Democratizing the use of AI with FUJITSU-MONAKA - fltech - 富士通研究所の技術ブログ

Introduction

The growth of artificial intelligence (AI) is mainly driven by three factors – Algorithmic Innovation, Data & Compute. AI development is at an inflection point, as researchers developed larger models and larger data sets, the size of the models and volume of the data sets have grown beyond the limits of a single compute facility to run efficiently. Currently AI performance is prohibitive due to the cost of both time and resources, thus there have been multiple attempts to design cheaper alternatives for efficient AI at Scale. These challenges are giving life to “Heterogeneous Computing”. The increasing demand for data intensive workloads has led to a rise in the use of accelerated software.

Fujitsu’s Computing Excellence

Fujitsu is developing FUJITSU-MONAKA, a 2nm, Arm-based CPU tailored for data centers. It leverages Fujitsu's 60-year legacy in processor development and incorporates open-source software technologies. FUJITSU-MONAKA aims to deliver both power efficiency and high performance, contributing to carbon-neutral green data centers with double the energy efficiency of competitors and double the performance for AI workloads.

FUJITSU-MONAKA Goals

FUJITSU-MONAKA is set to revolutionize computing with its groundbreaking 3D many-core architecture. The system will seamlessly integrate confidential computing, prioritizing security for sensitive data. Anticipated to be highly reliable, energy-efficient, and user-friendly, FUJITSU-MONAKA is expected to redefine the computing experience. Projections for 2027 suggest an impressive 2x increase in application performance & 2x power efficiency compared to competitors, establishing FUJITSU-MONAKA as a pinnacle of innovation.

FUJITSU-MONAKA Microarchitecture

FUJITSU-MONAKA adopts the innovative 3D many-core architecture.

All last-level cache functionalities are strategically housed in a distinct 5nm SRAM die located at the bottom, ensuring a streamlined and efficient architecture.
The architecture is tightly coupled through the utilization of Through-Silicon Vias (TSV), enhancing the communication coordination between the core and SRAM dies.
The design adopts a 3D many-core approach, optimizing core count, reducing latency, and maximizing overall throughput for enhanced processing capabilities.
Remarkably, the 2nm area, responsible for high-performance processing, accounts for less than 30% of the total die size contributing to cost efficiency.

Fujitsu’s contributions to open-source software

Fujitsu has a long history of collaborating with open-source communities by contributing key developments in mission-critical systems. FUJITSU-MONAKA Software R&D Unit, Fujitsu Research of India Pvt. Ltd. (FRIPL), is continuing the processor legacy and leading software development for the new processor. Open-source technology holds an important key to overcoming the many challenges the world is facing and creating sustainable value for the environment, economy, and people’s well-being. Fujitsu’s vision draws on our extensive global experience working with and collaborating with open-source communities.

Fig 4: Fujitsu’s key contributions to open-source community

Fujitsu has also made major contributions to oneAPI. Fujitsu and Riken optimized and ported oneDNN deep learning library for the Armv8-A instruction set, enabling it to run at high speed on Arm-based supercomputer. Fujitsu continues oneDNN optimizations for FUJITSU-MONAKA with JIT kernel development by efficiently leveraging Arm SVE to achieve further improved performance in training & inference. Fujitsu has recently also successfully contributed Arm SVE enablement and tuning of oneAPI Data Analytics Library (oneDAL) to speed up AI-ML algorithms. Our teams are also contributing actively towards software innovation in Math Kernels, OpenBLAS, Arm Compute Performance Libraries, PyTorch, TensorFlow and to build optimized AI solutions.

FUJITSU’s Unified Acceleration Initiatives

Fujitsu became an inaugural member of The Linux Foundation's Unified Acceleration Foundation (UXL) on October 5, 2023. The UXL's goal is to allow a single source code to generate native binaries optimized for a variety of accelerators, including CPUs, GPUs, and FPGAs. Through our active participation in the UXL's open-source community, we are enabling customers to operate various accelerators, including FUJITSU-MONAKA, using standard protocols. This initiative is aimed at expanding hardware options for our customers. UXL Foundation tries to simplify accelerator programming, uniting industry experts to drive innovation and implement cross-platform solutions.

Fig 5: Fujitsu’s Goals & focus areas with UXL Foundation

Conclusion

FUJITSU-MONAKA's commitment to open ecosystems and power-efficient CPUs will play a crucial step in democratizing AI. Fujitsu's forward-thinking approach, with the development of the FUJITSU-MONAKA CPU, aligns with the vision of a carbon-neutral society. The expansive range of AI applications, from healthcare to defense, underscores the transformative impact of accessible AI. Fujitsu is committed to building the future of human centric and sustainable digital transformation which needs a holistic approach to technology development and open collaboration with the international community.

Acknowledgements

NEDO Project | “Technology Development of the Next Generation Green Data Center” for the “Green Innovation Fund Project/Construction of Next Generation Digital Infrastructure”.
NEDO is “New Energy and Industrial Technology Development Organization”, a national research and development agency in Japan.
Fujitsu has been selected for the national initiative along with NEC Corporation, AIOCORE Co., Ltd., KIOXIA Corporation, FUJITSU Optical Components Limited and KYOCERA Corporation.
This presentation is based on results obtained from a project, JPNP21029 subsidized by the New Energy and Industrial Technology Development Organization (NEDO).

Comment from developers

FUJITSU-MONAKA Software R&D unit team members

Our project focuses on AI framework engineering and performance optimization for FUJITU-MONAKA with cutting-edge AI technologies such as machine learning, deep learning, and natural language processing (NLP), including computer vision, generative AI, and large language models (LLMs). Our work streams also cover real-time data processing and big data analytics, and all our teams actively collaborate with the open-source software (OSS) community to drive innovation in advanced computing with HPC AI core technologies, delivering accelerated software for data-intensive workloads.