Deep Dive into Math Libraries

NVIDIA’s GPU-accelerated Math Libraries, which are part of the CUDA Toolkit and the HPC SDK, are constantly expanding, providing industry-leading performance and coverage of common compute workflows across AI, ML, and HPC. We'll do a deep dive into some of the latest advancements in the Math Libraries, with a focus on accelerating sparse deep learning for structured and unstructured sparsity, and dense solvers and FFTs on multi-GPU and multi-node supercomputers. We'll also discuss our newer libraries, such as the device extensions that enable kernel fusion; new image processing and decoding technologies such as JPEG2000; newer NVIDIA Performance Libraries for Grace CPU; and direct sparse solvers. We’ll highlight how these library capabilities can help you exploit various new hardware capabilities, such as mixed-precision and sparse tensor cores on NVIDIA Hopper architecture GPUs, and particularly leverage the Grace Hopper platforms.