• The new data center accelerator can process four times as many simultaneous 1080p video streams, totaling 32, while consuming less energy.
  • The majority of the computational capacity of the Alveo MA35D is provided by two onboard chips that AMD refers to as VPUs.

Advanced Micro Devices Inc. unveiled a new data center accelerator, the Alveo MA35D, enabling businesses such as streaming service providers to process video more efficiently.

The MA35D is the successor to a product formerly known as the Alveo U30. It can process four times as many as 1080p video streams simultaneously, totaling 32 while consuming less energy. According to internal AMD experiments, the MA35D has 66% lower “typical power consumption” than the Alveo U30.

Streaming services do not deliver video content to customers in original form but in a compressed file format. Video in compressed formats require less storage space. As a result, data may be transmitted over the network more bandwidth-efficiently, lowering expenses.

Encoding is the process of altering the file format of a video stream. AMD’s new MA35D accelerator was created exclusively for encoding. It can also decode, converting a video stream into its original file format.

A computing module that packs one or multiple chips into a chassis that is smaller than a standard server is known as an accelerator card. AMD claims that a standard one-rack-unit server can accommodate eight MA35D cards. A computer with eight MA35D accelerators can manage up to 256 video streams in parallel.

Much of the processing power of the Alveo MA35D is provided by two onboard chips that AMD refers to as VPUs. A five-nanometer process was used to manufacture these application-specific integrated circuits (ASICs). An ASIC is a custom-built chip designed to handle a particular workload.

Optimizing a processor for a particular workload increases its performance. Engineers can construct a chip with many multiplication-optimized circuits if an application performs multiplications frequently. These chips are known to be more efficient than a general-purpose processor with fewer specialized circuits.

ASICs, such as the VPUs that enable AMD’s new accelerator, are purpose-built for a specific workload. They can be more closely matched to an application’s requirements than FPGAs, another type of chip commonly used for such tasks. The AMD MA35D accelerator is entirely ASIC-based, whereas its predecessor, the Alveo U30, also contained inferior FPGA components.

Dan Gibbons, General Manager of AMD’s AECG Data Center Group said, “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. 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.”

In addition to its two ASIC-powered VPUs, the MA35D includes additional compute modules. There is also an artificial intelligence engine capable of 22 trillion operations per second. According to AMD, the engine executes AI models that automatically enhance video stream quality and compress them to conserve bandwidth.

The chip manufacturer is presently providing MA35D samples to early customers. In the third quarter, AMD company intends to commence high-volume shipments.