A new European project – Low Energy Toolset for Heterogeneous Computing (LEGaTO) – seeks to develop a software stack that improves energy management in support of heterogeneous computing.
The idea is a familiar one. The limitation of transistor size reduction brought about by the decline of Moore’s Law has prompted growing use of heterogeneous computing architectures to squeeze more performance out of systems. One stumbling block has been relatively slow progress in advancing power management technology to manage systems that use multiple types of processing.
LEGaTO is a three-year project funded by the European Commission with a budget of more than €5 million. The effort will be based at the Barcelona Supercomputing Center. According to the announcement, “The project will strive to achieve this objective by employing a task-based programming model which is energy efficient by design coupled to a dataflow runtime, while simultaneously ensuring security, resilience and programmability.”
Specifically, the LEGATO project aims to:
- Improve the energy efficiency of heterogeneous hardware by an order of magnitude through the use of the energy-optimized programming model and runtime;
- Reduce the size of the trusted computing base by at least an order of magnitude;
- Reduce meantime failure rate fivefold while decreasing the energy consumption;
- Improve FPGA programmer productivity fivefold by leveraging novel features of dataflow hardware design.
Osman Unsal and Adrian Cristal, coordinators of the LEGaTO project are quoted in the announcement: “Moore´s Law is slowing down, and as a consequence hardware is becoming more heterogeneous. In the LEGaTO project, we will leverage task-based programming models to provide a software ecosystem for Made-in-Europe heterogeneous hardware composed of CPUs, GPUs, FPGAs and dataflow engines. Our aim is one order of magnitude energy savings from the edge to the converged cloud/high-performance computing (HPC).”
Link to release: https://legato-project.eu/sites/default/files/press_release.pdf