[ad_1]
With the {industry}’s first five-nanometre Software Particular Built-in Circuit (ASIC) -based media accelerator, Superior Micro Units has set the course for a brand new course in a content-hungry world of interactive media companies. Girish Malipeddi, director of product administration and advertising at AMD, spoke to Yashasvini Razdan from Electronics For You, explaining the necessity for environment friendly media accelerators in a world dominated by GPUs and CPUs and what it means for India.
Q. What are the important thing options and advantages of the Alveo MA35D media accelerator for video transcoding workloads, and the way does it examine to different {hardware} options AMD provides?
A. The Alveo MA35D, our industry-first 5-nanometer ASIC-based media accelerator, is designed to deal with interactive streaming at scale, making it preferrred for high-volume video transcoding workloads. This purpose-built chip incorporates all the required performance for environment friendly cloud-based video streaming.
As for its capabilities, the Alveo MA35D totally helps broadly used encoding and transcoding frameworks like FFmpeg and GStreamer, together with standard libraries resembling LibAV. This compatibility permits software program engineers emigrate their purposes to our platform utilizing these APIs simply. Moreover, we provide a lower-level C/C++ API for builders preferring to create their very own software program stack or framework, making certain portability throughout completely different {hardware} platforms and catering to varied wants.
Evaluating our choices at AMD, whereas AMD EPYC CPUs present wonderful video high quality at low bitrates for purposes with low stream counts and acceptable latency, GPUs are extra related for graphics-intensive purposes or multi-application eventualities. Nonetheless, for pure video transcoding workloads requiring excessive density, low latency, low price, and low energy consumption, the ASIC-based Alveo MA35D media accelerator is the proper match. It excels in eventualities like single-stream to tens of millions of customers, software-based stay sports activities streaming, interactive media, cloud gaming, and stay e-commerce.
Q. Why did you incorporate ASIC structure as an alternative of the same old FPGA structure to construct this chip?
A. After efficiently launching our first-generation FPGA-based Alveo U30, we gained useful insights and expertise within the media accelerator household. Whereas the Alveo U30 provided flexibility with its programmable logic, it turned evident that larger scalability, decrease energy consumption, and value discount had been important for the next-generation product.
To attain better scale, density, and energy effectivity, we made the strategic resolution to construct a customized ASIC for this software(MA35D). This alternative aligns with the precept noticed in conventional AI workloads, the place GPUs outperform software program options. The ASIC design incorporates a number of hardened IP blocks, together with the ABR scaler and compositor, enabling seamless scaling of video streaming. Whereas this enterprise represents our first foray into ASICs, it’s pushed by the realisation that as video workloads develop, customised options turn into essential for optimum efficiency.
Q. Does this imply that the ASIC options are higher than FPGA options?
A. The prevalence of ASICs or programmable logic relies on the particular purposes and workload necessities. Customized ASIC options usually provide superior value factors, density, and scalability when the workload is well-defined and understood. They turn into significantly advantageous when coping with larger volumes and stuck performance.
However, with their programmable logic, adaptive computing units like FPGAs present flexibility for purposes the place completely different functionalities could also be required on the endpoints. FPGA and adaptive System-on-Chips (SoC) provide a extra versatile and adaptable structure, making them appropriate for numerous use instances.
The selection between ASICs and adaptive computing relies on the workload data degree and the applying’s particular wants.
Q. How does your media accelerator deal with video decoding/encoding? What codecs and codecs does it assist?
A. The Alveo MA35D, an Software Particular Built-in Circuit (ASIC), incorporates two ASIC-based purpose-built video processing models on a single card, revolutionising stay interactive streaming companies at scale. Every chip options 4 encoders that assist 4K 60, or decrease decision streams. The highest two encoders are multi-format, offering assist for HEVC, H.264, and AV1, whereas the underside encoder is devoted solely to AV1.
Remarkably, with an influence consumption of beneath 35 watts, the Alveo MA35D can encode at extremely quick speeds, as quick as eight milliseconds for 4K content material. Compared, typical software program options or CPUs require roughly 25 watts per stream. This effectivity interprets to a low price per stream, considerably lowered energy consumption, and improved video high quality at decrease bitrates. Service suppliers can leverage these benefits to optimise their video transcoding infrastructure and ship high-quality, low-latency streaming companies extra effectively.
