Explain how multicore processors enable parallel processing
Describe the structure and purpose of a GPU
State Amdahl’s Law and explain its implications
Describe co-processors and their role in specialist processing
Multicore
Multicore Processors
A multicore processor contains multiple independent CPU cores on a single chip. Each core can independently fetch, decode and execute instructions, enabling true parallel execution of different threads.
Advantages
Parallel execution of multi-threaded applications. Better performance per watt than a single fast core. Multiple tasks (e.g. OS + application) run simultaneously.
Limitations
Benefits only realised if software is multi-threaded. Inter-core communication adds overhead. Not all problems can be parallelised.
Amdahl’s Law
Amdahl’s Law
Amdahl’s Law states that the speedup from parallelism is limited by the sequential fraction of the program that cannot be parallelised.
If 90% of a program is parallelisable and 10% is sequential: maximum speedup with infinite cores ≈ 10× (1 / 0.1). Adding more cores beyond a point yields diminishing returns.
Implication: parallel programs must minimise serial sections. A 50% serial program cannot be sped up by more than 2× regardless of core count. This is why software design is as important as hardware.
GPU
Graphics Processing Units (GPUs)
A GPU contains thousands of smaller, simpler cores designed for SIMD (Single Instruction, Multiple Data) execution — the same operation applied to many data elements simultaneously.
Designed for highly parallel tasks: rendering pixels, 3D geometry transformation, physics simulation.
Also used for GPGPU (General Purpose GPU) computing: machine learning (matrix operations), scientific simulation, cryptocurrency mining.
CPU cores: 4–64, powerful, general purpose. GPU cores: 1,000–10,000+, simple, parallel-specialist.
Co-processors
Co-processors
A co-processor is a specialised processing unit that assists the main CPU with specific tasks, offloading computationally intensive work to dedicated hardware.
FPU (Floating Point Unit): dedicated floating-point arithmetic; integrated into modern CPUs.
NPU (Neural Processing Unit): accelerates AI/ML inference; found in modern SoCs (Apple Neural Engine, Google Tensor).
DSP (Digital Signal Processor): optimised for audio/signal processing in embedded systems.
Co-processors free the main CPU for general tasks while specialist work runs simultaneously on optimised hardware.
Exam Practice
OCR H446 Style · 4 marks
Explain why GPUs are better suited than CPUs for rendering graphics. Include reference to the architecture of each in your answer.
[4 marks]
1
GPUs have thousands of simpler cores compared to the small number of powerful cores in a CPU.
1
GPUs use SIMD — the same instruction is applied to thousands of data items (pixels) simultaneously.
1
Rendering requires many identical operations on independent pixel data — highly parallelisable.
1
CPUs are designed for complex, sequential tasks with branching; the GPU architecture matches the parallel, data-regular nature of graphics rendering.
Common Mistakes
Don’t Lose Marks
!
Saying a GPU replaces the CPU — GPUs complement CPUs. The CPU handles general sequential tasks; the GPU handles parallel data-processing tasks. Both are needed.
!
Ignoring Amdahl’s Law when asked about parallel processing benefits — always mention that the sequential portion limits maximum speedup, regardless of core count.
!
Saying multicore processors always double speed with each added core — inter-core communication overhead, memory bandwidth limits, and Amdahl’s Law all cap the actual gain.