Participating Applications

ATEME: Video compression

ATEME AVC/H.264 research encoding library will be tested. This software library includes the latest developed algorithms that provide the best video quality for a given bit-rate but is currently much too slow to be used in commercialized products.

BMAT: Audio identification

The tools developed in the project will be used to implement and test data-intensive computing algorithms applied to audio processing. The implementation will be focused to a new product which aims to identify, in real-time, the music content for musical events. BMAT expects significant performance improvements with use of GPUs and hybrid methods ensuring an easy to program and maintain these systems. The results of the project will be implemented within Vericast product line or derivative products that can be designed along the project.

Bull: HPL (High-Performance Linpack Benchmark)

Bull will contribute to the experimentation and validation of the H4H development environment, in particular for performance analysis and comparison and will also study the corresponding power consumption, using:

  • Benchmarks such as HPL (A Portable Implementation of the High-Performance Linpack Benchmark for Distributed-Memory Computers),
  • Open source simulation applications such as BigDFT, a DFT (Density Functional Theory)

Dassault Aviation: Aircraft design

From its CFD (Computational Fluid Dynamics) software, AeTHER, DA will consider several key numerical kernels - namely Residual computation, Matrix assembly, Linear solver - which will be considered as applications to be studied with H4H technology, in particular, for the Linear solver kernel.

DataLab: health care environment applications

DataLab will use some of the tools developed in the project to analyse the possibilities of improvement of the performance and fault tolerance capabilities of our application support services. The results of the project will be implemented in the clients’ applications, providing significant improvements in performance and availability of their applications.

Efield: Electromagnetics

Efield will study one or more of the parallel hybrid Efield solvers with the goal to use the tools developed by the H4H project to achieve improved parallel efficiency when using these solvers for various applications, in particular large scale simulation of complex radar cross section and sensor integration problems on various platforms.

GNS: Metal Forming

GNS will examine 2 types of applications:

  • GENERATOR2 is a crash model preprocessing tool that allows the user to set various positions of dummies defined in regulations simply and quickly. The H4H technology and specially the technique concerning offloading to GPUs, will be of great interest to achieve realistic results for dummy positioning (including interactions with the deformable seats) within an interactive (real-time) response.
  • The INDEED software system offers a unique range of innovative high-precision calculation models for the simulation of forming processes. Upgrading the parallelization of INDEED by combining MPI and OpenMP parallelization on CPUs and GPUs will improve scalability of the software and will allow the computation of much more complex models (including the elastic deformation of the forming tools) in a praxis relevant time.

GWT: Simulation Codes

GWT will evaluate open-source simulation codes that are used and further developed at TU Dresden by engineering scientists. For example Semtex is a 'classical' quadrilateral spectral element DNS code that uses the standard nodal GLL basis functions and (optionally) Fourier expansions in a homogeneous direction to provide three-dimensional solutions. GWT will do case studies and will try to improve these parts by means of parallelization, usage of the H4H-optimized libraries and possible offload of codelets to the accelerators.

INTES: Mechanical Design

INTES will study task and data parallelization strategies to be conducted for different application domains, such as radiation or eigenvalue problems. Hierarchical methods will be evaluated to map the different parallelization levels onto the algorithms. Scheduling strategies for CPU/GPU workload distribution will be of special interest. For large models, node-local post processing of the partitioned model could be of interest to reduce data transfer.

MAGMA: Casting process simulation

MAGMA will evaluate different parallelization strategies during different phases of the metal casting simulation process. Also more general scheduling strategies are of interest, to distribute the workload between nodes, Cores (CPUs) and GPUs. Furthermore we will evaluate the SAMG solvers for various computational domains and therefore will benefit directly from all achievements gained in adopting and optimizing SAMG for the heterogeneous platform.

RECOM: Combustion Optimisation

The 3D-combustion simulation software RECOM-AIOLOS, applied by RECOM and jointly developed with a Research Group at Stuttgart University, is a reliable, fast, and cost-effective tool for design optimization and problem solving in the power and process industry. The software is designed to predict combustion performance and pollutant formation in industrial-scale combustion systems.

REPSOL: Seismic imaging and reservoir simulation

The software developed for the Kaleidoscope project will be tested: RTM and Wave Form Inversion, which provide a very high quality subsurface image. In the short term reservoir simulation software will be developed for new projects and will incorporate H4H as a proof of concept.