NVIDIA HPC-X is a comprehensive software package that includes Message Passing Interface (MPI), Symmetrical Hierarchical Memory (SHMEM), and Partitioned Global Address Space (PGAS) communications libraries, along with various acceleration packages. This toolkit enables MPI and SHMEM/PGAS programming languages to achieve high performance, scalability, and efficiency, ensuring that communication libraries are fully optimized by NVIDIA Quantum InfiniBand networking solutions.

Features

• MPI: A high-performance, optimized implementation of Open MPI, leveraging NVIDIA’s acceleration capabilities for seamless integration with commercial and open-source application software packages.
• OpenSHMEM: A one-sided communications library supporting parallel programming features like point-to-point and collective routines, synchronizations, atomic operations, and shared memory paradigms.
• SHARP: NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol improves MPI operations by offloading them from the CPU to the switch network, reducing data traversal and MPI operation time.
• UCX: Unified Communication X is an open-source communication framework providing low-overhead, point-to-point communication paths for near-native, hardware-level performance with cross-platform support.
• UCC: Unified Collective Communication is an open-source framework for high-performance applications, leveraging InfiniBand In-Network Computing acceleration engines.
• NCCL/SHARP and UCX Support: NCCL-RDMA plug-ins enable RDMA and switch-based collectives with the NVIDIA Collective Communication Library, enhancing performance.
• ClusterKit: A node-assessment tool for high-performance clusters, ensuring health and performance through well-known techniques and tests.

Benefits

• Enhanced Performance: Offloads collective communications from MPI onto NVIDIA Quantum InfiniBand networking hardware, increasing CPU availability and application scalability.
• System Efficiency: Ensures node-level and system-level health and performance, maximizing application performance with underlying hardware architecture.
• Optimized Networking: Fully optimized for NVIDIA Quantum InfiniBand networking solutions, supporting any interconnect based on InfiniBand or Ethernet standards.