Jensen Huang on the new HPC at SC18, 2018
1:11:38 “So before I go into some new stuff, the new HPC market, the new HPC market which started out with scientific computing with some early Internet companies playing with distributed computing and Hyperscale which then opened up the whole open source started with Hadoop and MapReduce and then the Berkeley work, Berkeley AMP Lab with Mate, Zaharia with Spark and…” Share
NVIDIA CEO Jensen Huang’s SC18 keynote sets out his view of a changing high-performance computing market. He links the end of Moore’s Law to GPU acceleration, mixed precision and AI; introduces the RAPIDS data-science stack built around Apache Arrow and Dask; demonstrates DGX-2 scale-up computing and NGC containers; and presents the T4 GPU for cloud inference, including a Google Cloud deployment.
Passages
14:42 “I think that the next two major breakthroughs on top of acceleration are going to be the combining of multi precision mixed precision computing, iterative solvers, as well as deep learning and AI for preconditioning to advance and to accelerate the resolution of the refinement. Really, really exciting work.” Share
18:29 “In order for us to achieve speed up these days, we have to design everything from the ground up for acceleration, but work with application developers in what the supercomputing industry calls co development to think about the entire stack, from the application to the solvers, to the libraries, to the architecture of the system, to the system architecture of…” Share
35:43 “Rapids is the world’s first fully GPU accelerated data science pipeline. From beginning to end, from the ingestion of data to the data processing, the joins, the filters, the group buys to machine learning, XG boost to random forest, K mean and K nearest.” Share
Search what Huang said from Claude, ChatGPT or your own code. See plans