MulticoreWare achieves 18% performance boost on the x265 video codec with Intel® AVX-512 Instructions for high quality encoding of 4K HDR content

  • On April 9, 2018

Saratoga, CA / April 9, 2018 – The benefits of deploying Intel® Xeon® Scalable processors for encoding OTT and broadcast content to HEVC just became more compelling with the combined efforts of MulticoreWare Inc. and Intel Corporation to accelerate the popular open-source HEVC encoder, x265, with Intel® Advanced Vector Extensions 512 (Intel® AVX-512). With a careful selection of kernels and algorithms, using Intel AVX-512 instructions delivered an additional speed-up of up to 18% for high-quality offline encoding of 4K HDR content on Intel Xeon Scalable processors, on top of a 1.56X speed-up from improvements to the architecture over previous-generation Intel® Xeon® processors.

The x265 video codec is the most popular open-source HEVC encoder, deployed in production workflows by several leading OTT and broadcast vendors. To address the complexity of HEVC encoding, MulticoreWare has leveraged its expertise in microarchitecture-aware optimizations to implement a combination of multi-threading, and hand-tuned assembly kernels in accelerating x265. This has resulted in an encoder that is near-identical in efficiency to the reference HEVC encoder (HM) but at significantly improved encoding times that are ideal for offline encoding. For live encoding, MulticoreWare’s UHDkit product employs multiple instantiations of x265 with tight integration to encode 4K HDR content at 60fps and optimize ABR work flows.

The Intel Xeon Scalable platform presents a revolutionary improvement in processor technology for the video encoding use-case. In particular, Intel AVX-512 offers a significant step forward in vector computing, enabling twice the throughput when compared to its predecessor. MulticoreWare, with support from Intel, has leveraged the support of the new vector instructions to accelerate the existing kernels of x265. This acceleration required a carefully selection of the kernels and algorithms for acceleration with Intel AVX-512 to avoid the pitfalls of using vector instructions with increased widths. The result is a speed-up of up to 18% for encoding 4K HDR content, enabling improved throughput for companies using x265 and UHDkit for encoding.

Our continued enhancement of x265 with complex implementations such as the Intel AVX-512 optimization would not have been possible without support from Intel. Access to technical information and support from Intel are paramount for this successful implementation in an open source project like x265,” said AGK Karunakaran CEO of MulticoreWare.

MulticoreWare’s implementation of Intel AVX-512 is one of the first applications of this technology to HEVC encoders, and it validates the value of this important technology in Intel Xeon Scalable processors. We’re excited to be working with MulticoreWare on getting this important development into the open source community, and we’ll be working with them to help the rest of the industry benefit and learn how to apply these optimizations to the entire HEVC encoding ecosystem,” said Lynn Comp, Vice President of the Data Center Group, and General Manager of the Visual Cloud Division at Intel Corporation.

A demonstration of these improvements will be available at MulticoreWare’s booth (SU-14708), and Intel’s Booth (SU-3110), in the South Upper Hall of NAB 2018 in Las Vegas.

About MulticoreWare

MulticoreWare is a software company with products in Broadcast, OTT, Autonomous Vehicles and Video analytics such as x265, UHDkit for Video encoding, UHDcode and VP9 for video decoding, LipSync and TextSync for Video Quality analysis. MulticoreWare has a worldwide team of computer scientists and experts in high performance and heterogeneous computing to accelerate video, machine learning, neural networks, computer vision, compiler technologies, and more. Visit https://www.multicorewareinc.com for more information.

Intel and Xeon are registered trademarks of Intel Corporation in the United States and other countries.

 

0 Comments