Release Notes: This release adds high-performance MPI communication from NVIDIA GPU device memory (to/from other devices and host memory) with IPC, collectives ,and datatype support; CPU binding granularity at socket and numanode level; checkpoint-restart and run-through stabilization with Nemesis; suspend/resume; and enhanced integration with SLURM and PBS. Network Fault Resiliency (NFR) has been added.
Release Notes: This release fixes a data validation issue in GPU transfers, tunes CUDA block size to 256K for better performance, enhances error checking for CUDA library calls, and fixes a mpirun_rsh issue while launching applications on Linux Kernels.
Release Notes: This release adds iWARP interoperability between Intel NE020 and Chelsio T4 adapters, space optimization in regards to buffer usage, and MPI communication from NVIDIA GPU device memory (including intra-node point-to-point communication for multi-GPU adapters/node and RDMA-based inter-node point-to-point communication to/from GPUs). Optimizations for collectives and one-sided communication.
Release Notes: Enhancements include a Nemesis-based interface, process manager support, rail binding with processes for multirail configurations, message coalescing, large data transfers, dynamic process migration, fast process-level fault-tolerance with checkpoint-restart, network-level fault-tolerance with Automatic Path Migration, RDMA CM support, iWARP support, optimized collectives, multi-pathing, RDMA read- and write-based designs, polling and blocking-based communication progress, multi-core optimized and scalable shared memory support, and LiMIC2-based kernel-level shared memory support.
Release Notes: The Shared-Memory-Nemesis interface was added, providing native shared memory support on multi-core platforms where communication is required only within a node. Support for 3D torus topology with appropriate SL settings was added. Quality of Service (QoS) support with multiple InfiniBand SL was added. GPU acceleration support was added. Fast Checkpoint-Restart support with an aggregation scheme was added. Fault tolerance support was improved. Dynamic detection of multiple InfiniBand adapters was implemented, and they are used by default in multi-rail configurations. Multithreading was enhanced. There were other enhancements and bugfixes.
Release Notes: This release adds message coalescing, hot-spot avoidance, application-initiated systems-level checkpointing, APM support, multi-rail support for iWARP, RDMA read, and blocking support. Assorted bugfixes.
Release Notes: This release adds RDMA CM-based on-demand connection management for OpenFabrics Gen2-* interfaces, uDAPL on-demand connection management, message coalescing support to enable reduction of per Queue-pair send queues, a Hot-Spot Avoidance Mechanism (HSAM) for alleviating network-congestion in large scale clusters, RDMA Read utilization for increased overlap of computation and communication for OpenFabrics devices, and support for an OpenFabrics Gen2-iWARP interface and RDMA CM.
Release Notes: Enhance udapl initialization. Minor bugfixes and code cleanups.
Release Notes: New message coalesing, hot-spot avoidance, application-initiated systems-level checkpointing, APM support, multi-rail support for iWARP, on-demand connection management for iWARP and uDAPL (including Solaris), RDMA read, and blocking support. The software was also updated to MPICH2 1.0.5p4.
Release Notes: This release added checkpoint/restart, OpenFabrics Gen2/iWarp, RDMA CM-based connection management support, shared memory optimizations for collective communication operations, and uDAPL support for the NetEffect 10GigE adapter.