What you need to know or do on the SCC<\/h2>\n\n- To use OpenACC, compile your Fortran code with the Portland Group Inc. (PGI) compiler,
pgfortran<\/span><\/code> (or pgf90<\/span><\/code>, pgf95<\/span><\/code>). You will need to load a module in order to use the PGI compiler:\nscc1%<\/span> module load nvidia-hpc\/2023-23.5<\/span><\/span><\/code><\/pre>\n<\/li>\n- After this, you can proceed with compilation. For example:\n
scc1%<\/span> pgfortran -o<\/span> mycode<\/span> -acc -Minfo<\/span> mycode.f90<\/span><\/code><\/pre>\nIn the above example, -acc<\/span><\/code> turns on the OpenACC feature while -Minfo<\/span><\/code> returns additional information on the compilation. For details, see the man page of pgfortran<\/span><\/span>:<\/p>\nscc1%<\/span> man pgfortran<\/span><\/code><\/pre>\n<\/li>\n- To submit your code (with OpenACC directives) to an SCC node with GPUs:\n
scc1%<\/span> qsub -l gpus=1<\/span> -b y mycode<\/span><\/code><\/pre>\nIn the above example, 1 GPU (and in the absence of a multiprocessor request, 1 CPU) is requested.<\/p>\n
Additional examples of GPU batch jobs are available here<\/a>.<\/li>\n<\/ol>\nDemonstration of Performance<\/h2>\n\n- The following examples demonstrate a matrix multiply (C = A * B<\/i><\/b>) using either multi-threaded OpenMP or OpenACC on a single GPU.
<\/span><\/code>\n\n- For OpenMP application:\n
scc1%<\/span> pgfortran<\/span> -mp matrix_multiply.f90<\/a> <\/span>-o mm_omp<\/code><\/pre>\n<\/li>\n- \n
For OpenACC application:<\/p>\n
scc1%<\/span> pgfortran<\/span> -acc matrix_multiply.f90 <\/span>-o mm_acc<\/code><\/pre>\n <\/i><\/p>\n<\/li>\n<\/ul>\n<\/li>\n- The following demonstrates timing comparisons for OpenACC, OpenMP, MPI:\n
![\"Bar](\"https:\/\/www.bu.edu\/tech\/files\/2013\/05\/gpu8ktime.jpg\")
Based on the given data, we can conclude that for 3 sets of four, eight, and sixteen CPUs, the following results were obtained using OpenACC, OpenMP, and MPI: For four CPUs, OpenACC took 19 seconds, OpenMP took 117 seconds, and MPI took 156 seconds. For eight CPUs, OpenACC took 19 seconds, OpenMP took 72 seconds, and MPI took 66 seconds. For sixteen CPUs, OpenACC took 19 seconds, OpenMP took 25 seconds, and MPI took 27 seconds.<\/p><\/div>\n
The above figure shows the timings comparison of a matrix multiply using a single GPU (via OpenACC) against two other parallel methods: OpenMP and MPI. The figure below shows the timings of matrix multiply using 1, 2, and 3 GPU devices.
\n![\"Bar](\"https:\/\/www.bu.edu\/tech\/files\/2013\/05\/acctime.jpg\")
<\/li>\n<\/ol>\n
OpenACC Tutorial<\/h2>\n
Please refer to the RCS tutorial slides for OpenACC programming<\/a>.<\/p>\nRelevant Links<\/h2>\n
pgfortran<\/span><\/code> (or pgf90<\/span><\/code>, pgf95<\/span><\/code>). You will need to load a module in order to use the PGI compiler:\nscc1%<\/span> module load nvidia-hpc\/2023-23.5<\/span><\/span><\/code><\/pre>\n<\/li>\n- After this, you can proceed with compilation. For example:\n
scc1%<\/span> pgfortran -o<\/span> mycode<\/span> -acc -Minfo<\/span> mycode.f90<\/span><\/code><\/pre>\nIn the above example, -acc<\/span><\/code> turns on the OpenACC feature while -Minfo<\/span><\/code> returns additional information on the compilation. For details, see the man page of pgfortran<\/span><\/span>:<\/p>\nscc1%<\/span> man pgfortran<\/span><\/code><\/pre>\n<\/li>\n- To submit your code (with OpenACC directives) to an SCC node with GPUs:\n
scc1%<\/span> qsub -l gpus=1<\/span> -b y mycode<\/span><\/code><\/pre>\nIn the above example, 1 GPU (and in the absence of a multiprocessor request, 1 CPU) is requested.<\/p>\n
Additional examples of GPU batch jobs are available here<\/a>.<\/li>\n<\/ol>\nDemonstration of Performance<\/h2>\n\n- The following examples demonstrate a matrix multiply (C = A * B<\/i><\/b>) using either multi-threaded OpenMP or OpenACC on a single GPU.
<\/span><\/code>\n\n- For OpenMP application:\n
scc1%<\/span> pgfortran<\/span> -mp matrix_multiply.f90<\/a> <\/span>-o mm_omp<\/code><\/pre>\n<\/li>\n- \n
For OpenACC application:<\/p>\n
scc1%<\/span> pgfortran<\/span> -acc matrix_multiply.f90 <\/span>-o mm_acc<\/code><\/pre>\n <\/i><\/p>\n<\/li>\n<\/ul>\n<\/li>\n- The following demonstrates timing comparisons for OpenACC, OpenMP, MPI:\n
Based on the given data, we can conclude that for 3 sets of four, eight, and sixteen CPUs, the following results were obtained using OpenACC, OpenMP, and MPI: For four CPUs, OpenACC took 19 seconds, OpenMP took 117 seconds, and MPI took 156 seconds. For eight CPUs, OpenACC took 19 seconds, OpenMP took 72 seconds, and MPI took 66 seconds. For sixteen CPUs, OpenACC took 19 seconds, OpenMP took 25 seconds, and MPI took 27 seconds.<\/p><\/div>\n
The above figure shows the timings comparison of a matrix multiply using a single GPU (via OpenACC) against two other parallel methods: OpenMP and MPI. The figure below shows the timings of matrix multiply using 1, 2, and 3 GPU devices.
\n<\/li>\n<\/ol>\n
OpenACC Tutorial<\/h2>\n
Please refer to the RCS tutorial slides for OpenACC programming<\/a>.<\/p>\nRelevant Links<\/h2>\n