Staying a single of the Alliance for Open Media founding associates, Intel has carried out a ton to promote the AV1 codec and make it far more obtainable to content makers/vendors and finish-customers. Intel was the initially to supply hardware AV1 decoding with its Xe-LP GPUs in 2020. This week it unveiled model 1. of its fast open-source Scalable Video Technologies AV1 encoder and decoder for CPUs. SVT-AV1 works with all present day processors.
The AV1 open-supply video codec was designed for ultra-large-def resolutions, wide coloration gamut, and higher-dynamic array enhancements. AOMedia stated in 2018 that its AV1 was 30% a lot more economical than present codecs (implying generally on H.265/HEVC that is designed for related 4K+ information), which is a massive deal. But one difficulty with hugely-economical codecs is that they are exceptionally hungry for resources and usually need components acceleration to get the job done correctly. Meanwhile, modern CPUs have a large amount of resources and new guidance that can be utilized to decoding and encoding, which is precisely what SVT-AV1 does.
SVT-AV1 is a scalable conventional-agnostic encoder/decoder library that can choose gain of the multi-core mother nature of contemporary CPUs and AVX2 guidance. The SVT-AV1 also provides additional AVX2 optimizations to increase overall performance, impression top quality improvements, fast-decode for far more preset levels, and S-frames guidance, reports Phoronix.
Intel’s SVT-AV1 libraries are supported on modern-day x86 equipment (Intel 5th Generation Core ‘Broadwell’ and bigger) functioning Apple’s macOS, Microsoft’s Home windows, and Linux.
Intel and Netflix originally commenced the SVT-AV1 project to develop a generation-good quality AV1-encoder with functionality concentrations applicable to several apps, from premium video-on-demand to serious-time and stay encoding/transcoding. In August 2020, the SVT-AV1 encoded/decode library was adopted by AOMedia’s Software Implementation Operating Group (SIWG) to make AV1 extra common. The SVT-AV1 version 1. release marks a milestone in the advancement of the encoder/decoder libraries.
Edition 1. of SVT-AV1 encoder/decoder libraries is a milestone and excellent news for articles creators and close-users. On the other hand, for corporations like Netflix, Intel now provides Arctic Audio-M accelerators dependent on DG2 silicon that can manage eight simultaneous 4K streams and supports hardware-accelerated AV1 encoding and decoding.
The single-tile Intel Arctic Seem 1T features an Xe-HP GPU with 384 EUs and 16GB of HBM2E memory, giving a peak bandwidth of up to 716 GB/s (which most likely usually means that we are dealing with two stacks of HBM2E that use a 2048-bit interface). The accelerator is a quick solitary-slot whole-height card rated for a 150W TDP.
Intel’s Arctic Seem 2T card carries an Xe-HP GPU with two tiles, 960 EUs (480×2 to be additional precise), and 32GB of HBM2E DRAM. The accelerator utilizes a total-length, full-peak (FLFH) variety element and is rated for a 300W TDP shipped employing a single eight-pin electricity connector. (One thing to maintain in mind is that IgorsLab edited the pictures of the playing cards to protect the resource.)
Intel’s Xe-HP architecture is a considerably cry from the firm’s Xe-LP architecture we know from the Iris Xe shopper-grade GPUs. The Xe-HP card supports a lot more floating-issue formats (e.g., FP16, FP32, FP64 for common reason, bfloat16 structure for AI/ML computing), extra compute-unique guidelines, DP4A convolution instruction for deep mastering, and Intel’s XMX extensions.
The datacenter-oriented Xe-HP GPUs use all-new execution units (EUs) with a variety of IPC enhancements, characteristic HBM2E memory assistance, and are produced utilizing Intel’s effectiveness-optimized 10nm SuperFin process know-how. In short, the Xe-HP is not the Xe-LP or Xe-HPG on steroids, but one thing totally distinctive.
Intel now authorized some of its consumers to preview its Arctic Seem compute playing cards carrying one-tile and twin-tile Xe-HP implementations. Intel declared a quad-tile Xe-HP implementation final calendar year and even shown just one of this sort of accelerators in action offering more than 42 FP32 TFLOPS of effectiveness. Having said that, the business is not ready to sample it appropriate now or is only sampling it with select prospects.
Intel’s Xe-HP ideas are not absolutely obvious as the organization has under no circumstances thorough them. In the meantime, the EU count of these two playing cards is somewhat lower than anticipated (assuming that a person Xe-HP tile features 512 EUs). At existing, we don’t know how aged these playing cards are and which configurations Intel plans to ship.