As the codec share of more complex algorithms such as HEVC and AV1 gradually increases, the realization of high-quality video encoding and decoding requires hardware support, and the software method is not the best solution on both the server and mobile terminals. FPGA-specific chips can reduce latency and cost. At the NAB 2018 conference, senior multimedia technology consultant Jan Ozer interviewed Ngcodec CEO Oliver Gunasekara about the current status and future of hardware coding on HEVC, VP9 and AV1. LiveVideoStack made an excerpt of the article.
Jan Ozer: Can you tell me something about what you showed on NAB?
Oliver Gunasekara: Our company is doing video encoding technology, as well as the next generation video compression technology. Shown here is the world's most powerful HEVC and VP9 live video encoder.
Jan Ozer: I have heard the word "the strongest performance" 12 times today. How do you prove that your encoder is the most powerful?
Oliver Gunasekara: We have a mysterious advantage: unlike friends, we do not use software, we use programmable hardware in the cloud for coding, that is, FPGA. The advantages of using hardware design may be slightly unfair to others, so we can achieve the most powerful encoder.
Jan Ozer: Can you explain the specific meaning of the word "in the cloud"?
Oliver Gunasekara: Our coding service is based on AWS F1. The FPGA chip is provided in the F1 example, and we will arrange the HEVC encoder on it for the user to use.
Jan Ozer: Compared to running an encoder on other devices, how much does it cost to run on AWS F1 per hour?
Oliver Gunasekara: I remember the usage fee for F1 instances is $ 1.65 per hour. This cost covers an FPGA chip and 1/8 Intel E5 processors. The usage fee of Intel C4 processor is roughly the same. It seems to be around $ 1.50. Although the cost of the two is not much different, but in order to match the performance of the x265 encoder, we need to use 20 instances with a C4 processor, then it will cost more than $ 32 per hour, and currently (using the E5 processor ) It only costs about $ 3 per hour.
Jan Ozer: Which users are currently using your HEVC encoder?
Oliver Gunasekara: It is a pity that there are few users, and this is also the current problem. We therefore decided to provide VP9 encoding services at the same time. In our view, the market is currently showing a split trend. High-end markets tend to use HEVC encoders. As we all know, Apple supports the use of HEVC encoders. The iPhone is very popular, but due to the growing user base of Android and Chrome, the mass market is currently using a large number of VP9 encoders. In fact, almost all chipsets support both encoders. However, due to royalties, OEMs (original equipment suppliers) will only enable VP9 encoders. You will find that although a large percentage of the Android market supports HEVC encoding, VP9 encoding is supported by all Android and iOS markets. Only the latest devices in recent years support HEVC encoding.
Jan Ozer: What caused the market share of HEVC encoders to increase slowly? What is your opinion on this?
Oliver Gunasekara: Fundamentally, this is caused by the patent fee framework and return on investment. When you use software coding, the complexity of HEVC coding is much higher than AVC coding, so the coding cost is also higher. Especially when you are broadcasting live (real-time coding), you are limited by the computing power available on the device. So this may result in the final bit rate not much higher than H.264 encoding. Although your costs have increased dramatically, the gain in bandwidth gained is very, very small. At the same time you also bear the pressure of patent fees. We believe that through FPGA coding, the cost of coding services has been greatly reduced, while improving the quality of coding. It is hoped that this adjustment will accelerate the market's adoption of HEVC encoders and VP9 encoders.
Jan Ozer: Do you have any new news about the AV1 encoder?
Oliver Gunasekara: We have been in the AOM alliance for almost a year. We are very excited about this. Just like other technologies, (the AV1 encoder is introduced to the market) it will take some time. In our view, if there is not enough support for a large number of hardware decoders, it makes no sense to deploy AV1. By the end of the year, most mainstream browsers will support the AV1 decoding function, but before obtaining hardware support for mobile devices, the AV1 encoder has a shortcoming in power consumption, which cannot attract people's attention. It is expected that consumers will start using mobile devices that support AV1 decoding in 2020.
Jan Ozer: Given that AV1 encoding is very slow, your value proposition in hardware encoding should be quite firm, right?
Oliver Gunasekara: We plan to provide real-time AV1 encoding services while effectively improving our VP9 and HEVC encoding services. If only 5-10% improvement is made, I think this is not enough to attract users to use our service. We set the improvement target of the VP9 encoder to 25%, and our VP9 encoder benchmark is our most advanced HEVC encoder. But we will not rush to achieve success, this improvement will take some time. It is likely that we will find a solution before deployment next year. The final deployment will be in 2020, when it will be supported by the next generation of Xilinx FPGA chips.
Jan Ozer: What is your business model? How to make money?
Oliver Gunasekara: Our service is based on the SaaS model. Currently users can get a seven-day free trial on the AWS market. No fees will be paid during this period. Seven days later, users need to pay $ 1.25 per hour to use our encoder.
Jan Ozer: You do n’t provide hardware, right?
Oliver Gunasekara: Of course. We are not responsible for providing hardware. Users need to purchase the hardware by themselves, and then pay our usage fee to run our software on their hardware.
FUME EXTRA 1500PUFFS
Shenzhen Xcool Vapor Technology Co.,Ltd , https://www.szxcoolvape.com