EPiGRAM-HS: Programming Models for Exascale Systems

EU-Project – Exascale Programming Models for Heterogeneous Systems

As supercomputers move towards the exascale era and become more powerful every year, it is also becoming increasingly difficult to program them. EPiGRAM-HS is a project funded by the European Commission (in the framework of Horizon 2020) with the goal of developing a programming environment for heterogeneous exascale systems. Our institute is on board.


Heterogeneity With Advantages and Challenges

Exascale supercomputers are capable of exaFLOP and are the next milestone in the computing power of mainframe computer systems. Exascale – that means 10 to the power of 18 floating point operations per second. Extremely powerful machines. To increase the computing power or to reduce the energy consumption, heterogeneous systems are used, i.e. there is more than one type of processor or cores like GPUs and FPGAs.

This complexity of heterogeneous systems has many advantages. However, heterogeneous systems are more difficult to program because many codes cannot simply be transferred from one type of machine to another. The potential cannot be fully exploited because the developers do not yet know how to handle the machines.

In the EPiGRAM-HS project, six partner organizations are working together to provide a programming framework for the application experts. Each partner in the project is responsible for providing a small part of the framework. EPiGRAM-HS started in September 2018 and is funded for three years by the European Commission.



Application Reference – From Weather to E-Health

More productive and powerful software has a strong social impact on many levels. For example, weather forecasts can be made faster and more accurate. Space weather and numerical fluid dynamics can provide important data and help to develop more effective technologies for the future - also with regard to the climatic changes caused by climate change. Exascale systems will also support medical applications in the future. Optimized detection of cancer using deep learning applications is just one example under the keyword e-health.

Overview graphic of the EPiGRAM-HS project.
Overview graphic of the EPiGRAM-HS project.

Our Part: Tasks and Objectives of the Fraunhofer ITWM

We extend our HPC tools (GPI and GPI-Space) to integrate deep learning applications on heterogeneous systems with FPGAs (field-programmable gate arrays).

In detail, we are working on the following points:

  • We extend our task-based programming system GPI-Space for the use in heterogeneous computing clusters. In the future we will be able to integrate FPGAs into the workflow. The development will be tested with deep learning applications.
  • We extend the parallel programming model GPI in a way that communication between FPGAs is possible and tested. The communication is controlled by the host CPU. Furthermore we will parallelize at least one common deep learning framework (TensorFlow or PyTorch) using GPI. The extensions will also be tested with deep learning applications.

Video: Overview of the EPiGRAM-HS project

Privacy warning

With the click on the play button an external video from www.youtube.com is loaded and started. Your data is possible transferred and stored to third party. Do not start the video if you disagree. Find more about the youtube privacy statement under the following link: https://policies.google.com/privacy

Our expert Dr. Valeria Bartsch explains in the video the challenges and tasks that arise in the project.