About Tech. Prog. Muti-Core ARM Proc Coprocessors GPUs HPC Cluster App. Kernels Registration

• Topics of Interest • Tech. Prog. Schedule • Topic : Multi-Core • Topic : ARM Proc • Topic : Coprocessor • Topic : GPGPUs • Topic : HPC Cluster • Topic : App. Kernels • Lab. Overview • Key-Note/Invited Talks • Home




hyPACK-2013 Topics of Interest : Mode-5 - HPC GPU Cluster

The Mode-5 programme focus is to understand programming issues of HPC GPU cluster and develop codes using different programming on host-cpu and device GPUs. The programming such as MPI, Pthreads and OpenMP with CUDA enabled NVIDIA GPU as well as OpenCL (AMD/NVIDIA) will be used. The tuning and performance aspects of programs on HPC GPU Cluster will be carried out. Topics of interest are listed below.


Mode-5 : HPC GPU Cluster : Programs based on host-cpu and devices GPUs (CUDA/OpenCL)

  • Simple example programs on Multi-Core Processors with NVIDIA - GPU Computing CUDA SDK will be made available.

  • Write Programs for performance Characteristics of host-memory, PCIe bus and network interconnect performance of HPC GPU Cluster

  • Programs on Matrix Computations based on MPI, Ptheads, OpenMP on host-cpu and CUDA enabled NVIDIA GPUs & OpenCL (AMD & NVIDIA) on device GPUs.

  • Measurements of bandwidths of PCIe Gen 2 X16 Slots of HPC GPU Cluster using CUDA /OpenCL APIs.

  • Low level Benchmarks measuring performance characteristics focusing on one-to-one ratio of CPU cores to GPUs of HPC GPU Cluster.

  • Test Suites focus on CPU-GPU single node performance on HPC GPU Cluster using CUDA /OpenCL prog.

  • Test Suite focus on Resource Allocation for Sharing and Efficient Use based on CUDA Wrapper library and GPU device Virtualization

  • Development of Programs based on host-cpu (Pthreads) and devices (CUDA) using assigning unique GPUs to host threads based on CUDA compute-exclusive mode & normal mode as well as affinity mapping.

  • Development of programs to check the health of the HPC GPU Cluster pre-job allocation and post-job de-allocation in HPC GPU Cluster programming environment

  • Development of programs to check the resources available for all devices in each node of HPC GPU Cluster

  • Development of MPI-CUDA Programs on HPC GPU Cluster, allotting one or more MPI threads per each or multiple GPUs

  • Demonstration of Open Source Software NVIDIA - MAGMA & Numerical Linear Algebra ( LINPACK) Benchmarks on HPC GPU Cluster.

  • Test programs based on MPI on host-cpu and using host-memory memory (pinned/pageble) of CUDA enabled GPUs on device GPUs in HPC Cluster environment.

  • Develop MPI based test suites to launch multiple kernels on CUDA enabled NVIDIA single and multiple GPU devices on each node of HPC GPU Cluster.

  • Develop MPI-CUDA based test suites on HPC GPU Cluster to launch multiple kernels on cUDA enabled NVIDIA single & multiple GPU devices on each node of HPC GPU Cluster.

  • Develop test suites on HPC GPU Cluster based on MPI programming on host-cpu and OpeNCL prog, on device GPUs to launch multiple kernels on GPU devices on each node of HPC GPU Cluster

  • Develop test suites to monitor health of a HPC GPU Cluster with OpenCL programming environment focusing on GPU device memory availability on single & multiple GPU devices

  • Develop test suites to monitor health of a HPC GPU Cluster focusing on inter-job memory cleaning for pre- job allocation and post-job de-allocation situations in HPC Cluster environment based on OpenCL programming environment.

  • MPI Test suites for sharing GPUs among multiple cores of HPC GPU Cluster based on OpenCL programming environment.




Centre for Development of Advanced Computing