Copyright © Unsplash / Christian Lue
Project profile
Italy, Germany, France, Spain
01/04/2022 - 31/03/2024
Motivation and Goals
- Energy efficiency and thermal control
- Develop an innovative two-phase cooling approach at node and rack leve
- Fully integrate this with an optimized multi-level runtime resource management system
- Sustained application performance
- Efficiently exploit highly concurrent accelerators (GPUs and FPGAs) focusing on data/stream locality, efficient algorithms and programming models, tuned libraries and innovative IP blocks
- Seamless integration of reconfigurable accelerators
- Extend field-proven tools for the design and implementation of accelerator code (like Vitis and OmpSs@FPGA) to support mixed-precision computing and power monitoring/control and innovative methodologies and IP blocks
- Create new hardware IP
- Develop IP blocks for mixed-precision arithmetic, data compression, power monitoring and control, task scheduling, and low-latency communication
- Integrated Development Platforms
- Develop two architecturally different, heterogeneous development vehicles (IDVs), one as a dedicated testbed for two-phase cooling technology, and one supporting the wider range of project innovations
Highlights/Status
- The two-phase cooling system was integrated into the energy development vehicle
- A thermal model for driving power management decisions was developed
- Prototype tools and IPs (e.g., to integrate accelerators or to generate on-line power monitors) are available in a preliminary version
- Preliminary versions of applications are ready
Copyright © Textarossa
Project coordinator
