WHITE PAPER
Massive Disaggregated Processing for Sensors at the Edge
mrcy.com 2 2
EDGE APPLICATIONS NEED FLEXIBLE,
HIGH-PERFORMANCE PROCESSING
Must be able to rapidly allocate, and re-allocate, computing
resources to process data steams for multiple applications
Advanced computing resources are moving from data centers
to edge systems, adding efficiency and new capabilities to
applications ranging from petroleum exploration to radar
signal processing. These high-performance edge systems
must be able to rapidly allocate, and re-allocate, parallel
processing resources to handle data streams from multiple
sensor sources through various types of algorithms, including
deep learning/machine learning neural networks for AI.
The system is the network
Networking speeds are keeping up with the constantly
expanding data streams, as communications standards like
PCIe Gen 5 and 200/400+ GbE delivering huge leaps in data
transfer bandwidth. Effective edge systems will exploit
those leaps, recognizing that data movement and data
stream processing are functions distributed throughout
a network; essentially, the system is the network.
Edge applications need support for more powerful, deployable computing subsystems
that can process extremely high bandwidth, ever-growing sensor data streams and exploit
the rapidly emerging capabilities of Artificial Intelligence (AI).
This paper highlights a novel architectural approach that addresses this growing need by
combining innovative data processor unit (DPU) technology with high-performance graphics
processing units (GPUs) in a rugged, SWaP-optimized configuration—without the need for an
x86 CPU host.
Data Center-Level Performance
