Path Activities

On 16 and 17 May 2018, Prof. Spencer Sherwin (Imperial College London) visited cfaed and gave a talk on recent developments of the Spectral Element Method achieved in his group.He discussed about common research questions related to the Spectral Element Method which is used in the ORCH path as an application challenge for running codes on wildely heterogeneous hardware. At this occasion Spencer Sherwin also participated in the defense of Karsten Bock dealing with high-order surface representation.

The PhD students, postdocs and investigators of the Orchestration Path met from 3rd to 5th July for the annual retreat in Altenberg on the ridge of the Eastern Ore Mountains (Osterzgebirge). Most of the time was dedicated to intense working sessions about collaborative projects and future research ideas. Additionally, the invited talk by Diana Göhringer, the new professor for adaptive dynamic systems at the faculty of computer science, and the participation of Steffen Lange and Jens Karschau two postdocs from the Biological Systems Path have been used to discuss and initiate future collaborations. We also enjoyed the nice weather on a short hiking tour to Kahleberg.

In the first week of July, 2016 the Orchestration Path held its annual retreat in the monastery St. Marienthal (Ostritz). The PhD students, postdocs and investigators met and discussed recent developments and results achieved, worked on collaborative projects and exchanged ideas for future cooperations. Overall, it was a very intensive week with inspiring talks by the investitors, an invited Talk by Reinhard Berger of the Molecular Functional Materials Group on Synthetic Carbon Structures for Integration on the System Level, many fruitful discussions within and outside the working sessions completed by a boat&bike tour as team building event.

In April, 2016 the Orchestration Path invited Tal Ben-Nun from Hebrew University of Jerusalem as a visior. In his talk he reported on his recent and ongoing work on memory access patterns [1].

Abstract: GPUs play an increasingly important role in high-performance computing, accelerating applications ranging from machine learning and computer vision to quantitative finance. In spite of their popularity, GPU programming remains a challenging task. Manual management of the multi-level memory hierarchy, inter-device synchronization, and various development constraints often generate lengthy, error-prone, and architecture-specific code. Therefore, it is imperative to develop new programming paradigms for efficient utilization of GPU and multi- GPU systems, without sacrificing code simplicity and portability. The talk presents Memory-Oriented Programming (MOP) - a data-centric programming model organized around memory access patterns. MOP categorizes parallel algorithms by their input and output parameters, where each parameter is characterized by a single pattern. Using these patterns, which cover most existing GPU applications, we show that it is possible to write short and intelligible code that attains high performance on a variety of GPU architectures and multi-GPU nodes. Furthermore, we demonstrate that the model can be used to increase the resilience of multi- GPU nodes by providing transparent failure protection mechanisms. We show that the resulting code is on par, and in some cases outperforms, existing manually optimized production-grade libraries, exhibiting near-linear speedup on a single node with multiple GPUs. Performance is measured on fundamental computational operations, as well as real-world applications in deep learning and non-negative matrix factorization.

[1] Tal Ben-Nun, Ely Levy, Amnon Barak, and Eri Rubin. 2015. Memory access patterns: the missing piece of the multi-GPU puzzle. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC '15). ACM, New York, NY, USA, , Article 19 , 12 pages. DOI=http://dx.doi.org/10.1145/2807591.2807611