Furthermore, the media accelerator’s distinctive density additionally reduces the fee, energy, and area necessities for video transcoding in knowledge facilities. By utilising this {hardware} resolution, cloud suppliers can improve their video transcoding capabilities, supporting as much as 32 AV1 and 16 H.265 (HEVC) or H.264 (AVC) 1080p60 streams per card, all inside a compact, low-power type issue.
Along with encoding capability, the chip additionally contains decoding functionality to deal with streams from numerous endpoints, essential for purposes like watch events, the place a number of streams should be decoded, mixed right into a composite picture, and shared with contributors.
Q. Why are you giving extra choice and prominence to the AV1 encoding format in your chip?
A. The choice for AV1 encoding on our chip relies on the popularity of its growing adoption and demand sooner or later. Whereas AV1 adoption remains to be in its early phases, we anticipate important traction over the subsequent few years. Main {industry} gamers, together with main sensible TV producers like Samsung and LG, have already included AV1 decoding assist, and newer Android telephones and laptops are following swimsuit.
The rising recognition of AV1 is attributed to its advantages, resembling a considerable 50% discount in bitrate for streaming, leading to price financial savings for service suppliers. As decoding functionality turns into extra prevalent on endpoint units, leveraging AV1’s effectivity and cost-effectiveness is a logical alternative for the {industry}.
Q. Might you clarify how the AV1 customary differs from the HEVC and H.264?
A. From an financial standpoint, AV1 provides a major benefit as it’s an open customary with out royalties or licensing charges. Which means anybody can use AV1 with out incurring extra prices. However, requirements like H.264 and H.265 contain substantial royalties and patent challenges, requiring corporations to pay charges for implementing these codecs.
AV1 has gained robust assist from main gamers resembling Google, recognising the rising want for environment friendly streaming algorithms because of the exponential enhance in interactive content material and media companies. AV1 excels at decreasing bitrate whereas sustaining video high quality, making it a superior possibility in comparison with legacy H.264, H.265, and different codecs. Every codec has its strengths, however AV1’s open and royalty-free nature makes it an interesting alternative, and lots of North American cloud service suppliers are already evaluating it for adoption.
Q. What’s the distinction between an open customary and open supply?
A. The idea of open supply is completely different and primarily applies to the software program facet of the world. Open supply refers to creating software program the place anybody can contribute to its growth and make it accessible to all customers. It fosters a big neighborhood of builders who can collaborate and contribute to the software program’s enchancment and innovation. In open supply, no restrictions or royalties are related to utilizing or contributing to the software program.
This idea barely differs from the dialogue on AV1, which focuses on the codec being an open customary with out royalties. The open-source mannequin allows the creation and sharing of software program that may be freely accessed and enhanced by a various neighborhood of builders.
Q. Will we see the adoption of open requirements enhance, and can that impression the pricing of the merchandise sooner or later?
A. We consider that adopting open requirements, resembling AV1, that are royalty-free, will incentivise many service suppliers to embrace these requirements. Service suppliers can scale back bills and leverage open requirements’ advantages by eliminating recurring royalty prices.
Relating to the impression on product pricing, the semiconductor {industry} has a pure pattern of reducing prices over time. This discount is pushed by elements resembling growing density, developments in semiconductor course of applied sciences, and the flexibility to pack extra performance right into a given chip. Whereas the royalty side is a small part of the fee discount, the semiconductor {industry} has persistently witnessed important price reductions over a long time as extra performance is added to the options.
Q. How is a normal processing unit (GPU) for video transcoding completely different from a media accelerator chip?
A. There are certainly different options accessible available in the market for video transcoding, resembling GPUs and CPUs. GPUs are generally used for video transcoding and provide scalability because the density will increase. Nonetheless, GPUs are primarily designed for rendering and gaming, with solely a small portion of the chip devoted to video transcoding. This may result in disadvantages resembling bigger chip dimension, larger price, and elevated energy consumption in comparison with devoted video transcoding options like our ASIC.
