Japan's FugakuNEXT: A Leap Towards Zetta-Scale AI-HPC Supercomputing

Japan's Ambitious Leap into Zetta-Scale Computing

Japan is set to reclaim its position at the forefront of high-performance computing with the announcement of FugakuNEXT, a cutting-edge hybrid AI-HPC supercomputer. Unveiled at a ceremony in Tokyo on August 22, this ambitious project marks a significant collaboration between RIKEN, Fujitsu, and Nvidia, aiming to surpass exascale and venture into zetta-scale territory 1

1

2

2

3

3

.

A Powerful Fusion of Technologies

FugakuNEXT will integrate Fujitsu's custom-designed MONAKA-X ARM-based CPUs with Nvidia's state-of-the-art GPUs. This marks a departure from its predecessors, as it's Japan's first flagship system to adopt GPUs, reflecting the growing global trend of unifying HPC and AI computing 1

1

. The system will utilize NVLink Fusion, a high-bandwidth bridge between CPU and GPU, to create a seamless integration of these technologies 1

1

3

3

.

Unprecedented Performance Goals

RIKEN has set ambitious performance targets for FugakuNEXT:

1. A fivefold increase in hardware performance over the original Fugaku
2. A twentyfold improvement in software and algorithmic performance
3. A projected peak performance of around 600 exaFLOPS in FP8 sparse precision

These advancements are expected to deliver a massive hundredfold increase in real-world application performance compared to today's machines, potentially making FugakuNEXT the world's first zetta-scale supercomputer 1

1

2

2

.

Energy Efficiency and AI Integration

Despite its significant performance leap, FugakuNEXT is designed to operate within the same forty-megawatt power envelope as its predecessor, Fugaku, underscoring a strong commitment to energy efficiency 1

1

. The system will also fully integrate Nvidia's software stack, including CUDA-X libraries, TensorRT for inference, and NeMo for large language model training, reflecting the growing role of AI as a scientific instrument 1

1

3

3

.

Applications and Impact

FugakuNEXT is positioned as an integrated AI-HPC platform, aiming to accelerate discovery across diverse fields:

1. Climate modeling and disaster resilience
2. Drug development and healthcare research
3. Advanced manufacturing and industrial design
4. Earth systems modeling for earthquake and severe weather prediction 1

   1

   3

   3

The system is designed to automate various aspects of the research process, from hypothesis generation to experiment simulation and validation, potentially revolutionizing the scientific method 1

1

.

Strategic Importance for Japan

For Japan, FugakuNEXT represents more than just a technological achievement. It's a strategic investment aimed at:

1. Strengthening domestic semiconductor technologies
2. Positioning Japan as a global leader in AI and HPC innovation
3. Serving as a national platform for research and industry 1

   1

   2

   2

The project, backed by Japan's Ministry of Education, Culture, Sports, Science and Technology (MEXT), is projected to cost over 110 billion Yen ($740 million USD) 1

1

3

3

.

Timeline and Future Prospects

FugakuNEXT is expected to come online around 2030 at RIKEN's Kobe campus on Port Island 1

1

. While the exact specifications of the Nvidia GPUs to be used are yet to be announced, the early collaboration between RIKEN, Fujitsu, and Nvidia suggests a carefully planned development process 2

2

3

3

.

As FugakuNEXT takes shape, it promises not just to succeed Fugaku but to redefine the role of national supercomputers in shaping both science and society. Its success could establish Japan as a central force in the zetta-scale era and provide a blueprint for future high-performance computing systems worldwide.