AMD Unveils First 5nm ASIC-based Media Accelerator Card for Advanced Interactive Media Services

The tl;dr version: the Alveo MA35D media accelerator introduces notable improvements in channel density, ultra-low-latency, and video quality. Whether it’s a “game-changer” depends on the industry’s adoption rate and the overall impact it has on interactive media services, but it certainly has the potential to make a significant difference.

The newly launched media accelerator card, AMD Alveo MA35D, is equipped with two 5nm, ASIC-based video processing units (VPUs) supporting the AV1 compression standard, specifically designed to power live interactive streaming services at scale. As live content currently dominates over 70% of the global video market, new low-latency, high-volume interactive streaming applications like watch parties, live shopping, online auctions, and social streaming are emerging.

The Alveo MA35D ASIC includes a variety of features: it has quad encoders, dual decoders that support AV1, HEVC, VP9, and H.264, dual adaptive bitrate scalers, and dual compositor engines. In addition, it comes with video QoE and look-ahead engines, a quad-core general-purpose microprocessor, LPDDR5 memory, and a dedicated AI processor. These functional blocks within the ASIC are connected through an on-chip network, while the card communicates with the host via a high-speed PCIe Gen 5 interface.

Alveo MA35D offers a higher channel density of up to 32x 1080p60 streams per card, enhanced power efficiency, and ultra-low-latency performance, addressing the escalating infrastructure costs associated with compute-intensive content delivery. The card provides up to 4x higher channel density, 4x max lower latency in 4K, and 1.8x greater compression efficiency compared to its predecessor, Alveo U30 media accelerator.

The purpose-built VPU in Alveo MA35D accelerates the entire video pipeline, minimizing data movement between the CPU and accelerator, reducing latency, and maximizing channel density. It supports mainstream H.264 and H.265 codecs and includes next-generation AV1 transcoder engines for up to a 52% reduction in bitrate compared to a software implementation.

Supporting up to 32 1080p60, 8 4Kp60, or 4 8Kp30 streams per half-height, half-length (HHHL) card, it can deliver 256 channels when used in a 1U rack server equipped with eight cards. This enables maximum stream efficiency per server, rack, or data center. Additionally, the card operates at roughly 1 watt per stream, which is a significant reduction in power consumption per channel. When compared to the previous generation Alveo U30, the Alveo MA35D boasts 4X higher channel density, 1.8X greater compression, and 4X lower latency. These enhancements in performance and density make the Alveo MA35D an attractive option for customers looking to scale their operations with optimal efficiency.

The accelerator also features an integrated AI processor and dedicated video quality engines to improve the user experience at a lower bandwidth. AI-enabled optimization techniques include region-of-interest (ROI) encoding for text and face resolution, artifact detection, and content-aware encoding for predictive insights for bitrate optimization.

The Alveo MA35D media accelerator can cost-effectively scale interactive media services by maximizing the number of channels per server while minimizing power and bandwidth-per-stream. The platform is compatible with the AMD Media Acceleration software development kit (SDK), which supports widely used FFmpeg and Gstreamer video frameworks.

AMD has started sampling Alveo MA35D media accelerators, with production shipments anticipated in Q3. An Early Access Program is available for qualified customers, providing comprehensive documentation and software tools for architectural exploration.

Video and Media Accelerators

Media accelerators, or video accelerators, are specialized hardware devices designed to offload video processing tasks from the CPU to dedicated processing units. They are particularly useful in applications where efficient video processing, encoding, decoding, and streaming are required. Their role in the supply chain of streaming video gets more and more important as the role of interactivity, and the general growth, of streaming continues to consume the internet.

One of the key components of a media accelerator is the dedicated Video Processing Unit (VPU). VPUs are specifically designed for video-related tasks, such as encoding, decoding, and transcoding. By handling these tasks more efficiently than general-purpose CPUs, VPUs can enhance video performance and reduce latency, allowing for smoother playback and improved video quality.

Media accelerators can also improve display performance by implementing advanced video codecs, such as H.264, H.265, and AV1. These codecs allow for better video compression and decompression, which leads to reduced bandwidth consumption and storage requirements. This is especially important for high-resolution content, where large amounts of data must be transmitted and displayed in real-time.

In addition to VPUs and advanced codecs, media accelerators can integrate AI processors and dedicated video quality engines. These components enable intelligent optimization techniques, such as region-of-interest (ROI) encoding, artifact detection, and content-aware encoding. By employing these techniques, media accelerators can dynamically adjust encoder settings and improve perceived visual quality while minimizing bitrate requirements. This results in a better viewing experience without increasing the amount of data transmitted.

By offloading video processing tasks from the CPU, media accelerators free up valuable processing resources for other tasks. This can lead to improved overall system performance, enabling smoother multitasking and more responsive applications, especially in cases where the CPU may be overwhelmed with simultaneous tasks.