In distinction, our resolution offers sooner scalability and decrease energy consumption, requiring roughly 35 watts on the server degree. GPUs usually eat round 75 to 100 watts and require two slots of PCI, whereas our resolution suits right into a single slot. It’s price noting that there are additionally rivals on this area, together with these providing FPGA-based options.
Q. Can a media accelerator chip be used as an alternative of a CPU?
A. Sure, a media accelerator chip can certainly be used as an alternative of a CPU for video processing. CPUs have a general-purpose design and are essential for server performance, together with video processing. They’re appropriate for software-based options attributable to their programmability and are nonetheless a viable alternative for coping with a single stream or a reasonable variety of streams reaching tens of millions of customers.
Nonetheless, because the stream density will increase, different options turn into extra environment friendly. CPUs are usually power-hungry, costly, and occupy important bodily area. In distinction, devoted video processing options like our ASIC can present a extra environment friendly method. By including a number of MA35D playing cards to a server, the capability to deal with streams will increase exponentially, enabling a whole lot of streams to be processed on a single server. This provides important benefits when it comes to scalability and cost-effectiveness.
Q. How do you keep low bitrates on the chip?
A. The chip incorporates {hardware} blocks, together with adaptive bitrate scalers, to assist completely different resolutions for every stream. The decision is adjusted primarily based on the consumer’s accessible bandwidth, optimising the video high quality.
A compositing engine allows mixing a number of streams, and an AI processor is utilised to reinforce pixel high quality utilizing AI strategies. On-chip High quality of Expertise (QoE) engines measure goal metrics like PSNR and SSIM, offering suggestions to the end-user software for real-time video high quality adjustment or bitrate optimisation.
Moreover, a lookahead engine is employed, permitting improved bitrate compression by analysing as much as 40 frames prematurely. These functionalities are linked to the host processor by way of PCIe Gen 5 or Gen 4. Our hardware-based method ensures environment friendly and high-quality video processing, attaining important bitrate financial savings in comparison with legacy requirements like H.264. The MA35D achieves a 24% bitrate discount for HEVC, a 47% discount for H.265, and a powerful 52% discount for AV1, all whereas sustaining or enhancing visible high quality. AI strategies play an important position in enhancing pixel high quality and enhancing the visible expertise whereas sustaining low bitrates.
Q. Might you elaborate on the rationale why you’ve included AI?
A. The incorporation of AI in our product serves to reinforce video pixel high quality. Whereas many discussions revolve round application-level AI, resembling face recognition or cloud computing, AI has different useful purposes, together with video pixel high quality enhancement.
Conventional codec instruments might be extremely compute-intensive, making their implementation in software program difficult. As codecs like H.265 and AV1 demand extra computational sources, the applying of AI strategies turns into important for environment friendly processing.
AI strategies provide extra alternatives to enhance pixel high quality with out considerably growing die dimension or introducing latency. Strategies like areas of curiosity might be utilized to reinforce particular areas of curiosity within the video with out affecting the general bitrate.
Moreover, AI can be utilized for super-resolution scaling, leveraging AI to scale smaller-resolution photos to larger resolutions with improved pixel high quality and lowered computational necessities.
These AI strategies in video codec enchancment are analogous to machine studying, educating the codec to realize higher encoding outcomes over time.
Q. Please share any use instances the place AI has improved the video expertise.
A. AI has been instrumental in enhancing the video expertise in numerous purposes. As an illustration, in video conferencing or Zoom calls, our AR processor identifies areas of curiosity, resembling faces, and allocates extra bits to these areas whereas disregarding the background. This method considerably improves the picture high quality of faces with out affecting the general bitrate, leading to a greater visible expertise.
Equally, AI strategies are utilized in gaming to cut back textual content artifacts and improve particular areas of curiosity, enhancing the general visible expertise for avid gamers.
In content-aware encoding throughout stay streams, resembling information channels with minimal motion, AI-based frame-to-frame comparability can determine areas with little change. This permits for important bitrate discount by dynamically adjusting the bitrate, generally known as variable bitrate. AI strategies play a essential position in detecting these low-action frames and optimising the bitrate accordingly.
These are some widespread use instances the place we leverage AI strategies to enhance the video expertise by enhancing pixel high quality and optimising bitrates.
