Graphics Processing Unit

• What is a GPU? #1

  • A graphics processing unit (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
• What is a GPU? #2

  • This is useful when processing graphics (e.g. transforming points in a polygon or shading 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 
  • A GPU will usually be multicore and can have up to 76 cores 
• What can a GPU be used for besides graphics? #1

  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 
• What can a GPU be used for besides graphics? # 2 

  Data Mining 

  • Data mining is the process of analysing large amounts of data to find patterns 
  • The main computational tasks are sorting, searching, pattern recognition, statistical analysis and graph algorithms 
• What can a GPU be used for besides graphics?  #3

  Performing Complex Numerical Calculations 

  • Matrix multiplication and inversion can be done in parallel 
  • Numerical Simulations 
  • Physics and engineering simulations often involve solving complex maths models, which can be done in parallel 
  • Solving Differential equations 
  • Solving differential equations involves computations which can be performed in parallel
• What can a GPU be used for besides graphics? #4

  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 
• What can a GPU be used for besides graphics? #5

  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. insurance pricing, modelling risk, calculating bills 
  • This is done by GPUs rather than CPUs due to being set up for parallel processing 
• What types of task are GPUs suited for? #1

  GPUs are suited to certain tasks that utilise: 

  • 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 generalised which makes GPUs suitable for a wide range of complex calculations besides graphics processing 
• What types of task are GPUs suited for? #2

  • Multiple cores  
  • Although a CPU can have multiple cores, these are optimised for serial processing 
  • GPUs have smaller cores but these are optimised 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
• What types of task are GPUs suited for? #3

  • 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 
• 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