FCSCST1133Data Engineering Ecosystem

SCST1133 · Topic 01

Supercomputer Architecture

See how thousands of compute nodes cooperate through high-speed networks, storage and software.

01Introduction to Computer and System
Advanced infrastructure

A supercomputer is a system of systems

Its processors become useful only when compute nodes, interconnects, storage, cooling and parallel software operate together.

Architecture of a modern supercomputer
Figure 20. Compute nodes combine CPUs and accelerators through a high-speed system fabric.

Architecture layers

LayerRole
Compute nodeExecutes a portion of a parallel workload using CPU, GPU and memory.
InterconnectMoves messages and data between nodes with low latency.
Parallel storageFeeds large datasets to many nodes and stores results.
SchedulerAllocates nodes and execution time to approved jobs.
Power and coolingMaintains reliable operation under sustained load.
Parallel softwareDivides work and coordinates results across nodes.

Frontier example

Frontier at Oak Ridge National Laboratory uses more than 9,400 compute nodes. Each node combines one AMD EPYC CPU with four AMD Instinct MI250X accelerators.