Konrad-Zuse-Zentrum für Informationstechnik Berlin (ZIB)

ZIB logo

Principal Investigators:

Dr. Thomas Steinke

Dr. Thomas Steinke is head of the HPC dept. at the Zuse Institute Berlin (ZIB). His research interest is in high-performance computing, heterogeneous systems for scientific and data analytics applications, and parallel simulation methods. Thomas co-founded the OpenFPGA initiative in 2004, and he leads the Intel® Parallel Computing Center(s) (Intel® PCC) at ZIB. He received the doctoral degree in Theoretical Chemistry from the Humboldt-Universität zu Berlin in 1990.

 

Florian Wende

Florian Wende is part of the Distributed Algorithms and Supercomputing department at Zuse Institute Berlin (ZIB). He is interested in accelerator and many-core computing with application in Computer Science and Computational Physics. His focus is on load balancing of irregular parallel computations and on close-to-hardware code optimization. He received a Diploma degree in Physics from Humboldt Universität zu Berlin and a Bachelor degree in Computer Science from Freie Universität Berlin.

 

Matthias Noack

Matthias Noack is part of the Distributed Algorithms and Supercomputing group at Zuse Institute Berlin (ZIB). His interests include parallel programming models, heterogeneous architectures, and scientific computing. He developed the Heterogeneous Active Messages (HAM) framework, which provides efficient offloading, local and over fabric, for multi- and many-cores. Matthias currently focuses on runtime compilation techniques, portable programming methods for vectorization, as well as optimization and scaling of the Hierarchical Equations of Motion (HEOM) method.

Description:

The proposed activities include the continuation of our code modernization work with the VASP, DM-HEOM, and PALM codes. We want to address the challenges involved with near-future Intel® Xeon® CPU designs with many cores and Intel® Advanced Vector Extensions 512 (Intel® AVX-512) support (e.g. Cascade Lake). A new focus is on a nonlinear optimization code which is expected to benefit from wide SIMD and many-core parallelization and in particular through variable precision computations on FPGA platforms.

The memory-intensive codes VASP and PALM are our candidates for exploring different configurations of the memory hierarchy (DRAM and NVRAM). The communication dominating parts of VASP and DM-HEOM will be optimized by using MPI 3 features (e.g. one-sided communication and neighborhood collectives) to increase communication efficiency and scalability.

Our IPCC work is linked to the IXPUG community and part of our collaboration with NERSC.

Finally, this proposal helps us to strengthen the knowledge and expertise of code developers of the Northern German HPC Alliance (HLRN) through our dissemination work and workshop activities.

Publications:

Related Website:

For more complete information about compiler optimizations, see our Optimization Notice.