Call for presentations and participants: Forum on the Use of GPU and Accelerators for HPC

Thursday 27 October 2011, 10:00 – 17:00, hosted by Universität Basel

Setting the Scope

In the last years the interest in using Graphics Processing Units (GPU) for HPC computing has been growing. Modern GPUs are designed to support computer games in the real time calculation of their visual effects. Therefore modern GPUs are highly parallel multi core systems optimised for floating point operations and high throughput. They pack more computing power in the same space than a traditional CPU with reduced power consumption and cost. While those characteristics are perfectly fitted for HPC computing others impose problems.  The more problematic features of GPU arise since in its classic use neither precision is a top priority nor is there much need for access control to the GPU.

With the introduction of CUDA, OpenCL and similar products a larger number of programmers were enabled to use GPUs. This is leading to an increasing number of programs making use of GPU capabilities and leads to an increased demand of GPUs in HPC computing.

The use of GPU in HPC clusters pose new challenges like the integration in the queuing system and the programming languages to use. The goal of this forum is to exchange the information about the usage of the GPU in a HPC environment.

Key Questions

  • Are GPUs a current hype or the technology of the future?
  • How can we make use of this new technology?
  • How to integrate the GPUs in the queuing systems?
  • Which system tools are you deploying to monitor and report the HW status of the GPGPUs accelerators integrated into your HPC Cluster? How do you handle GPU drivers failure detection and recovery, also from a batch queuing system point of view?
  • How to gain usage metrics per user?
  • What are your experiences in the integration of GPU in the HPC clusters?
  • What could be the possible metric in order to measure the usage and the performance of the GPU?
  • What are the limitations of this technology (benefit versus overhead)?
  • Development tool CUDA versus OpenCL: experience, advantage, disadvantage?
  • Are your GPGPU cluster customers adopting GPU-Direct technology ? If yes, in which application field ? In which particular application ? Did they measure any interesting application performance improvement ?

Location / Hosting

The Topic Forum will take place at the Biozentrum at the University of Basel in the room 106.

Please refer to this link to get information about how to reach Biozentrum and a map of the site:


  • Martin Jacquot, URZ, tel 061 267 22 66
  • Michele De Lorenzi, CSCS, tel 091 610 82 08


This event is only open to hpc-ch community members and guests.