NETINT Codensity AV1 Transcoder Technical Capabilities

This article provides a technical overview of NETINT Codensity hardware transcoders optimized for the AV1 video codec. It details the underlying ASIC architecture, throughput capacity, latency benchmarks, power efficiency metrics, and software integration options that enable real-time, high-density AV1 video infrastructure for enterprise streaming platforms and data centers.

ASIC-Driven AV1 Architecture

NETINT's video processing architecture relies on dedicated Application-Specific Integrated Circuits (ASICs) rather than general-purpose compute units like CPUs or GPUs. In handling AV1, which is computationally expensive due to advanced partitioning, directional intra-prediction, and loop restoration filters, the hardware offloads the entire encoding and decoding pipeline. The architecture provides deterministic hardware execution, ensuring consistent encode times without the compute jitter typical of software-based encoding.

Resolution and Throughput Capabilities

NETINT video processing units (VPUs) deliver high-throughput, multi-stream transcoding performance per module:

• Resolution Support: Real-time processing for standard resolutions up to 4K (3840x2160) and 8K (7680x4320).
• Frame Rates: Native support for 60 frames per second (fps) at 4K, with scaling capabilities to support high-frame-rate inputs up to 120 fps depending on the resolution.
• Stream Density: A single module can process multiple concurrent 1080p60 or 720p streams simultaneously. When deployed in dense configurations within a 1U or 2U server, systems can process hundreds of real-time AV1 broadcast channels concurrently.
• Cross-Codec Transcoding: Dedicated hardware pipelines support direct transcode workflows between legacy formats (H.264/AVC, H.265/HEVC) and AV1 without transferring raw video frames back to host system memory.

Latency and Compression Efficiency

The transcoders feature an ultra-low latency operating mode designed for interactive workflows like cloud gaming, video conferencing, and live event broadcasting. The hardware pipeline achieves sub-frame encoding latency, bypassing the traditional frame-buffering bottlenecks found in multi-pass software encoders.

In terms of compression efficiency, the silicon implements full AV1 Main Profile features, matching or exceeding libaom and SVT-AV1 quality benchmarks at significantly higher speeds. This yields a 30% to 50% bitrate reduction compared to standard H.264 at equivalent visual quality metrics (VMAF, PSNR).

Color Profiles and Video Standards

• Bit Depth: Native processing of 8-bit and 10-bit video profiles.
• Chroma Subsampling: YUV 4:2:0 format handling.
• High Dynamic Range (HDR): End-to-end hardware pass-through and processing for HDR10, HDR10+, and Hybrid Log-Gamma (HLG), including dynamic metadata preservation.

Form Factors and Power Efficiency

NETINT transcoders are deployed using standard enterprise storage and expansion form factors:

• Interfaces: Available in U.2 (2.5-inch NVMe) and PCIe Add-in Card (AIC) form factors. The U.2 form factor leverages standard hot-swappable NVMe drive bays, allowing high-density scaling without proprietary chassis modifications.
• Power Consumption: The ASIC architecture consumes a fraction of the power required by CPU/GPU clusters. Individual U.2 modules typically operate within an 8W to 20W power envelope, drastically lowering total cost of ownership (TCO) and cooling requirements in data center environments.

Software Stack and Integration

The hardware interfaces with standard open-source multimedia frameworks to minimize integration overhead:

• Host Interface: Connects over standard PCIe interfaces using NVMe-like command sets optimized for video transport.
• Framework Integration: Comprehensive integration with FFmpeg and GStreamer via dedicated hardware wrappers and plugins.
• APIs: Native C/C++ software development kits (SDKs) are provided for direct hardware control, custom rate control tuning, and low-level pipeline orchestration across Linux environments.