Demi journée GPGPUDemi-journée sur le thème des accélérateurs graphiques
Thursday, October 7th, CERFACS Conference Room - 14h00
Inscription obligatoire par email à l'adresse suivante : firstname.lastname@example.org
14h00 Architecture GPU Fermi et écosystème NVIDIA pour le calcul hybride
Jean-Christophe Baratault - NVIDIA
14h45 L'Hybride computing avec AMD Fusion.
Hervé Chevanne - Software Engineer - ATI
15h30 PGI Compilers & Tools for Heterogeneous Manycore Systems
PGI has a long history of supplying optimizing/parallel compilers & tools for multi-core x86 HPC systems based on the lastest CPUs from Intel and AMD. In the past year, the PGI compilers have been enhanced to support porting and optimization of x86 applications to systems incorporating NVIDIA GPU accelerators. This talk will provide an overview of these new features and how they can be used to maximize performance and program x86+GPU platforms as integrated heterogeneous manycore systems.
Speaker : Douglas Miles - responsible for all business and technical operations of The Portland Group (PGI). He has worked in various positions over the last 25 years in HPC applications engineering, math library development, technical marketing and enginneering management at FPS, Cray Research, PGI and STMicroelectronics.
16h15 Coffee break
16h30 Incremental migration of C anfd fortran applications to GPGPU using HMPP
17h15 GPU Debugging Made Easy with Allinea DDTGPGPU is a tremendous opportunity to many application fields. Migrating legacy software to GPGPU is a complex process that requires mastering the technological risks (e.g. loss of code portability, extensive code restructuration, debugging complexity) as well as costs. In this talk we present a methodology based on HMPP (Heterogeneneous Multicore Parallel Programming) to set up an incremental approach that reduces the cost and risks when porting codes ti GPGPU.Francois Bodin - CAPS
GPU represent a new holy grail for HPC users and NVidia Cuda is a popular route for developing applications tthat exploit their power.Naturally, where there is development, there will also be bugs. This presentation will introduce the new CUDA enabled version of Allinea DDT and will show some of the powerful features that are in the product to help track down CUDA bugs "easily".Allinea DDT parallel debugger has been setting the standard for usability for many years and is tearing up scalability records. It is used on the world's largest systems - debugging over 220 000 processes in some cases - so all GPU threads should be within reach of DDTLindon Locks - Technical Pre sales Consultant - Allinea