Advanced Micro Devices Inc. today debuted a new data center accelerator, the Alveo MA35D, that will help companies such as streaming providers process video more efficiently.

The MA35D is the successor to an earlier product called the Alveo U30. It can process four times as many 1080p video streams at once, for a total of 32, while using less power.

Streaming providers don’t distribute video content to users in its original form, but rather package it into a compressed file format. Video stored in compressed formats takes up less storage space. As a result, it can be sent over the network in a more bandwidth-efficient manner, which reduces costs. 

The task of changing a video stream’s file format is known as encoding. AMD’s new MA35D accelerator is specifically designed to perform encoding. It can also carry out decoding, which is the task of translating a video stream back into its original file format.

An accelerator card is a computing module that packages one or more chips into a chassis smaller than a standard server. In the case of the MA35D, AMD says eight of the cards can be incorporated into a standard one-rack-unit server. A machine with eight MA35D accelerators can manage up to 256 parallel video streams.  

Most of the Alveo MA35D’s processing power is provided by two onboard chips AMD refers to as VPUs. They’re application-specific integrated circuits, or ASICs, made using a five-nanometer process. An ASIC is a custom chip designed to run one specific type of workload.

Optimizing a processor for a specific workload makes it faster. If an application frequently performs multiplications, engineers can build a chip with a large number of multiplication-optimized circuits. Such a chip would perform better than a more general-purpose processor that doesn’t have as many specialized circuits.

ASICs like the VPUs that power AMD’s new accelerator are built from the ground to run one type of workload. They can be aligned with an application’s requirements more closely than FPGAs, another type of chip commonly used for such tasks. AMD’s MA35D accelerator features a fully ASIC-basic design, whereas the previous-generation Alveo U30 it replaces also included slower FPGA elements.

“We worked closely with our customers and partners to understand not just their technical requirements, but their infrastructure challenges in deploying high-volume, interactive streaming services profitably,” said Dan Gibbons, general manager of AMD’s AECG Data Center Group. “We developed the Alveo MA35D with an ASIC architecture tailored to meet the bespoke needs of these providers to reduce both capital and operating expenses.”

Besides its two ASIC-powered VPUs, the MA35D also features other compute modules. Among them is an artificial intelligence engine that can perform up to 22 trillion operations per second. According to AMD, the engine runs AI models that automatically increase the quality of video streams and compress them to save bandwidth.

The chipmaker is currently sampling the MA35D to early customers. AMD company plans to begin high-volume shipments in the third quarter.

Image: AMD