General purpose computers have completely revolutionised the modern world. Their beating heart — the CPU — can perform any computable task and has been getting ever faster since it’s inception decades ago. However, this generality comes at a cost, and the seemingly never-ending CPU speed increases have seemingly come to an end.
This is where the concept of a hardware accelerator comes in, which gives up generality to gain extra speed at a more specific set of tasks. We discuss AI hardware accelerators, which have recently been undergoing a renaissance, and have enabled calculations, and thus applications, that were otherwise completely unfeasible.
Specifically, we cover AI accelerators from widely-available GPUs to custom-programmable FPGAs to bespoke ASICs, such as Google’s shiny new TPUs that are specially designed for deep learning tasks. To explain how these systems can run circles around our trusty CPU friend, we relate how they work to what happens under the hood of commonly-used machine learning algorithms.
Finally, we briefly touch upon possible future accelerator technologies such as optical processing units (OPUs) and quantum processing units (QPUs).
After learning what computer is best for machine learning, we will look at how to get hold of some of this power without having to buy actual hardware. We will review why cloud based computing might be a good option, look at what the main providers are offering, explore some of the services and tools that are on offer and look at how to get started.
Yolk Recruitment are kindly sponsoring the food* and drink, Tramshed Tech are graciously providing the venue, and our thanks to ARTIMUS for their continued sponsorship of the group.
Look forward to seeing you there!
*please message us >in advance< with any specific dietary requirements – we usually provide Domino’s pizzas, as well as beers, cider and soft drinks