The Alveo-based card offloads video processing functions from the CPU to reduce latency and maximize channel density.
Use of video streaming encoder cards in the data center is on the rise, and AMD is the latest to tackle the demands of high-volume streaming.
Even before the pandemic forced everyone to work from home, videoconferencing usage was climbing. Once schools and businesses became dependent on Zoom calls, video streams started clogging data centers and network pipes across the country. Reliance on video among consumers also took off as TikTok, Twitch, and Facebook became broadcast platforms.
With users demanding broadcast-quality video – no one wants blurry, blocky, poor resolution – enterprises are left to deal with increased strain on server CPUs.
Cue video streaming cards. Intel recently came out with one called Arctic Sound, based on its Xe GPU architecture. Now AMD is releasing its own accelerator called the Alveo MA35D video processing unit (VPU).
The MA35D offloads video processing functions from the CPU to the VPU, minimizing data movement between the CPU and VPU, reducing overall latency, and maximizing channel density with up to 32x 1080p60, 8x 4Kp60, or 4x 8Kp30 streams per card
The platform delivers up to a 52% reduction in bitrate for bandwidth for AV1 transcoding savings versus a comparable software implementation. That’s because at the core of the processor are two custom AV1 encoder blocks built using a 5nm process technology node. The result is the ability to transcode up to 32 1080P/60 bitstreams at only 1W per channel.
The card supports a variety of popular video streaming codecs, including H.264/AVC, H.265/HEVC, VP9, and AV1, the last of which is becoming a popular streaming protocol for its high compression, low latency, and royalty-free usage.
Kevin Krewell, principal analyst with TIRIAS Research, said that with the explosion of video streaming for meetings and interaction, a video accelerator is needed. Video compression in and of itself is not particularly taxing to the CPU, but when you multiply it by hundreds or thousands of users, that adds up.
“These cards are designed for data centers, where you’re dealing with a massive amount of video streaming in and out. So the efficiency of a hardware encoder is far greater than any CPU or trying to do it in software,” he said.
AMD switched from an FPGA to an ASIC design after getting feedback from cloud providers doing video, he said. “As good as the FPGA was at performing encoding of video streams, an ASIC based on that same technology will always be more efficient. Plus the fact they went to a five nanometer process node made it more efficient,” he said.
The AV1 protocol is growing in popularity because it is an open standard and no royalties have to be paid to use it, making it a more appealing compression protocol than H.264, and that benefits the Alveo card, Krewell added.
The Alveo MA35D card is sampling now and will be available in production in Q3 of this year.