Strategies for energy efficiency in HPC
Energy efficiency needs to be addressed in HPC at all layers of the stack, from application development up to hardware deployment and system operation. While significant gains have been achieved through advances in hardware architecture, improving how existing systems are operated remains an equally important opportunity. This presentation offers a broad overview of established approaches to improving energy efficiency, including hardware innovations, system software techniques, and application settings. It will then focus on operational practices aiming at obtaining the maximum throughput out of existing infrastructures, at the minimum possible energy cost. In particular, it discusses how runtime decisions based on system state and workload characteristics can be leveraged to reduce energy waste while maintaining performance. The presentation draws on work carried out in the SEANERGYS project, which develops a software suite combining monitoring, data analytics, and scheduling to achieve better energy-aware system operations.

Towards Real-time AI at the Edge

Edge AI applications including Agentic AI systems require real-time performance. In addition, in many applications, energy efficiency becomes a critical metric. To meet these requirements, recently many edge platforms have been proposed including GPUs, FPGAs, NPUs and heterogeneous architectures such as AI PCs. These devices are being used along with multi-core and novel memory technologies to realize advanced platforms to accelerate edge inference and support diverse applications requiring low latency. We will review emerging technologies for edge inference and advances in reconfigurable computing over the past three decades leading up to current innovations in FPGA accelerators for AI. We will illustrate parallel architectures and algorithms for commercial, defense and space applications. Using our algorithm-architecture co-design methodology to realize high performance accelerators for these applications, we demonstrate the role of modeling and algorithmic optimizations to develop highly efficient Intellectual Property (IP) cores for FPGAs and realize end to end application acceleration. We illustrate our methodology by developing high performance designs for graph machine learning, long context LLM inference, Mixture of Agents inference and SAR ATR. We conclude by identifying opportunities and challenges in exploiting emerging heterogeneous architectures composed of multi-core processors, FPGAs, integrated GPUs, NPUs and accelerators.

Distributed Machine Learning: Bridging Cloud and Edge Systems
Edge-cloud solutions are being used to collect and analyze large amounts of data generated by IoT devices in various application domains, such as urban mobility, smart cities, healthcare, and augmented reality. We must be able to combine techniques and algorithms of data analysis and machine learning with the scalable architectures of Cloud systems and Edge technologies. This approach can reduce latency and network congestion associated with traditional cloud-based machine learning techniques by processing data locally on edge devices before sending it to the cloud for further analysis. This keynote discusses distributed machine learning and proposes a reference architecture to adapt distributed machine learning algorithms at the edge-cloud continuum. Real applications are presented, and the main open research issues are discussed.

Tasks Based Hybrid Workflows for Emergent Technologies

The Barcelona Supercomputing Center hosts a twin installation based on transmon qubits, superconducting technology funded by Spain. Another annealing system co-funded by EuroHPC JU and Spain is also expected. Significant research and development is done by BSC researchers from the Quantic groups in topics related to quantum computing, which are closer to its physics nature. Complementary to this research, and in collaboration with our colleagues, the Workflows and Distributed Computing group is doing research activities related to software aspects related to the integration of HPC + Quantum Computing (QC).

The talk will describe the steps taken towards integration of HPC and QC at the BSC. We will describe how a traditional workflow environment is being extended to support hybrid HPC+quantum computing workflows.The current solution enables the hybrid execution of workflows in CPUS, GPUs and Quantum systems (onsite and through cloud access). The talk will present these topics and our plans for the near future.