Beyond Shady AI with OpenCL: Massively Parallel Algorithms on GPGPUs

Today, even mobile devices include powerful GPUs, servers often include clusters of GPGPUs, and the latest generation of consoles contain shiny compute units. What opportunities do these recent changes in hardware offer AI and game programmers? Which techniques would reduce the amount of power your game uses, speed up offline pre-calculations by an order of magnitude, or significantly increase the amount of data you can process in real-time? This in-depth tutorial will explain (almost) everything you need to know about modern GPUs, how they perform on a variety of AI algorithms, and how best to leverage their power from standard languages such as OpenCL.

Using examples in decision making, visibility and path-finding, you'll see where there's low hanging fruit for massive parallelism as well as applications to avoid completely! You'll also discover what's in store for the near future, such as unified architectures, and turning standard C++ programs into hybrid compute kernels.

Talk objectives:

This session is intended for programmers keen on improving the performance of their games/code with new hardware, or simply reduce power consumption. It will cover details about GPU hardware, API concepts behind compute languages like OpenCL, and how to prepare for the future to integrate such techniques into C++ codebases.

Target audience:

Anyone technically-minded interested in artificial intelligence for games, or game programmers unfamiliar with OpenCL or GPUs keen to learn something new!