Intel Agilex® 7 M-Series FPGA Network-on-Chip (NoC) User Guide

ID 768844
Date 12/04/2023
Document Table of Contents Parallel Computing NoC Application

In a parallel computing system you may use several types of processors and accelerators to optimally execute a computational effort, using task-based and data-based parallelism. Interconnect and memory performance play a key role in a parallel computing system where the NoC subsystem can provide high-bandwidth and low-latency communication between the processing elements and memories.

A parallel computing system may have data that transfers between an external host and the FPGA over a PCI Express* (PCIe) link. Once you bring data into the FPGA fabric, you can then store this data in HBM2e or external memory, for example DDR5, over the NoC subsystem.

Every processing element connected to a NoC can access all channels of the global memories attached to that NoC, regardless of the physical location of these channels. Processing elements often have highly sequential memory read access patterns that benefit from use of a feature in Intel Agilex® 7 M-series devices known as the fabric NoC. This feature allows the NoC to deliver read results via M20K memories that are located close to the PE logic in the FPGA core fabric.