may negatively impact NVENC-based video encoding performance. For best results, install the latest supported driver version.
After the driver installation is complete, see Set up and run Transcoder in Wowza Streaming Engine for information about configuration options.
EVA transcoding driver requirements
With Wowza Streaming Engine 4.9.7, we migrated to MainConcept EVA to take full advantage of its integrated NVIDIA GPU acceleration. EVA transcoding with NVIDIA GPU requires NVIDIA display drivers 572.60 (Windows x86) and 570.124.04 (Linux x86) as the minimum on the server running Wowza Streaming Engine. For more details, see the following links:
- Windows 572.60 driver details and supported products
- Linux 570.124.04 driver details and supported products
For more information, see Set up and run Transcoder in Wowza Streaming Engine.
CUDA 12
WSE 4.8.22+ supports NVIDIA GPU-accelerated video transcoding on CUDA 12 for improved performance. This supports the use of the latest NVIDIA drivers for transcoding at a greater scale and speed. The NVIDIA microarchitecture of your hardware must support CUDA 12. NVIDIA driver versions must be at least 525.60.13 for Linux and 527.41 for Windows. Any previous or unsupported drivers will cause Wowza Streaming Engine to revert to default CPU transcoding. For information about CUDA-enabled hardware, see NVIDIA CUDA GPUs.
CUDA 12 and Wowza Streaming Engine 4.8.23With Wowza Streaming Engine 4.8.23, NVIDIA users with custom encoding, decoding, and scaling properties should review their implementations for updated CUDA 12 parameters. To confirm if your custom properties and settings should be updated, reference the NVIDIA NVENC Preset Migration Guide.
CUDA 11
WSE 4.8.14+ supports NVIDIA GPU-accelerated video transcoding on CUDA 11 for improved performance. This upgrade supports the use of the latest NVIDIA drivers for transcoding at a greater scale and speed. The NVIDIA microarchitecture of your hardware must support CUDA 11, and using NVIDIA driver version 460.00 or later is required. For information about CUDA-enabled hardware, see NVIDIA CUDA GPUs.
With this upgrade, Kepler GPUs and CUDA decoding are not supported, as the Kepler architecture is now deprecated. The quickest way to identify these cards is the K at the beginning of their name, for example, k4200. For NVIDIA accelerated decoding, use the NVCUVID (also known as NVDEC) implementation instead.
CUDA 10
WSE 4.8.13 and earlier support CUDA 10 and NVIDIA drivers 440.00 and earlier. For older hardware, you can downgrade to WSE 4.8.13 and run NVIDIA server driver 440 with CUDA 10.x. See Resolved: Wowza Streaming Engine does not support CUDA 11 (NVIDIA drivers 450.00 and later) for more information.
See the following sections for hardware and driver information specific to accelerated NVENC encoding, NVCUVID decoding, or CUDA scaling.
NVIDIA Video Codec SDK Encoder accelerated encoding
Wowza Streaming Engine leverages the NVIDIA Video Codec SDK Encoder (NVENC) API to access the high-performance hardware video encoder in NVIDIA graphics cards. NVENC-based video encoding is faster and consumes less power than legacy CUDA-based or CPU-based encoding. Not all NVIDIA cards support NVENC. For supported hardware, see the NVENC Encoding GPU support matrix on the NVIDIA website.
Notes:
- Older graphics drivers for your NVIDIA hardware may limit NVENC-based video encoding to approximately 30 simultaneous encoding sessions. Update your graphics driver to the latest version to avoid this limitation.
- You can use more than one NVIDIA graphics card for NVENC accelerated encoding by specifying which card to use in your Transcoder template with the GPU ID setting. See Template details - Encode for more information.
For instructions on how to set up NVENC accelerated encoding, see the following articles:
NVIDIA Video Codec SDK Decoder accelerated decoding
Most modern NVIDIA graphics cards have fixed-function hardware that uses the NVIDIA Video Codec SDK Decoder (also known as NVCUVID or NVDEC) for accelerated decoding. For supported hardware, see the NVDEC Decoding GPU support matrix on the NVIDIA website.
For instructions on how to set up NVCUVID/NVDEC accelerated decoding, see Template details - Decode.
NVIDIA CUDA-accelerated video scaling
Wowza Streaming Engine 4.6.0 and later supports using CUDA-based GPU resources to scale video, leveraging the NVIDIA CUDA API. This reduces the overall CPU usage of a given set of Transcoder sessions. The software is compatible with NVIDIA CUDA cards of Tesla technology or greater.
For instructions on setting up NVIDIA CUDA video scaling, see Template details - Scale.
