FCSCST1133Data Engineering Ecosystem

SCST1133 · Topic 01

GPU and Parallel Computing

Understand why CPUs coordinate systems while GPUs accelerate highly parallel workloads.

01Introduction to Computer and System
Advanced hardware

Match the processor to the workload

Performance depends on workload structure, memory access and data movement—not the processor name alone

CPU and GPU parallel processing comparison
Figure 17. CPUs emphasise flexible low-latency control while GPUs emphasise parallel throughput.
GPU system anatomy and data flow
Figure 18. GPU performance also depends on storage, system memory, interconnects and VRAM.

CPU and GPU comparison

DimensionCPUGPU
Primary strengthControl, serial logic and branchingMany similar operations in parallel
Core designFewer powerful, general-purpose coresMany throughput-oriented processing units
Memory emphasisLow-latency cache hierarchyHigh bandwidth to accelerator memory
Typical workOperating systems, applications and coordinationAI, graphics, simulation and analytics
System roleCoordinates the complete workflowAccelerates suitable parts of the workflow

Key idea

A GPU does not replace the CPU. The CPU prepares and coordinates work, while the GPU accelerates suitable operations.