How to maximize HPC productivity, performance, and portability (PRT220)
summary
NVIDIA HPC SDK installs in minutes from developer.nvidia.com/hpc-sdk or AWS Marketplace and is free. It compiles the same code for x86, Arm and Power (with or without GPUs) using -stdpar=gpu or CPU modes, and shows concrete wins: rewriting with standard-parallelism yields 2–10x speedups; A100 runs a C++ example ~60x faster than an OpenMP GCC baseline; Graviton3posts ~1.45x runtime and ~1.7x price-performance versus Ice Lake; Graviton/T4 combinations give the best cost performance in several miniapp studies.
AI-written summary · about
chapters (15)
- 0:00Intro
- 2:17The NVIDIA platform
- 3:03The NVIDIA developer ecosystem
- 3:57NVIDIA HPC software initiatives
- 8:11NVIDIA HPC SDK is ready for Arm
- 9:36NVIDIA HPC SDK supports AWS Graviton
- 10:33NVIDIA HPC compilers
- 12:03NVIDIA HPC SDK programming models
- 14:54Accelerated standard languages
- 16:47C++ standard parallelism
- 18:54Fortran standard parallelism
- 20:38NVIDIA math libraries
- 22:55cuSPARSE improvements in key HPC kernels
- 29:03NVIDIA HPC SDK in the AWS Marketplace
- 29:57Installation on Amazon Linux 2
related talks
34mHPC on AWS: Innovation without infrastructure constraints
20mAWS On Air ft. AWS Graviton3 Processors and Amazon EC2 C7g Instances
22mGraviton
21mState of ARM-based HPC
01h 00mfor virtually every application
46mWhen AI Meets High Performance Computing
47mCloud Native and Sustainable, Reproducible Scientific Computing
40mThe POWER of Supercomputers at Your Fingertips