Emerging codecs including H.265/HEVC and VP9 offer higher compression rates with the same or better visual quality and encoding costs. Implementations vary – the compression rate and visual quality of a given encoder can be dramatically different. NGCodec has developed state-of-the-art encoders, using Xilinx FPGAs, providing over 30% better compression with H.265/HEVC and VP9 compared to existing H.264/AVC encoders targeting live workloads. For video streaming, better compression means reduction in bandwidth. We estimate that cost savings from using next-generation codecs range from 15% to 20% even when applied only to the most popular subset of videos or channels. In addition, better visual quality can improve user and content creator satisfaction. This paper provides a quantitative framework for estimating the economic benefit of adopting NGCodec’s H.265/HEVC and VP9 vs legacy H.264/AVC for live streaming.
NGCodec is a Silicon Valley-based startup that develops next-generation Cloud-based live video encoders. Unlike almost all competing encoders, NGCodec does not use software running on X86 CPU’s, instead, NGCodec uses Xilinx’s programmable hardware called Field Programmable Gate Arrays (FPGAs) running in public clouds including AWS, Alibaba, Huawei, Tencent, Baidu etc.
Visual Quality and User Growth are Driving Exponential Bandwidth Costs
The emergence of advanced video codecs promises to improve visual quality and reduce the bandwidth required for real-time video streams. Our customers are demonstrating strong interests in emerging video categories including live game and camera broadcasts combined with community social streams. Over the next four years, global video streaming bandwidth is expected to grow rapidly. Mobile, social streaming is expected to grow at an 18.9% CAGR, and live video and game streaming is growing at an 26.8% CAGR. Cisco predicts that overall live video bandwidth will grow 15-fold between 2016 and 2021.
As live video streams increase in popularity, their bandwidth and encoding requirements become a larger portion of service operating expenses. Successful app-based live streaming services are seeing an increase in annual bandwidth costs exceeding 75% year over year as their services ramp up and they work to improve visual quality. As TVs, PCs and mobile phones increase in resolution, users are increasingly sensitive to poor quality streams. Leading streaming platforms are increasingly offering high bitrate content, with 8-10mbps for leading providers at resolutions of 1080p60 and beyond. The combined result is fast growth in bandwidth and encoding requirements for video services, and a corresponding increase in operating costs.
The primary drivers of cost for a video encoding infrastructure are video encoding computation, IT power consumption and stream size/ bit rate. Advancements in video encoding promise to improve visual quality and reduce bandwidth requirements. Video service providers can reduce costs by reducing the size of streams and through the adoption of efficient video encoders. The performance advantages of H.265/HEVC and VP9 over the 15-year-old H.264/AVC codec can be dramatic. Below, NGCodec used its RealityCodec real-time encoder to create a 1.5 Mbps 1080p30 stream. The stream is high quality and has a surprisingly low bitrate. This is just an example of the dramatic improvement in visual quality and bandwidth sipping potential of next-generation codecs.
x264 H.264/AVC at 3 Mbps
NGCodec H.265/HEVC at 1.5Mbps
Visual quality as measure by VMAF is significantly better using the NGCodec Xilinx FPGA-based hardware encoder spanning the most common bitrates for the next-generation codecs including H.265/HEVC and VP9. This chart measures average results for four varied video clips. The gap between any two lines is the bandwidth savings between those encoding approaches.
The Combined Reach of the H.265/HEVC and VP9 Codecs - Almost 100% of Today’s Client Platforms
The VP9 codec is a royalty-free, open source codec. It is not under the control of a single vendor. VP9’s performance characteristics are similar to the H.265/HEVC standard. VP9 is the precursor to AV1 which is being standardized by the Alliance of Open Media, and already has gained a wide adoption, supported by the entire Android ecosystem and by the Chrome, Firefox, and Edge browsers. Active users on these VP9 platforms represent approximately 70% of all PC and Smartphone users. The similar, but not royalty free H.265/HEVC codec is further supported by all iOS devices for the last few years, giving the combination of H.265/HEVC and VP9 almost 100% client platform support.
Given that the next generation of codecs can reduce bandwidth and improve visual quality, we see that the costs of operating services, including bandwidth costs and even encoding costs might be reduced. These factors are important in translating the performance of the emerging codecs into cost savings calculations and are introduced below:
CLIENTS: The adoption of video clients supporting next-generation codecs. Today, H.265/HEVC and VP9 have support in almost 100% of client platforms.
