Heterogeneous System Architecture From A HPC Usage Perspective
Several programming models that are being developed by the HPC (High Performance Computing) community have recently considered extensions and additions to utilize GPUs implicitly and explicitly. This has been accomplished both within modifications to existing HPC languages and libraries and conception of new languages and libraries just to support heterogeneous systems. One common technique adopted by the HPC community is to extend an existing programming model to work heterogeneous computing systems. In this process, one of the biggest challenges is to maintain data consistency across their network, CPU and GPU usage. In this discussion we describe the details on how addressing such issues have been simplified by HSA (Heterogenous System Architecture) architecture, HSA runtime, and OpenCL 2.0, and what a HPC programmer needs to be doing to be able to utilize HSA features before and after OpenCL 2.0. MPI Applications, Global Arrays, OpenMP, OpenACC are utilized as programming model implementation examples in the discussion.
Speaker: Vinod Tipparaju
AMD Fusion Developer Summit 2012. For more information, visit: http://bit.ly/AFDS_2012
Subscribe
If you enjoyed this article, subscribe to receive more just like it.
