Shared Memory HW / CUDA Libraries

A talk in two parts:

Part 1: Shared memory Hardware.

http://photos3.meetupstatic.com/photos/event/5/6/2/4/event_310762052.jpeg

On how hardware can help share data among cluster nodes with minimal latency and maximum throughput through HW boards.

Part 2: CUDA libraries

There is a good momentum around CUDA development partly thanks to the effort Nvidia has put into developing generic + domain specific libraries and the contribs from the developer community.

http://photos4.meetupstatic.com/photos/event/3/a/e/a/event_260655082.jpeg

Julien Demouth (Nvidia DevTech) will present several generic and core libraries running over CUDA runtime including Hemi, Thrust, CUDPP, CUB, and NVPP, dealing more specifically with:

• their features and how to use them best
• how they can work together
• some pitfalls to avoid
• use in a production environment

Talk in French, slides in English.