Applications

The encoder learns to watch.

YUAN's next-generation 4K60 smart encoder fuses broadcast-grade capture, AV1 compression, and on-board AI — so the stream describes itself as it is being made.

15 µs: capture latency with NVIDIA GPUDirect RDMA
4K60P: broadcast-grade capture — 4:4:4, 10-bit, VANC
70/100: TOPS options on Jetson Orin NX for in-encoder AI

The challenge

More pixels, more hours, and someone has to tag it all.

Broadcast television is moving from HD to 4K and from 4G to 5G at the same time. Every hour of programming becomes H.264/H.265 streams on local or cloud storage — massive volumes of data to hold and convert, and extensive video content that still needs manual tagging and labelling before anyone can edit or replay it.

Meanwhile AI edge computing has matured to the point where video and text convert fluently in both directions. Encoding and AI are no longer separate domains; the encoder is the natural place for them to merge.

More pixels, more hours, and someone has to tag it all.

The capture

True 4K60P, fifteen microseconds from the lens.

The encoder pairs NVIDIA Jetson Orin NX with YUAN's SC710 capture cards: 4:4:4 chroma subsampling, 10-bit depth, and VANC capture — genuinely broadcast-grade acquisition, not a scaled-up consumer path. NVIDIA GPUDirect RDMA takes overall capture latency down to 15 microseconds.

Jetson Orin NX brings AV1 compression: quality comparable to H.265 at lower bitrates, which cascades into real savings on bandwidth, storage, and transmission across the whole chain.

True 4K60P, fifteen microseconds from the lens.

The transport

Speak every protocol in the truck.

YUAN's NexVDO SDK integrates the broadcast industry's transmission standards in one place: RTSP, RTMP, TS, NDI, SRT, Dante AV, and YUAN's own SkyLink X ultra-low-latency protocol. Capture, encode, and transmit become one streamlined process rather than three vendors' boxes.

CBR streaming keeps network transmission rates under stable control — the unglamorous requirement that separates a broadcast product from a demo.

The intelligence

Metadata written while the frame is still warm.

With 70 or 100 TOPS on board, the encoder runs AI and generative AI algorithms during the encode itself. Stream Information Extraction compresses audio encoding and AI-generated information in synchronisation, embedding what the AI sees into the stream's metadata as it is written.

For post-production, that changes the job: rushes arrive already searchable, important timestamps already marked. The fusion of AI and 4K encoding redefines what the next generation of encoders is specified to do.