Building A 300 Channel Video Encoding Server

NETINT, Supermicro and Ampere work together to create a high-performance, low-cost live video server

Overview

The demand for high-definition live videos has surged, bringing huge pressure to operational costs and user experience. Traditional x86 processors struggle to cope with the intensive video processing tasks required by modern streaming media. NETINT, Supermicro and Ampere Computing collaborated to redesign video transcoding servers for higher performance, lower cost and higher efficiency.

Solution

NETINT combines their Quadra VPU with Ampere's Altra Max processor to create a smaller, faster, and more cost-effective server. This new server architecture supports advanced video processing capabilities, including AI inference tasks and automatic subtitle generation using OpenAI's Whisper.

Key Features

• High performance: Ability to transcode multiple video streams simultaneously (for example, 95 1080i30 streams, 195 720i30 streams).
• High cost-effectiveness: Operational cost reduction is 80% compared to traditional x86-based solutions.
• Advanced processing: Support interlaced conversion, software decoding and AI inference tasks.
• Flexible control: Manage through FFmpeg, GStreamer, SDK or NETINT's Bitstreams Edge application interface.

Technical Innovation

• Customized ASIC: NETINT's proprietary ASIC for high-quality, low-cost video processing.
• Ampere Altra Max Processor: Providing unprecedented efficiency and performance, optimized for intensive computing environments.
• Optimization software: Utilize the latest FFmpeg version and Arm64 NEON SIMD instructions to significantly improve performance.

Achievements

The result of this cooperation is a breakthrough live video server with the following advantages:

• The throughput is 20 times higher than that of the software on x86.
• Operational costs have been greatly reduced.
• Extend system functions and support video formats that are not native to NETINT VPU.
• Real-time and accurate live transcription through automatic subtitle generation.

Detailed explanation of architecture

NETINT VPU is responsible for intensive video encoding and transcoding processing, while powerful host CPUs can perform other functions that the VPU does not support on hardware, such as interlaced conversion and software decoding, and AI inference tasks. NETINT recently announced the industry's first automatic subtitle feature using OpenAI Whisper, which is optimized for the Ampere Altra Max processor to enable real-time and accurate transcription of live broadcasts. The server performs video interlaced conversion and transcoding in a high-density, high-performance, and cost-effective manner that the x86 processor cannot achieve.

Video engineers can control the server through FFmpeg, GStreamer, SDK or NETINT's Bitstreams Edge application interface, which facilitates deployment and replacement of existing transcoding resources or use in new projects.

This case study discusses how NETINT, Supermicro and Ampere engineers optimize systems to provide a redesigned video server. This server can transcode 95 1080i30 streams, 195 720i30 streams, 365 576i30 streams, or combine 100 576i, 100 720i, and 10 1080i, 40 1080p30, 40 720p30, and 10 576p streams. The server decodes 96 input 1080i30 H.264 or H.265 streams with 96 inputs, and 320 inputs 1080i MPEG-2 streams, extending system functionality.

"The combination of Ampere Altra Max processor and NETINT VPU makes Supermicro 1U servers show new value."

——Alex Liu, co-founder of NETINT

NETINT selection

To address customer concerns about limited CPU processing capacity and soaring power consumption costs, NETINT has developed custom ASICs with the goal of achieving the highest quality and lowest cost video processing and encoding. NETINT redesigned the live video transcoding server by combining the NETINT Quadra VPU with Ampere's Altra Max processor to create a smaller, faster server with 80% lower operating costs and 20 times faster throughput.

Performance Challenges and Optimization

The initial results, while encouraging, failed to meet the goals set by NETINT. Through optimization of hardware and software, the team finally achieved the goal, and the Ampere Altra Max processor has only 50-60% CPU utilization, leaving room for future features. The article describes in detail the process of using mpstat, BCC softirq tools and perf tools for performance analysis and optimization, and ultimately solves the performance bottleneck problem.

Results and Conclusions

The Ampere Edition of the NETINT 300 channel live video server, jointly launched by NETINT, Supermicro and Ampere, achieves unprecedented performance and efficiency in a Supermicro 1U server chassis. The server also supports AI inference processing, such as NETINT's first automatic subtitle feature using OpenAI Whisper, which was shown at NAB 2024.

(Image description: FFmpeg Arm64 interlaced conversion optimization, after upgrading from FFmpeg 5.0 to 6.0, the performance improvement on Ampere Altra Max processor is as high as 2.9 times.)

The article ends with NETINT, Supermicro and Ampere companies, and provides more information to obtain.

The above is the detailed content of Building A 300 Channel Video Encoding Server. For more information, please follow other related articles on the PHP Chinese website!