The NVIDIA HPC-Benchmarks collection provides four benchmarks widely used in the HPC community optimized for performance on NVIDIA accelerated HPC systems. These benchmarks include HPL-NVIDIA, HPL-MxP-NVIDIA, HPCG-NVIDIA, and STREAM. The benchmarks are designed to measure and optimize the performance of high-performance computing systems equipped with NVIDIA GPUs and CPUs.

Features

• HPL-NVIDIA: Solves a dense linear system in double precision arithmetic on distributed-memory computers equipped with NVIDIA GPUs.
• HPL-MxP-NVIDIA: Uses mixed precision arithmetic with Tensor Cores for solving dense linear systems.
• HPCG-NVIDIA: Performs multigrid preconditioned conjugate gradient iterations using double precision floating point values.
• STREAM: Measures sustainable memory bandwidth with synthetic benchmark programs optimized for NVIDIA GPU architectures.

Benefits

• Optimized Performance: Ensures high performance on NVIDIA accelerated HPC systems.
• Multinode Support: Supports multinode configurations for scalable benchmarking.
• Multi-Architecture Compatibility: Compatible with NVIDIA Ampere, Hopper, Blackwell GPU architectures, and NVIDIA Grace CPU.
• Comprehensive Tools: Includes scripts, sample input files, and tuning guides for efficient benchmarking.