Nvidia upgrades PGI 2014 accelerator compile suite
OpenACC C++ support advances the cause of GPU-based supercomputers
Just a few months after first releasing its PGI 2014 set of compilers and tools, Nvidia subsidiary The Portland Group (PGI) has expanded OpenACC C++ support in the high-performance computing compile suite, and put older versions of Windows on the EOL list.
The PGI 2014 release back in February added support for the latest version of the OpenACC acceleration offload language to the platform, as well as support for Nvidia's Tesla and AMD's Radeon GPUs and APUs.
The new 14.4 release is now available for Linux (workstation, server and a cluster development kit), Windows (workstation and server, and a Visual Fortran version), and a workstation version for OSX.
The full list of features in PGI 2014 14.4 is here, but high points include new CUDA 5.5 and CUDA 6 features, GPU-side debug in OpenACC under Allinea DDT, better PGI multi-core performance on Sandy Bridge, and integrated prebuilt versions of NetCDF and HDF5.
Nvidia got into the compiler business in 2013 with the acquisition of The Portland Group, in recognition of the growth of GPU-based supercomputers.
Nvidia and PGI were two of the original partners in setting up OpenACC, which El Reg then noted is “an open standard for adding directive hints to compilers to help them parallelize applications for CPUs, GPUs, and any other kind of parallel execution engine.”
PGI 2014 is a big chunk of this strategy, since it aims to decouple the accelerator programming from specific hardware, instead letting HPC users work with accelerators on a more generic level and skip device-specific tuning. ®