GPUs - Graphics Perspective

What is a Graphics Processing Unit (GPU)?

• Designed to handle more complex mathematical and geometric calculations required for graphics rendering

• On average, have more transistors and a larger density of computing cores with more ALU than a normal CPU

• It is a specialised processor designed to handle computations related to images, video, and graphics rendering

• Also used in AI, data science, and scientific computing

• Like "graphic engines" that help computers draw faster

GPU vs CPU

Feature	CPU (Central)	GPU (Graphics)
Purpose	General-purpose processor designed to handle a wide variety of tasks Optimised for sequential task execution and complex logical operations Ideal for running OSes, apps, and general computing tasks	Specialized processor designed for parallel data processing Developed for rendering graphics but now used for ML and video processing Optimized for tasks requiring high throughput
Architecture	Fewer cores (2–64) Each core is highly powerful and optimized for single-threaded performance Features a large cache memory for faster access to frequently used data	Thousands of smaller, less powerful cores Designed for parallel processing of large data sets Limited cache per core compared to CPUs(high-speed memory)
Performance	Performs better for tasks requiring high sequential processing or logic Handles I/O operations and general-purpose tasks effectively	Delivers superior performance for tasks that can be parallelized High floating-point calculation performance - ideal for simulations and AI workloads
Parallelism	Designed for low to moderate levels of parallelism (multi-threading or multi-core)	Massively parallel architecture, ideal for handling thousands of threads simultaneously
Memory	Access to large system memory (RAM) and a larger cache per core for fast data retrieval Lower memory bandwidth compared to GPUs	Uses high-speed, specialized memory High memory bandwidth enables fast data transfers during parallel processing tasks
Power Consumption	Consumes less power compared to GPUs for equivalent workloads	Consumes more power than CPUs, especially under heavy loads
Applications	Best for general-purpose tasks like running OS, browsing, word processing, and sequential applications	Best for specialized, parallel tasks like gaming, video rendering, machine learning &simulations Used in AI, NNs & high-performance computing
Cost	Relatively cheaper for general-purpose computing	More expensive, especially high-end GPUs used in gaming, AI, or data centers

What do GPUs Accelerate?

• Rendering 2D & 3D images

• Image transformations

• Lighting, shading, textures

• Motion and frame calculation

• Real-time updates (for games, simulations, etc)

GPU Architecture Basics

• Thousands of small cores for parallel processing

• Memory blocks (VRAM – Video RAM) to store pixel & vertex data

• Shaders: Small programs run on GPU to determine lighting, colour, etc

Types of GPU

Dedicated GPU :

• Separate hardware (card) installed in desktops

• NVIDIA RTX 4080

Integrated GPU:

• Built into the CPU

• Uses shared memory

• Intel UHD Graphics

Cloud GPUs:

• Provided by cloud services for remote computing

• Used in AI and large simulations

Shaders

Programs which run on the graphics card to do something and there are lots of different kinds

Types of Shaders

• Pixel/fragment shaders: these are run for every pixel on the screen; eg, change colour or brightness

• Vertex shader: run for every vertex being rendered to transform the 3D coord into a 2D screen coord; eg, normal perspective transform, distortion effect

• Geometry shader: run for every primitive (eg, triangle); eg, tessellate for smoother curves, do tricks for shadows

Why are Shaders important?

They allow developers to:

• Simulate lighting and shadows

• Create water, fire, fog, and glass effects

• Apply textures (images) to 3D models

• Animate objects in real-time

Shader Languages

Language	Used With	Purpose
GLSL	OpenGL	High-level shading language based on C Widely used in games and simulations
HLSL	DirectX (Windows)	Used in many PC/console games
Cg	NVIDIA GPUs	Cross-platform (now outdated)

Rendering

converting 3D models into 2D images

How do GPUs support in Rendering

• Project 3D → 2D

• Handle colour blending

• Light/shadow calculation

• Frame buffering

Result: Smooth, real-time visual output in games and software

Popular GPUs in Graphics

Nvidia GeForce

• Common in gaming and rendering

• Technologies: RTX (Ray tracing), DLSS (AI upscaling)

AMD Radeon

• Good performance for budget

• Supports OpenCL and DirectX

General Purpose GPU (GPGPU)

• This was the idea that you could accelerate non-graphics programs using a graphics card if they happened to use the same calculations

• This had to be done via “tricks” like encoding values into textures, uploading them, calculating the answer into a new texture, and downloading and decoding that into your data

• Lots of overhead but could still be faster

• Very quickly people and manufacturers realised this was a good thing and create ways of doing this without “tricks”

Why GPU is Important for you

• Essential to Modern AI and ML due to their ability to manipulate large data efficiently

• Essential for graphics and game development, understanding them is key in game development area

• Used in Cloud & Data Center Environments, it helps to use cloud resources efficiently

• Understand parallel algorithms (write GPU code using OpenCL) helps to understand parallel algorithms, memory hierarchy, and optimization, which are highly valuable for performance-critical applications

• Expands Career Opportunities  Proficiency in GPU computing opens doors in: AI/ML engineering, Data science, Scientific computing and Graphics and visualization

• Foundation for Future Technologies - Emerging fields like quantum simulation and AR/VR systems rely on GPU - Understanding GPUs prepares for future tech trends