Cray recently announced the birth of our new computing research organization for Europe, the Middle East and Africa (EMEA), the Cray EMEA Research Lab (CERL). Our investment in Europe is not new (the Cray®-1 and every machine since found a European home), but an explicit focus on research is a big and bold move for our company. I am very honored to be leading that change, and I will explain here what you can expect to see from CERL.
Recent collaborations in EMEA
For our current customers, this move is very welcome, but may not have been a huge surprise. That’s because during the last five years we have become more involved in deep research collaborations in EMEA. Most notably, Centers of Excellence at EPCC, HLRS and recently also at KAUST have worked directly with users to improve application scalability and programmability. We are also an active partner in R&D efforts unattached to sales, such as an application autotuning development under CRESTA and an exascale programming models development called EPiGRAM. In addition, we were recently thrilled to be selected for a codesign project hosted by the Jülich Supercomputing Centre in Germany: The Human Brain Project’s pre-commercial procurement, in which a specialist visualization platform will be codeveloped.
Cray’s new lab will be 100 percent focused on such collaborative R&D and codesign efforts, providing a single point of contact and a pooling of expertise.
Widening the circle
My own research has focused on the new ways of describing data layout and data movement, and on the development of mathematical solvers to support that. Everyone involved in scientific computing knows how important but ambitious these topics are. In order to think outside the box, we’ve partnered with researchers at ETH Zürich who are not connected to HPC as well as with more traditional HPC application scientists. This has set a precedent for CERL projects, which will be focused on helping the community solve enormous technical challenges, and not directly oriented toward sales. We will take this approach with whoever possesses the necessary expertise, and we will develop proof of concept technologies only in combination with real HPC and analytics applications.
Toward open-source software
There is a growing need for open-source and collaborative software developments. At CERL we will develop open, collaborative software that is insulated from the traditional Cray software stack. This will give customers and partners a way to codevelop prototype software with CERL researchers and gives us the opportunity to move into new open software initiatives. By maintaining a distance between our proprietary products, CERL project details will be shared publicly and, when making significant contributions, even published in peer-reviewed journals. You can expect to hear more about all of the projects mentioned in this post in the Cray blog and other public forums.
Anticipating customers’ future challenges
Perhaps the strongest motivation for CERL lay in the difficulty in designing future hardware and software solutions. Current hardware trends indicate severe disruption in both future hardware and software, though the exact form of those changes is unknown. For example, the standard HPC programming model is widely considered to be insufficient for exascale system programming, though there is no consensus on its replacement. CERL researchers are investigating new technologies for high-level programming of supercomputers such as DSLs, Python and Chapel. We are also investigating new abstractions for data layout optimization and auto-parallelism, efficient usage of new memory technology (such as high-bandwidth and non-volatile memories), software infrastructures for interactive computing and programming models for High-Performance Data Analytics. These projects are worlds away from being mature products. The intent is that these collaborative projects help guide and shape Cray’s actual R&D programs so that future products tackle the deep technical challenges our customers are anticipating.
The European funding scenario has never been more vital, and with CERL we believe that Cray is well positioned to enter joint funding proposals in high-profile programs such as Horizon2020. That will certainly open new collaborations and partnerships for our company. I am excited about the challenges and opportunities that we will face under this new direction, and I look forward to collaborating with you.
The post Deepening Cray’s Involvement in Collaborative R&D and Codesign appeared first on Cray Blog.