MSVP: Meta's Scalable Video Processor
summary
MSVP is a server-grade video accelerator that processes up to 1 billion pixels per second while consuming less than 10 W, implemented as a 100 mm² 7 nm SoC with the encoder occupying >50% of die area. It decodes H.264/H.265/VP9/AV1, encodes H.264 and VP9, uses 25-tap 2D scalar filters, a three-stage motion search with ±512 horizontal, ±160 vertical search and parallelized RDO/rate estimation, and runs quality metrics (SSIM, MS-SSIM, VIF, PSNR, blur) ~25× per upload; results yield 9× faster with half the energy versus software.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:44Data-center processing (transcoding) of videos
- 2:04Quality, Egress and Compute Tradeoff for Video Encoding System
- 3:04Transcoder Pipeline
- 5:18Encoder Architecture
- 7:55System Level Overview
- 8:38Advanced ABR Encoding Family
related talks
20mBoosting Video Codec With RISC-V Vector Extension
33mArchitecting Scalable Multi-Modal Video Search
12mMeasuring live video quality with minimal complexity, now available for everyone!
24mSuper Resolution: The scaler of tomorrow, here today!
17mGemino: Practical and Robust Neural Compression for Video Conferencing
21mReal-Time Video Super Resolution for Live Streaming with WebGPU
22mAI Enhanced GPU Video Coding: Achieving Joint High Compression Efficiency and Throughput
24mAV1 at Netflix
