ASICs for Video Encoding: The Surest Way to Optimize Efficiency and Cost

by Anita Flejter at NETINT Technologies Inc.

Video content is everywhere, and it’s growing, driven by user-generated content platforms like YouTube, Instagram Reels, and TikTok. Optimizing video encoding is crucial for efficient storage, bandwidth utilization, and cost-effectiveness. 

This article, based on Dylan Patel’s lecture during our recent symposium, explores the need for Application-Specific Integrated Circuits (ASICs) in video encoding. It provides insights and best practices for businesses and individuals in video content creation and distribution.

Understanding the Video Encoding Landscape

Video encoding is a complex process that involves converting raw video footage into different formats and resolutions suitable for various devices and internet connections. Using traditional CPUs for video encoding at the massive scale required by today’s user-generated content platforms is economically unfeasible due to the sheer number of servers needed to handle the workload. 

If a platform like Google were to encode 500 hours of video per minute into the H.264 format, it would require a staggering 2.2 million servers to accomplish the task. This estimation assumes that each server is equipped with two CPUs and is operating at 100% utilization, which is rarely achievable in practice due to the spiky nature of video traffic and uploads.

The situation becomes even more challenging when considering more advanced video codecs like VP9 or HEVC, which offer higher compression efficiency but require more computational power. Encoding the same 500 hours of video per minute into VP9 or HEVC would need an astonishing 10 million servers. Wait till you see the estimate of servers needed for AV1, you better be sitting down.

This number is impractical and impossible to achieve. To meet such a demand, more servers would have to be manufactured worldwide. But the astronomical costs of powering, cooling, and operating such a vast infrastructure do not pencil out.

As video content continues to grow and evolve, with higher resolutions and more efficient codecs like AV1 gaining prominence, the computational requirements for video encoding will only increase. As suggested by Dylan Patel, a platform the size of YouTube, encoding videos in AV1 could require over 100 million servers, which is entirely unfeasible from a logistical and financial standpoint.

Given these limitations, it becomes evident that relying solely on traditional CPUs for video encoding at scale is not a viable option for businesses dealing with massive amounts of user-generated video content. The costs associated with the necessary infrastructure would be prohibitively high, making it challenging for platforms to remain profitable while delivering high-quality video experiences to their users. This is where specialized video encoding ASICs come into play, offering a more efficient and cost-effective solution to handle the growing demands of video encoding in the modern digital landscape.

• * The Challenge: Imagine having to encode 500 hours of video every minute, as Google did in 2019. Now, consider the vast array of formats and resolutions required, such as 4K, 1440p, 1080p, 720p, 480p, 360p, and 240p, to cater to different devices and internet speeds. Additionally, you need to support multiple video codecs like VP9, H.264, HEVC, and AV1, each with its strengths and weaknesses.
• * The Solution: Enter ASICs—specialized chips designed to accelerate specific tasks, in this case, video encoding. By offloading the encoding process to ASICs, companies can achieve significant performance improvements and cost savings compared to traditional CPU or GPU-based solutions.

The Power of ASICs in Video Encoding

1. 1. Performance Gains

– Efficient Encoding: ASICs are designed to excel at specific tasks like video encoding, resulting in faster processing times and higher throughput than general-purpose processors.

– Parallel Processing: ASICs can leverage parallel processing capabilities, allowing multiple videos to be encoded simultaneously, further improving overall efficiency.

1. 2. Cost Savings

– Reduced Infrastructure: Without ASICs, companies like Google would require millions of servers to handle video encoding workloads, making it financially unfeasible.

– Example: Google estimated that encoding videos into H.264 format for all resolutions would require 2.2 million servers, each with two CPUs. For the more efficient VP9 codec, the number skyrockets to 10 million servers – an impossibility.

1. 3. Enabling New Capabilities

– Higher Resolutions: ASICs can support the latest and most efficient video codecs like AV1, enabling higher resolutions and better compression while minimizing bandwidth and storage costs.

– Real-Time Encoding: ASICs can facilitate real-time encoding, enabling live streaming and low-latency video delivery, which is crucial for applications like cloud gaming and live events.

Overcoming the Challenges of Adopting ASICs

While the benefits of ASICs are compelling, their adoption can be challenging. Developing a custom video encoding ASIC is a highly expensive endeavor, with costs often running into hundreds of millions of dollars. 

