Source: MHV 2022 Day 1Speaker: Alex Liu, Founder and COO of NetintContent Organizer: Yin Wenpei. The ASIC solution aims to replace existing software encoding methods, and the speaker discussed various advantages of using ASICs.
Table of Contents
-
Big Data Presentation
-
In-Depth Comparisons
-
Different ASIC Solutions
-
Common Features of ASICs
-
Comparison of ASIC, GPU, and Software Encoding
-
Product Showcase
-
Data Centers
Big Data Presentation
First, let’s list three numbers:
-
ASICs are nearly 20 times more efficient than general encoding software running on CPUs. -
ASICs reduce carbon emissions by 80 times compared to software running on CPUs. -
Encoding each channel of 1080p60 live video with ASICs costs only $100.
In-Depth Comparisons
Next, we will compare performance across three general platforms: CPU, GPU, and ASIC.
We will compare their performance in four aspects:
-
Video Quality -
Bit Efficiency -
Operational Complexity -
Integration Flexibility
In terms of video quality, CPUs perform the best (High), GPUs are moderate (Medium), while ASICs can provide low, medium, and high video quality levels based on user needs.
The comparison of bit efficiency follows the same pattern as video quality.
For operational complexity, ASICs are the lowest, followed by GPUs, and CPUs are the highest.
Regarding integration flexibility, ASICs excel, GPUs are the worst, and CPUs are moderate.
Different ASIC Solutions
Common ASIC methods:
-
Embedding chips into consumer electronics panels and deploying dedicated software for commercial purposes achieves low cost and rapid market entry; -
The middle option includes some special solutions, such as using secure cameras for live streaming, but the configuration is cumbersome; -
The third option is to customize video solutions based on demand, using the same chips but redesigning all encoding engines, which provides a better user experience (though it requires significant investment, typically in the millions of dollars).
Common Features of ASICs
The world’s first ASIC video engine developed for x86 and ARM architectures serves data centers via U.2, PCIe, or Ruler protocols.
It supports the widest range of video encoding formats: AV1, HEVC, H.264, etc.
It offers near-semiconductor performance without sacrificing quality, bit efficiency, or software flexibility.
With this architecture, you can plug the device into any server and use it immediately without installing any drivers.
This architecture is also very flexible, providing rich interfaces for developers to create new features and functionalities.
Comparison of ASIC, GPU, and Software Encoding
As we can see, the annual costs are $131,000 for ASICs, while GPUs (Nvidia T4) and software (Intel SVT) require $1,100,000 and $5,800,000, respectively.
In terms of carbon emissions, ASICs only produce 29.3, while GPUs and software solutions require 410 and 2170, respectively.
ASICs also require significantly less server density than the other two options.
Product Showcase
We have launched both the first and second generation products. The first generation supports HEVC, H.264 encoding, and PCIe protocol, G4 architecture.
The second generation supports AV1, HEVC, and H.264 encoding, U.2, PCIe protocols, G5 architecture.
Data Centers
ASICs not only provide services for individual users but also develop more advanced features for data centers.
Users can build a transcoding factory with our products, providing a VPU cluster for data centers, allowing any server user to access this VPU cluster as if it were local, enhancing efficiency by 30%-40%, which is very convenient.
Attached is the presentation video:
