AC2 Laboratory focuses on research and development of massively parallel computing systems and their applications to real-world problems, particularly in the area of HPC (High Performance Computing) research based on GPUs and other accelerators. Given the performance demands of supercomputers, the limitations of available power capacity, and the growing trend toward decarbonization, improving the energy efficiency of supercomputers has become an urgent challenge, with the use of accelerators emerging as the mainstream solution.

Our lab is dedicated to advancing hardware and software for heterogeneous computing, including system architecture, data structures, algorithms, task scheduling, and programming models, with the goal of maximizing accelerator performance. Based on our past achievements in large-scale parallel computing, accelerator systems, and energy-efficient supercomputing, we are broadening our focus to include next-generation AI/HPC infrastructure. We are committed to continuing to explore and shape the future of supercomputing systems.

For example, we are currently conducting research such as the following.

• Wafer-Scale GNN Training: We study GNN training on GPUs and wafer-scale systems to understand when irregular graph workloads benefit from a simpler execution model and a large fast memory system.

• DPU-Assisted LLM Training: We study communication-data compression and offload on NVIDIA BlueField DPUs for multi-GPU LLM training, using large-scale training as a practical testbed for future communication-sensitive workloads.

• Heterogeneous Runtime: We investigate runtime principles for mapping sparse, irregular, and communication-sensitive workloads across GPU clusters, GPU+DPU systems, and wafer-scale or other emerging accelerator platforms.

For ongoing projects and research themes, please visit the Research page.

For publications and technical outputs, please visit the Publications page.

For available computing resources, please visit the Computers page.

Accelerated Computing

• GPU computing

• FPGA/DPU and reconfigurable computing

• Wafer-scale systems

• Emerging accelerator platforms

Sparse and Irregular Computing

• Dynamic graph learning systems

• Sparse matrices and adaptive meshes

• Irregular memory access and data placement

• Dependency-sensitive execution

Runtime and Programming Systems

• Runtime systems for heterogeneous computing

• Dynamic scheduling and resource management

• Accelerator-aware software stacks

• Performance portability, reproducibility, and co-design

AI for Science and Digital Twins

• Graph and scientific learning systems

• Digital-twin-oriented AI/HPC workloads

• Distributed training and world models

• Accelerating real-world applications

We share recent papers, posters, technical reports, talks, and lab activities here.

• 2026-05-16: An open lab event has been held at the 2026 Graduate School Entrance Examination Information Session (Yokohama Campus).

• 2026-04-18: A poster for AC2 Lab has been displayed at the 2026 Graduate School Entrance Examination Information Session (Ookayama Campus).