Back to Table of contents

Primeur weekly 2017-09-25

Special

2017 - Another year on the Road to Exascale: The rise of Machine Intelligence ...

Focus

ARM/Allinea alliance is building a community with promising HPC tools ...

Mont-Blanc prototypes form excellent testbeds for ARM-based but also other platforms ...

Exascale supercomputing

Department of Energy's Exascale Computing Project (ECP) names Doug Kothe as Director ...

Secretary of Energy Rick Perry announces High Performance Computing for Materials Programme to help industry develop new, improved materials for severe environments ...

Quantum computing

Cost-effective quantum moves a step closer ...

Researchers demonstrate quantum teleportation of patterns of light ...

Focus on Europe

DeepL anchors neural machine translator at Verne Global's HPC-optimised data centre ...

Intel awards Paderborn University a cluster fueled by Xeon processors and Intel Arria 10 FPGAs ...

A new way to enhance the capacity of memory devices ...

Middleware

Bright Computing announces support for Ubuntu ...

Altair extends its strategic relationship with HPE ...

Hardware

Mellanox appoints Shlomit Weiss to Senior Vice President of Silicon Engineering ...

Japan Advanced Institute of Science and Technology puts Cray XC40 supercomputer into production ...

MemComputing compute engine rivals supercomputer processing speeds ...

Japan's Yokohama City University selects Cray XC50-AC supercomputer for life sciences research ...

Preferred Networks launches one of Japan's most powerful private sector supercomputers ...

Cray awarded $48 million contract from the Korea Institute of Science and Technology Information ...

Supermicro launches new 6U SuperBlade disaggregated server systems supporting rack scale design and free air cooling ...

Applications

Berkeley Lab cosmology software scales up to 658,784 Knights Landing cores ...

Multiscale simulations help predict unruly plasma behaviour ...

The Cloud

Oracle's new SPARC systems deliver 2-7x better performance, security capabilities, and efficiency than Intel-based systems ...

ARM/Allinea alliance is building a community with promising HPC tools


25 Sep 2017 Frankfurt - During the Workshops Day at ISC'17 in Frankfurt, Germany on June 22, 2017, Oliver Perks from ARM, was the second speaker of the "Going ARM" Workshop. He talked about Allinea high performance tools to debug, profile and analyze applications. Allinea is now a part of ARM but has an heritage in HPC of over ten years built by a global team with industry experience. Allinea tools are represented in 7 of the Top 10 supercomputers serving multiple disciplines. Allinea also provides training, webinars and services and is involved in the European Commission's Horizon 2020 research.

Oliver Perks gave an overview of the Allinea tools. Allinea Forge Pro is a debugger and profiler for application developers. It has the same user interface and a deep application insight. Allinea Performance Reports provide an application performance summary for system administrators. It can also deliver historical performance tracking.

In December 2016 Allinea was acquired by ARM. For Allinea this means that it can count on the continuation of a cross-platform product as well as the strength to deliver the roadmap in a faster way. The benefits for ARM are a better support for ARM-based hardware and a step towards a coherent HPC tools ecosystem. The collaboration will also provide a wealth of HPC knowledge and experience.

The cross-platform tools consist of NVIDIA GPUs, the Intel Xeon family, AMD's Ryzen, ARM's Cortex and IBM Power. There are consistent tools across multiple architectures, always with the same user interface and experience, Oliver Perks explained. It is also possible to exploit new features by making the most out of the hardware with new CUDA profiling and hardware specific performance metrics.

The cross platform is important, Oliver Perks insisted, because the consistency of tools guarantees a lower barrier to entry to ARM, enables to perform architecture comparisons and allows to get faster performance.

There is also consistency in the porting codes. The debugging code is integral to porting and developers can compare variables and arrays. As for the performance of the porting codes, it is possible to understand the performance on the old platform and then measure and optimise the performance on ARM. The consistent view aids the porting. Oliver Perks highlighted the team-building aspect by mentioning the training and professional services, the possibility to work with other tools, and the support of the community.

Allinea is now part of the ARM HPC ecosystem, meaning that the ARM HPC essentials including the Linux user space compiler for HPC applications and ARM's performance libraries BLAS, LAPACK and FFT have been supplemented with the Allinea Forge multi-node profiler and debugger and Allinea Performance Reports for cross-platform application performance insight.

The goal is to generate useful and meaningful information, according to Oliver Perks. This can be done by using scalable and portable solutions and by taking care of the data collections and data presentation.

ARM can offer support through its DDT debugger support which has the same functionality as X86 with no watchpoints. ARM also provides MAP and performance reports using the same system metrics, with no CPU instruction time and with use of Linux Perf.

ARM can also provide an advanced profiler usage with custom metrics in which the platforms can differ. One can capture platform specific data, using the MAP framework. It is also possible to work with application specific data. Further advances are to be found in the data export with JSON export, data analytics by means of Python, and a continuous integration.

Oliver Perks assured that the custom metrics can help others. ARM wants to open source its metrics, share them with different users, and build a centralised repository and knowledge base. There are open source Python scripts for analysis, as well as a strong and weak scaling analysis over multiple files. More information can be found in the Github . The developers want to share tools experience to aid the porting.

Oliver Perks ended by casting an eye into the future announcing that there was so much more to come including selected rank profiling, metric disabling, Python debugging, and advanced GPU support.
Leslie Versweyveld

Back to Table of contents

Primeur weekly 2017-09-25

Special

2017 - Another year on the Road to Exascale: The rise of Machine Intelligence ...

Focus

ARM/Allinea alliance is building a community with promising HPC tools ...

Mont-Blanc prototypes form excellent testbeds for ARM-based but also other platforms ...

Exascale supercomputing

Department of Energy's Exascale Computing Project (ECP) names Doug Kothe as Director ...

Secretary of Energy Rick Perry announces High Performance Computing for Materials Programme to help industry develop new, improved materials for severe environments ...

Quantum computing

Cost-effective quantum moves a step closer ...

Researchers demonstrate quantum teleportation of patterns of light ...

Focus on Europe

DeepL anchors neural machine translator at Verne Global's HPC-optimised data centre ...

Intel awards Paderborn University a cluster fueled by Xeon processors and Intel Arria 10 FPGAs ...

A new way to enhance the capacity of memory devices ...

Middleware

Bright Computing announces support for Ubuntu ...

Altair extends its strategic relationship with HPE ...

Hardware

Mellanox appoints Shlomit Weiss to Senior Vice President of Silicon Engineering ...

Japan Advanced Institute of Science and Technology puts Cray XC40 supercomputer into production ...

MemComputing compute engine rivals supercomputer processing speeds ...

Japan's Yokohama City University selects Cray XC50-AC supercomputer for life sciences research ...

Preferred Networks launches one of Japan's most powerful private sector supercomputers ...

Cray awarded $48 million contract from the Korea Institute of Science and Technology Information ...

Supermicro launches new 6U SuperBlade disaggregated server systems supporting rack scale design and free air cooling ...

Applications

Berkeley Lab cosmology software scales up to 658,784 Knights Landing cores ...

Multiscale simulations help predict unruly plasma behaviour ...

The Cloud

Oracle's new SPARC systems deliver 2-7x better performance, security capabilities, and efficiency than Intel-based systems ...