Opening

Software acceleration for dummies - a survey of current technologies

• Magnus Peterson (Synective Labs)

Speeding up computational heavy software by replacing traditional CPUs with other types of programmable devices is becoming a more and more viable technique. GPUs, FPGAs and Cell processors, which are the most common ones, offer tremendous computing power, at lower cost and at lower power consumption than CPUs. Their supporting eco-systems have also matured considerable over the last years, making them increasingly interesting not just for experimental systems, but for a broad range of real life applications. This presentation will make a survey of the different technologies, of available hardware solutions and available programming tools.

Ct - a new paradigm for data parallel computing

• Hans-Christian Hoppe (Intel)

Intel's Ct Technology provides a comprehensive set of data parallel abstractions that greatly simplify the task of writing parallel applications and at the same time deliver forward-scaling performance across a wide variety of multi- and manycore platforms. Ct enables developers to incrementally introduce data parallelism in C++ programs, while avoiding ther risks of data races and/or deadlocks. The talk introduces the Ct concepts and discuses examples on how to use Ct effectively for HPC applications.

AMD & OpenCL: A Balanced Platforms Approach to Heterogeneous Computation

• James Hrica (AMD)

GPUs have been proven to offer benefits for accelerating computationally intensive algorithms. However, CPUs are by no means being made any less relevant for HPC workloads. Many applications require leveraging both the GPU and CPU to enable the greatest acceleration, i.e. heterogeneous computing. This presentation will provide an overview of AMD's vision for heterogeneous computing.

OpenCL and GPU High Level Programming Tools

• Stéphane Bihan (CAPS)

OpenCL is an initiative launched by Apple to ensure application portability accross various types of GPUs. It aims at being an open standard (royalty free and vendor neutral) developed by the Khronos OpenCL working group (http://www.khronos.org). This talk will give an overview of the language to program GPUs, portability and its place in and along with other programming tools.

Molecular dynamics simulations on GPUs

• Rossen Apostolov (Stockholm University)
• Szilard Pall (Stockholm University)

The Open Molecular Mechanics (OpenMM) library provides tools and consistent hardware-agnostic API for modern molecular modeling simulations with emphasis on hardware acceleration (currently GPUs only). The talk will present an overview of the platform and discuss some of the implemented algorithms, how they differ from the standard CPU ones and how performance is affected by them.

Nvidia/CUDA - applicability and problems by the example

• Hans Hacker (LRZ)

As part of the PRACE, a few kernels from the Euroben-benchmark had to be ported to all the available PRACE prototype architectures. This presentation focuses on the Nvidia/CUDA port and gives an overview of the used Nvidia Hardware, the experience with CUDA and the available toolkit. It illustrates the porting effort, various problems and the results by the example with three of those kernels, namely a dense matrix-matrix multiplication, a sparse matrix-vector multiplication and a 1D fast Fourier transformation.