The financial investment is primarily due to the complex nature of ASIC design and development, which requires significant time, resources, and specialized expertise. The process involves designing and optimizing the chip architecture, verifying the design through extensive simulations, and fabricating the chip using advanced semiconductor manufacturing processes.

Given the high upfront costs associated with custom ASIC development, it is only a viable option for companies that have massive video workloads and the potential for savings in the billions of dollars. These companies, typically large tech giants like Google, Meta (Facebook), and Chinese hyperscalers, generate and process an enormous amount of video content daily. For them, the potential cost savings from using a custom video encoding ASIC can easily justify the initial investment in chip development.

For companies with smaller video workloads or those whose potential savings are a few hundred million dollars, there may be better choices than developing a custom ASIC. In such cases, the cost of producing the ASIC could outweigh the potential savings, making it a less attractive option. Moreover, there is always the risk that the ASIC development process may encounter issues or even fail, leaving the company with a significant financial loss and needing to fall back on less efficient CPU-based encoding solutions.

Furthermore, even for companies with the necessary scale and potential savings, the decision to develop a custom video encoding ASIC should be taken seriously. It requires a thorough cost-benefit analysis, taking into account factors such as the projected growth of video content, the evolution of video codecs, and the company’s long-term strategic goals. 

While custom video encoding ASICs offer significant performance and cost benefits, their development is costly and complex. It is a viable option only for companies with massive video workloads and the potential for billion-dollar savings, as the long-term benefits can justify the high upfront costs and risks associated with ASIC development. For most other companies, partnering with third-party ASIC providers or leveraging alternative encoding solutions may be a more practical and cost-effective approach.

Embracing the Future of Video Encoding

As user-generated video content proliferates, optimizing the video encoding process becomes paramount for businesses and platforms. ASICs offer a compelling solution, delivering unparalleled performance, cost savings, and enabling new capabilities.

Western hyperscalers like Amazon, Google, Microsoft, Meta (Facebook), and Cloudflare have been slower to adopt the VP9 codec due to licensing restrictions. VP9, an open-source codec developed by Google, offers better compression than H.264 but has faced adoption challenges due to patent licensing issues. As a result, Western hyperscalers have relied more on the older H.264 codec, which has had a longer lifespan in these regions.

On the other hand, Chinese hyperscalers like ByteDance (TikTok), Tencent (cloud gaming), Baidu, and Alibaba have been quicker to adopt video encoding ASICs compared to their Western counterparts primarily due to differences in video codec adoption. There is more inclination towards adopting the H.265 (HEVC) codec, which offers better compression efficiency than the widely used H.264 (AVC) codec. The improved compression of H.265 allows for reduced bandwidth and storage requirements, making it more cost-effective for delivering high-quality video content.

The emergence of the AV1 codec, an open-source and royalty-free alternative to HEVC and VP9, is now driving the adoption of video encoding ASICs across the board. AV1 offers even better compression efficiency than HEVC and VP9, making it an attractive option for reducing bandwidth and storage costs while maintaining high video quality. 

As AV1 gains traction, hyperscalers recognize the need for dedicated video encoding ASICs to handle this codec’s increased computational requirements.

To successfully integrate ASICs into your video encoding workflow, follow these steps:

1. 1. Assess your video workloads and potential savings to determine the viability of an ASIC solution. If you are encoding using software on CPUs, the savings on the low end can be 80%.
2. 2. Conduct thorough research and partner with experienced ASIC design firms to mitigate development risks. Remember, if your savings from adopting ASICs amount to less than $200 million over 3-5 years, you should not consider building a chip and will want to buy a commercial VPU (video processing unit) solution that ASICs power.
3. 3. Embrace open standards and modular architectures to future-proof your investment. NETINT widely deploys standards for network architecture and video streaming efficiency so that almost any workflow can be upgraded.
4. 4. Continuously evaluate emerging video codecs and resolutions to stay ahead of the curve. Don’t assume that the codecs you deploy today will always meet your needs. AV1 is coming fast but is not economically viable in many applications. With ASIC-powered VPUs, AV1 is viable.

By leveraging the power of ASICs, companies can optimize their video encoding processes, reduce costs, and unlock new opportunities in the fast-moving video streaming ecosystem. 

Want it for yourself? Let's chat about ASICs and VPUs during NAB. Pick your time here: https://netint.biz/NAB

RELATED ARTICLES