Intel launches the AV1 video encoder and decoder for processors with higher efficiency and less power

Intel launches the AV1 video encoder and decoder for processors with higher efficiency and less power

Intel is a founding member of the Alliance for Open Media, so it’s no surprise that it has created a new and more efficient video codec to become more widely available to content creators, end users, and streaming providers. The company has released version 1.0, which offers scalable, open source video technology for encoding and decoding in CPUs. Intel’s further AV1 decoding appeared in the company’s Xe-LP GPUs as early as two years ago. Even better than Intel’s latest version is that it is fully compatible with all current processors.

Intel’s AV1 encoding and decoding capabilities will help end users and streaming data servers bring high-quality content in a lighter process.

The AV1 open source video codec is designed for high definition video resolutions, wide color gamuts and high dynamic range updates. AOMedia stated that its AV1 was 30% more productive than existing codecs (mainly suggesting H.265 / HEVC intended for comparative 4K + video content). One problem with exceptionally competent codecs is that they are incredibly resource hungry and generally require an increase in equipment speed to function accurately. However, current CPUs have a lot of next generation resources and instructions that the system can use to encode and decode, which is what SVT-AV1 does.

AMD Navi 24 GPU for Radeon RX 6500 XT is primarily designed for laptops with PCIe Gen 4 support

SVT-AV1 is a library of adaptable encoders and decoders that take advantage of the multi-center nature of today’s CPUs and AVX2 guidelines. Likewise, the SVT-AV1 adds further AVX2 enhancements for easier execution, image quality updates, quick interpretation for additional preset layers and S contours.

Intel’s SVT-AV1 libraries are supported on current x86 machines (Intel 5th generation Core “Broadwell” and later) running many operating systems, such as Apple’s macOS, Microsoft’s Windows, and the open source Linux operating system.

The delivery of SVT-AV1 1.0 variant denotes a result in the promotion of encoder / decoder libraries. Intel and Netflix initially started the SVT-AV1 venture to promote a quality AV1 encoder with relevant execution levels to different applications, from premium video on demand to constant and live encoding / transcoding. In August 2020, the SVT-AV1 coded / unveiled library was taken over by AOMedia’s Software Implementation Working Group (SIWG) to make AV1 more famous.

Variant 1.0 of the SVT-AV1 encoder and decoder libraries is an uplifting milestone and novelty for content creators and end customers. Despite this, Intel currently offers Arctic Sound-M gas pedals in light of DG2 silicon that can handle eight simultaneous 4K streams and supports equipment that accelerates AV1 encoding and routing for organizations like Netflix.

The Intel Arctic Sound 1T single-tile features an Xe-HP GPU with 384 EU and 16GB of HBM2E memory, offering excellent data throughput up to 716GB / s (which presumably implies we’re managing two heaps of HBM2E that use a 2048-cycle interface). The accelerator pedal is a short, single-opening full-level board rated for a TDP of 150W.

The AV1 specification of the open source codec is now available, offering much more efficient compression than HEVC and VP9

The Intel Arctic Sound 2T card carries an Xe-HP GPU with two panels, 960 EU (480 × 2 to be more exact) and 32GB of HBM2E DRAM. The accelerator pedal features a full-length, full-level (FLFH) structure factor and is rated for a piped 300W TDP using an eight-pin power connector. (One thing to remember is that IgorsLab altered the card images to safeguard their source.)

Intel’s Xe-HP engineering is a far cry from the organization’s Xe-LP design we know from shopper-level Iris Xe GPUs. Xe-HP board supports grave derivation point designs (e.g. FP16, FP32, FP64 for widely applicable, bfloat16 design for AI / ML processing), more explicit register indications, DP4A convolution guide for deep learning and XMX expansions from Intel.

The Xe-HP GPUs located in the data center utilize completely new execution units (EU) with several IPC enhancements, highlight HBM2E memory support, and are built using Intel’s lean 10nm SuperFin process innovation. Hence, the Xe-HP is not the Xe-LP or the Xe-HPG with steroids, but something unique.

Intel has currently allowed some customers to review its Arctic Sound figure boards that carry single-pane Xe-HP executions. Intel declared a quad-tile Xe-HP implementation last year and surprisingly showed one of these throttle pedals in real life, featuring north of 42 running FP32 TFLOPS. However, the organization is not ready to test or test it with select customers.

Intel’s Xe-HP plans are unclear as the organization never has any details. We have no idea how old these cards are and what configurations Intel plans to supply. Meanwhile, the EU count of these two cards is lower than expected (expecting a Xe-HP card to highlight 512 EU).

Leave a Comment

Your email address will not be published.