Q. How is that this product going to learn the Indian market? What’s the response you’ve obtained so removed from India?
A. Whereas the Indian market remains to be within the early phases for our present product, we now have gained some expertise with our older technology product. India has a major live-stream presence, significantly within the OTT platform area. Though our present product could goal barely completely different markets, we now have noticed adoption by native gamers in India for our earlier product, resembling IPTV suppliers and media corporations like Tata Sky and Dish TV. These corporations have recognised the financial advantages of our options in comparison with present choices.
As India generates a considerable amount of media content material, we anticipate elevated demand and a concentrate on addressing native wants and consumer preferences sooner or later. Presently, the Indian market depends closely on numerous cloud infrastructures, resembling AWS, because of the absence of its knowledge facilities. Nonetheless, because the market develops, native knowledge middle suppliers could play a bigger position and undertake {hardware} options like ours to cater to native wants. We’re already engaged in talks with many of those OTT platforms.
Q. How are OTT platforms using your product, as there aren’t many knowledge facilities in India?
A. Whereas knowledge facilities might not be native to India, many OTT platforms in India depend on platforms resembling AWS and Google Cloud, which offer knowledge middle companies in numerous areas, together with Asia. These platforms provide media libraries and infrastructure that OTT can leverage to host and ship content material. Whereas OTT platforms don’t essentially want native knowledge facilities, controlling their infrastructure permits for optimum implementation and customisation.
Nonetheless, the supply of cloud cases from suppliers like AWS and Azure provides flexibility and comfort for OTT platforms to utilise present infrastructure. Subsequently, whereas native knowledge facilities might be helpful, OTT platforms can nonetheless use cloud platforms to fulfill their necessities successfully.
Q. What are the particular wants and challenges of the Indian market relating to media acceleration expertise?
A. Whereas the structure for streaming is usually comparable throughout completely different areas, together with India, there could also be variations within the kind and high quality of content material being streamed. In India, a major quantity of standard-definition content material remains to be being consumed. Whereas our structure helps standard-definition content material, it’s anticipated that as extra folks undertake sensible TVs and the demand for high-definition content material will increase, there may also be a shift in the direction of larger high quality content material in India.
You will need to word that the structure we mentioned earlier can accommodate numerous purposes. Whereas particular purposes or companies could also be distinctive to the Indian market, the general structure stays relevant and adaptable to completely different areas and streaming wants.
Q. Are there any pricing or licensing issues with respect to the Indian market?
A. The pricing for the {hardware} stays the identical whatever the area. There aren’t any extra licensing charges or royalties related to the product. The pricing is constant globally. Nonetheless, when customers select to deploy the {hardware} of their knowledge facilities or go for cloud cases, there could also be extra pricing fashions particular to these cases. Nonetheless, the producer’s prompt retail value (MSRP) for the {hardware} stays constant worldwide.
Q. How do you guarantee accessibility?
A. AMD is a big firm with an in depth community of distribution channels worldwide. We promote our merchandise, together with media accelerator chips, by way of these distribution channels. Moreover, we now have direct channels for particular buyer wants, making certain a variety of choices for customers to buy our merchandise, together with India.
Q. Are there any enlargement plans regarding enterprise in India?
A. AMD actively promotes its product household to prospects worldwide, together with India. Recognising the potential for important development within the Indian market in the long term, we’re proactively reaching out and visiting India commonly. We additionally goal to extend our participation in commerce reveals and occasions in India over time. Being half of a bigger firm like AMD permits us to leverage its present presence and participation in commerce reveals in India, and we are going to proceed to develop our involvement sooner or later.
Q. How do you intend to develop in India with this product, in case your market could be very small?
A. Whereas the market in India could also be comparatively small at the moment, there are development alternatives on the horizon. As extra native content material is created and platforms like Fb Dwell, YouTube, or telecom suppliers resembling Jio Streaming acquire traction, the demand for environment friendly media acceleration expertise will doubtless enhance.
By catering to native content material creators and offering options that may accommodate their rising wants, we goal to foster adoption and develop our presence within the Indian market. Moreover, because the Indian market evolves, we’re ready to assist each native knowledge middle approaches and the continued utilisation of cloud platforms like AWS and Google Cloud, making certain accessibility for end-users whatever the enlargement path taken.
[ad_2]