Join us for an overview of the key features in the new CUDA Toolkit 7.5. Ujval Kapasi, director of product management, and members of the CUDA engineering team will share how you can store up to 2x larger datasets in GPU memory and reduce memory bandwidth requirements significantly using the new 16-bit floating point (FP16) data format. They’ll also cover the new cublasSgemmEx() routine that supports 2x larger matrices, cuSPARSE GEMVI routines for dense matrix x sparse vector operations that are ideal for natural language processing, and new instruction-level profiling capabilities that help you quickly pinpoint performance bottlenecks in your GPU code.

Finally, you’ll see some examples of the new (experimental) GPU lambdas feature in action.

CUDA Toolkit 7.5 Features Overview

REPLACE CONTENT WITH YOUR OWN

Presenter:

Ujval Kapasi, CUDA Product Manager, NVIDIA

Presenter Bio:

Ujval Kapasi is NVIDIA's CUDA Product Manager. Ujval received his Ph.D. in Electrical Engineering from Stanford University and his Bachelor of Science in Engineering from Brown University.