Making HPC Programming More Accessible with Cross-Platform Tools

Over the last few months, the Infrastructure team in Arm has been very busy, bolstered by a very warm and enjoyable summer! Some of the key news include the announcement of Arm Neoverse, exciting improvements in the GNU toolchain and the release of Allinea Studio 19.0! On the field, Arm hardware is gaining traction and key HPC sites have started to use production-grade clusters with Marvell's ThunderX2 cores, to the delight of the scientific community. 

And this is what matters to us! Within the Arm development tools team, the HPC community remains one of our core focus. We endeavour to serve researchers and computer scientists alike, and help them write and run applications efficiently on the hardware that is best suited to their needs, whatever architecture this might be.

In Forge and Performance Reports 18.3, we have expanded the capabilities of our profiling and performance analysis tools to support the latest Cray XC50 systems, which are being deployed throughout the world as we speak. 

In addition, we have added support for the latest IBM environments, including the IBM software stack, Power 9 and Nvidia CUDA 9. This new release is yet another example that demonstrates Arm's commitment to provide cross-platform support for tools.

MAP Summit

But there's more! In just a few weeks, Arm will be releasing Forge and Performance Reports 19.0. This new major version will include full native Python profiling and performance analysis support. With such a capability, we endeavour to help novice developers and HPC gurus make the most out of the capabilities that Python has to offer, without suffering the associated performance penalty costs. In Arm MAP 19.0, it becomes surprisingly easy to speed up computation kernels in all sorts of Python workloads. We were pleasantly surprised with the performance gains we have been able to muster, sometimes in as little as a few minutes, on traditional scientific simulations, training phases of machine learning models or larger frameworks.

If you are interested in learning more, please do reach out! We are always delighted to engage with the community and find out how we can help.

Anonymous
HPC