↓ Skip to main content
jeket.

Broadcast

Multicast at the Edge, Unicast at the Core
·9 mins
Engineering DMF MXL MXL Fabrics RDMA RoCE TCP Libfabric SMPTE-2110 Multicast Broadcast DGX Spark Networking
The core of a DMF is unicast, not multicast, and that changes the whole network problem. A study of MXL Fabrics over TCP and over RDMA verbs: latency, CPU, and the three different ways a media network breaks. The most useful result is that there is no single answer.
2110 at the Edge, MXL at the Core
·9 mins
Engineering SMPTE-2110 MXL DMF RDMA RoCE Libfabric NMOS Broadcast Kubernetes DGX Spark
It’s not 2110 versus MXL. It’s 2110 at the edge and MXL at the core, the two halves of the EBU’s Dynamic Media Facility. How MXL moves media with zero copies on a node and over RDMA between them, what we measured, and why you can bring your own media node without touching any of it.
Sizing the DGX Spark in Gigapixels
·9 mins
Engineering DGX Spark NVIDIA Blackwell SMPTE-2110 NVENC Broadcast ABR Encoding Sizing
ST2110 ingest, NVENC encode budget, and ABR ladder math for the DGX Spark, all expressed in a single universal unit: gigapixels per second.
SMPTE-2110 on the NVIDIA DGX Spark
·6 mins
Engineering SMPTE-2110 DGX Spark NVIDIA Blackwell Broadcast MOQ NMOS MXL Kubernetes
Why a $3.5K NVIDIA DGX Spark works as a real SMPTE-2110 ingest, processing, and encode node, and what it means for the economics of broadcast infrastructure.