From SIPB Cluedumps

Jump to: navigation, search

[edit] Introduction to Data-Parallel GPU Programming with CUDA

Date: November 9, 2010, at 3:00 PM
Presenters: Jiawen "Kevin" Chen
Location: 4-231
Abstract: Processor design very quickly approached a wall in the early 21st century. Due to power constraints, we can no longer increase clock speeds on processors while shrinking transistor sizes. The only way to increase performance today is to add additional processors, which demands a fundamentally parallel programming model. Although there are numerous forms of parallelism, "data parallelism" is by far the easiest to understand and exploit. The latest generations of graphics processing units (GPUs) are architectures designed for high performance on data-parallel tasks. This cluedump gives a tutorial on how to program a modern NVIDIA GPU using the CUDA API, with motivating examples in data analysis, image processing, and scientific computation. The session will be interactive with plenty of time for questions and live programming.
Personal tools