Video Processing Unit (VPU) IP

VeriSilicon's Hantro VC9800D enables up to 256 streams decoding with robustness, high throughput single-core solution, or multi-core solution, supporting VVC, AV1, HEVC, H.264, AVS3, AVS2, VP9 video formats, JPEG, and legacy formats.
 

VeriSilicon's Hantro is based on silicon-proven VeriSilicon Hantro technology and target to demand the high number steams, high throughput, low bandwidth data centers to address the video transcoding, AI server, desktop virtualization even cloud gaming applications.
 

Hantro VC9800D provides semiconductor manufacturers a minimum risk solution for integrating high-performance video capability into their chips.

Video Formats

• Unified architecture supports multiple formats
• VVC Main Profile 8/10bit
• AV1 main profile
• HEVC main10, main, and main still profiles
• H.264 up to constraint high10 profile
• VP9 profile 0 and profile 2 (10-bit)
• AVS3.0 main-10 bit profile
• AVS2.0 main/main10 profile
• JPEG (Include Progressive JPEG)
• Legacy formats: MPEG4, MPEG2, MPEG1, VC-1, H.263, VP8, VP7, VP6, RV10, RV9, RV8, AVS, AVS+, Sorenson, DIVX3/4/5/6, WebP

High Performance and Throughput

• Up to 8K@30fps/16x1080P@30fps performance with a single-core
• Up to 8K@120fps/64x1080P@30fps with multiple cores
• High BUS latency tolerance without mainstream decoding format performance impact
• Up to 256 streams in a single device
• Maximally offloads the system CPU
• Leading in PPA/Die Area for high performance solution

Versatile Inline Post Processing

• Bandwidth efficient solution by inline processing
• Multiple post processing outputs
• High quality resize and programmable scaling kernel
• RGB formats for AI models
• Compatible VSI GPU rendering formats for high working efficiency with VSI GPU
• Rich post-processing capability, such as blending, image stitch, color remapping, rotation, flip, etc

Features for Different Application Requirements

• Based on frame level switching of virtual machine tasks, providing ultra-low latency task scheduling capabilities for cloud computing applications
• Error concealment and robustness
• Independent multiple Post-processing channel for Video transcoding
• Support compatible AI input NCHW(Planar)/NHWC(Packed) output to best fit AI applications
• Matching OpenCV resize precision for AI applications
• Normalization for AI pre-processing
• DDR efficiency and bandwidth saving
• Low-latency decoding
• Security and DRM support

Flexible HW Configuration

• HW configuration can be done for most video formats
• Typical 1-4 cores or more in a device
• PP different channels can have different features and restriction
• Data caching and burst shaping for good DDR efficiency in specific system
• Bandwidth saving by reference frame & output frame compression
• Optional MMU, encryption

API support

• OpenMax-IL
• FFmpeg avcodec Plug-in
• VAAPI
• Hantro Video Private ctrlSW API
• V4L2 API