# Graphics Processing Unit (GPU) ## What is a GPU? - A GPU is responsible for **processing graphics** within the computer to **reduce the load on the CPU** - CPUs are general purpose processors whereas GPUs are designed specifically for graphics - GPUs are likely to have built-in circuitry or instructions for common graphics operations - GPUs can perform an instruction on **multiple pieces of data at one time** - This is useful when processing graphics (e.g. transforming vertices in a 3D model or applying color filters to pixels) which means it can perform transformations to on-screen graphics quicker than a CPU - The GPU can either be **part of the graphics card** or **embedded in the CPU** - Modern GPUs typically contain **hundreds or even thousands of smaller processing cores**, allowing them to perform many operations in parallel

A GPU is a specialized processor designed to handle graphics and parallel computations more efficiently than a general-purpose CPU

## What can a GPU be used for besides graphics? Besides graphics processing, a GPU can also be used for: ### 3D modelling - The GPU can be used to render lighting effects, textures and shadows ### Data modelling - As GPUs can handle many calculations simultaneously, they can handle large datasets and complex operations like sorting and filtering data ### Financial modelling - GPUs are used to simulate different scenarios in risk modelling, option pricing and other financial modelling types - Lots of simulations can be run in parallel ### Data Mining - Data mining is the process of analyzing large amounts of data to find patterns - The main computational tasks are sorting, searching, pattern recognition, statistical analysis and graph algorithms ### Performing Complex Numerical Calculations - Matrix multiplication and inversion can be done in parallel - Numerical simulations - physics and engineering simulations often involve solving complex mathematical models, which can be done in parallel - Solving differential equations involves computations which can be performed in parallel ### Machine learning - This involves training a computer on a massive amount of data which can be done in parallel - There are lots of matrix multiplications and other computations which can be performed - After the training, GPUs can be used to speed up the process of making predictions on new data ### Calculations on multiple data at the same time - There are a number of scenarios where calculations will be needed to be carried out on multiple data at the same time e.g. weather forecasting, climate modeling, pricing derivatives - This is done by GPUs rather than CPUs due to being set up for parallel processing

You don't need to know how to perform these complex calculations, just that GPUs can be used for non-graphics tasks that involve parallel processing

## What types of task are GPUs suited for? GPUs are suited to certain tasks that utilize: ### Specialist instructions - GPUs are designed to execute specialist instructions which are common in 3D graphics rendering such as operations on matrices, vectors and geometric transformations - These capabilities have been expanded over time and have been generalized which makes GPUs suitable for a wide range of complex calculations besides graphics processing ### Multiple cores - Although a CPU can have multiple cores, these are optimized for serial processing - GPUs have smaller cores but these are optimized for parallel processing - GPUs can perform many calculations simultaneously - ideal for tasks that can be broken down into smaller parts - This is useful in machine learning and situations where large amounts of data need to be processed ### SIMD processing - Single Instruction Multiple Data (SIMD) processing is computers that have multiple processing elements which perform the same operation on multiple data points simultaneously - GPUs support SIMD processing as they were originally designed to perform the same operations on multiple pixels or vertices simultaneously - this is a common requirement in image processing, simulations and machine learning

SIMD (Single Instruction Multiple Data) processing executes the same operation on multiple data points simultaneously using multiple processing elements

## What are the benefits of using a GPU? There are a number of benefits to using a GPU as well as a CPU (it isn't possible to only use a GPU as the CPU assigns tasks to the GPU) ### Parallel processing - GPUs can handle many tasks simultaneously as they are multicore processors ### Speed - As GPUs can use parallel processing, this speeds up tasks, particularly those involving large amounts of data or complex computations ### Efficiency - GPUs can perform more calculations per unit of power consumed in comparison to CPUs making them more energy efficient when it comes to parallel tasks

A computer cannot run on a GPU alone - the CPU must assign tasks to the GPU and coordinate operations