All times are in EST (local time in Atlanta)
Tuesday, 19 November 2024
10:15 AM - 10:45 AM
TSUBAME4.0: More of Everyone's Supercomputer toward HPC-AI Era in Science Tokyo
Toshio Endo, Institute of Science Tokyo
TSUBAME4.0 is a new supercomputer whose operation was started in April 2024 by Tokyo Institute of Technology. Now Tokyo Tech has been changed to Institute of Science Tokyo, due to merger. This talk presents overview and current status of TSUBAME4.0 with 952 PFlops in AI.
11:00 AM - 11:30 AM
Compute Node Partitioning Strategy in TSUBAME4.0
Akihiro Nomura, Institute of Science Tokyo
In TSUBAME supercomputer series, we partition compute nodes dynamically to accommodate as many jobs as possible. Since TSUBAME3.0, we have used Linux control groups (cgroups) to accomplish this partitioning, and we started using NIVIDIA's MIG(Multi Instance GPU) from TSUBAME4.0. In this talk, we report the design, implementation, and limitations of node partitioning and whether it worked or not.
2:30 PM - 3:00 PM
A New Era of Infrastructure Management Through the Fusion of Artificial Intelligence and Infrastructure Physics Engineering
Mohamed Wahib, RIKEN R-CCS
This project integrates Artificial Intelligence (AI) with High Performance Imaging (HPI) to enhance road infrastructure management in Japan. By leveraging advanced AI techniques such as machine learning and computer vision, alongside cutting-edge imaging technologies, we address critical challenges in monitoring, analyzing, and maintaining large-scale infrastructures. The fusion of these technologies enables detection of structural issues, predictive maintenance, and resource optimization. We will highlight the transformative potential of AI and HPI in improving safety, efficiency, and sustainability across sectors such as construction, energy, and transportation. We emphasize on the innovations in data processing, image analysis, and decision-making frameworks, emphasizing how this interdisciplinary approach can revolutionize infrastructure management.
Wednesday, 20 November 2024
4:00 PM - 4:30 PM
Interactive HPC: Scheduler Technologies and Use Case
Hiroki Ohtsuji and Masahiro Miwa, Fujitsu Limited
This presentation introduces our scheduler technologies, enabling interactive use of supercomputers. A case study demonstrates their effectiveness by accelerating a digital twin application for optimal e-scooter placement policy exploration on the TSUBAME4.0 supercomputer.
Thursday, 21 November 2024
1:00 PM - 1:30 PM
Continual Pre-Training on TSUBAME for a Target Language.
Kazuki Fujii and Taishi Nakamura, Institute of Science Tokyo
This booth talk introduces a methodology for continual pre-training Llama-3 specializing in target languages while maintaining English proficiency. We present technical aspects of training configurations for 8B and 70B models, focusing on efficient adaptation strategies that preserve the model's original capabilities. We will also show practical considerations and technical tips for successfully training these large-scale models.