CONTENT: The distribution of video popularity can impact savings. Most services report that 10% of content accounts for 60% to 90% of bandwidth. So encoding that top content tier with an efficient codec will provide the majority of the bandwidth benefit and cost savings.
ENCODER BIT RATE: The ability of an encoder to provide the same visual quality at a reduced bandwidth requirement forms the best, apples to apples comparison of encoder performance.
ENCODER PERFORMANCE PER WATT: The performance of video encoders to process video streams can vary dramatically. Hardware based encoders are generally more efficient but less flexible. Software based encoders offer more flexibility but generally consume significantly more power. Emerging FPGA and other silicon-based approaches perform encoding with specialized processors designed specifically for higher performance.
Understanding Bandwidth and Encoding Costs For Live Streaming
The formula for calculating the percentage of savings on bandwidth costs can be summarized as a product of the percentage of clients supporting the codec, the percentage of content encoded that is watched, and the relative efficiency of the codec.
Bandwidth Cost Savings = Percent of Clients Supported * Percent of Content Encoded that is Watched * Encoder Bit Rate Savings
The cost savings using a high performance H.265/HEVC and VP9 codec can be dramatic. Today, the NGCodec RealityCodec encoder using H.265/HEVC or VP9 provides 30% savings in bitrate over best-in-class H.264/AVC encoding at the same or better visual quality. It is also currently the highest performance H.265/HEVC and VP9 encoder able to encode H.265/HEVC or VP9 stream at almost exactly the same power consumption per channel as a commercial, high performance H.264/AVC CPU video encoder. The bandwidth cost savings is driven by the higher quality of H.265/HEVC and VP9 relative to H.264/AVC at the same bit rate. For many customers, the 20% most popular live streaming channels, generate 80% or more of the network traffic. This implies that a platform with 80% of the traffic using the top 20% percent of streams, can substantially reduce network traffic by encoding the top 20% of the streams using VP9 or HEVC. For instance, if VP9 with 70% platform coverage were employed to encode only the top 20% of streams for 80% of watched content, the cost reduction would be given by:
16.8% Bandwidth Cost Savings = Percent of Clients Supported [70%] * Percent of Content Encoded that is Watched [80%] * Encoder Bit Rate Savings [30%]
Estimates for bandwidth cost savings range from approximately 15% to 20% as we vary the percent of channels encoded and the platform coverage - H.265/HEVC plus VP9 cover approximately 100% of client platforms. To demonstrate the potential savings, the table below shows costs per Gigabyte ranging from 4 cents to .25 cents and calculates the cost per million concurrent streams at 2 megabits average bandwidth per year. Savings at 16.8% bandwidth reduction range from $52 Million to $3.2 Million per million streams. Regardless of bandwidth costs, by adopting H.265/HEVC and VP9 significant bandwidth cost savings can be achieved.
Let’s take a look at the cost difference between a hardware and software solution. Encoding costs with a hardware encoder are substantially lower than encoding with software. A high end Xeon E5 CPU with a commercial software-based encoder can process approximately 8 1080p60 H.264 channels at approximately 30% higher bitrate achieving “medium” to “low” preset quality. This same server would not be able to encode a single 1080p60 1080p60 stream using HEVC or VP9 low bit rate encoding.
As such, it is impossible to encode real-time HD streams with the Visual Quality achieved by the hardware-based encoder employing H.265/HEVC or VP9 codecs. The NGCodec Xilinx FPGA-based server solution can process 8 channels of 1080p60 H.265/HEVC or VP9 channels at a 30% lower bitrate achieving much better subjective visual quality - beyond H.264/AVC “slow” preset quality. Taking the improved visual quality as a large benefit, the encoding cost of an H.265/HEVC or VP9 solution is much lower than the cost of encoding to H.264/AVC/VP9/HEVC with software encoders.
Conclusion: Save Money While Improving Visual Quality
Creating the next successful video streaming application or operating an established live streaming service is becoming increasingly, and potentially exponentially more expensive. Live video service providers can save millions of dollars in bandwidth cost, thanks to a 30% reduction in bandwidth using hardware video encoding. Employing H.265/HEVC and VP9 for real time streams provides the best possible visual quality and similar encoding costs. Adopting these next generation codecs can improve user experiences while reducing costs, paving the way for future innovation in live streaming video. NGCodec is committed to further bit rate reductions and has a roadmap of VQ improvements which will be achieved through software updates, in addition next year NGCodec will deliver a live AV1 encoder which will allow the investment made in Xilinx FPGA infrastructure to be reused with even further compression gains.