The Future of AI Computing Aimed for by Fujitsu

Hello! This is the Fujitsu Public Relations Note Editorial Department.
In recent years, the rapid evolution and widespread adoption of AI technology have significantly changed our lives. While innovative services are being created one after another, the demand for computing resources to develop and use AI is increasing, and there are also concerns about the environmental impact caused by rising power consumption. To advance AI technology while being mindful of the global environment and to ensure that more people can benefit from AI, the evolution of computers is essential.
Against this backdrop, Fujitsu signed a memorandum of understanding today with Advanced Micro Devices, Inc. (AMD) to advance collaboration in the fields of AI and HPC.With this, platforms equipped with Fujitsu's next-generation CPU, 'FUJITSU-MONAKA,' will now be able to utilize AMD GPUs, expanding the options for AI computing.This time, we asked Naoki Shinjo, who leads the development of 'FUJITSU-MONAKA,' about the future of computing as envisioned by Fujitsu and the goals of the collaboration with AMD.

Interviewee: Naoki Shinjo, Head of Advanced Technology Development Division, Fujitsu Laboratories, Fujitsu Limited
　　　　　　　　Naoki Shinjo

First, please tell us about 'FUJITSU-MONAKA'.

'FUJITSU-MONAKA' *1 is a general-purpose CPU based on the Arm instruction set architecture that Fujitsu is developing for next-generation data centers and other applications, and it adopts cutting-edge 2-nanometer technology. By utilizing Fujitsu's proprietary technologies, such as our in-house designed microarchitecture and low-voltage technology, we are proceeding with development toward a 2027 release with the goal of achieving both high processing performance and power efficiency. Additionally, to ensure that its performance can be easily utilized to the fullest, we are promoting support for industry-standard software through collaboration with the OSS community.

*1 FUJITSU-MONAKA: This achievement was obtained as a result of a project subsidized by NEDO (New Energy and Industrial Technology Development Organization).

What is the goal of this collaboration with AMD?

The collaboration with AMD is to realize a shared vision of building a high-performance and power-efficient AI computing infrastructure and accelerating the democratization of AI.

'FUJITSU-MONAKA' is a CPU that boasts high power performance. By combining this with high-performance AMD GPUs, we believe we can realize an AI computing infrastructure that achieves superior performance and power efficiency for many AI services, contributing to the realization of green data centers.

What are the strengths of 'FUJITSU-MONAKA' in AI?

■ AI involves model training and inference processing

Generally, AI has a process of training to create a model and an inference process that performs data prediction or classification based on the trained model. The combination of 'FUJITSU-MONAKA' and GPUs will be utilized for training and inference processing of various AI models, including large language models. Since AI training requires vast computational resources, GPUs are often used, but GPUs are not necessarily optimal for inference processing. In particular, we have designed 'FUJITSU-MONAKA' to demonstrate high performance even on its own for inference processing.

■ 'FUJITSU-MONAKA' can demonstrate high AI performance

Inference processing does not require as much computation as training, so it can be sufficiently handled by a CPU, and in some cases, processing with a CPU is more power-efficient. 'FUJITSU-MONAKA' is equipped with AI-specific instructions *2 that can perform large amounts of computation with the precision necessary and sufficient for AI, and it is designed so that generative AI engines *3 and the like also demonstrate high performance on the CPU alone.This allows for superior power efficiency when using generative AI, such as large language models, because GPUs are not used for inference processing. Furthermore, because it is a CPU, there are benefits such as the ability to perform flexible application development and ease of deployment in edge computing environments and small-scale data centers. Additionally, 'FUJITSU-MONAKA' boasts higher performance, lower power consumption, and higher security compared to competing CPUs, providing a faster, greener, and safer inference environment.

■ Replacing physical simulations with AI

Also, use cases where AI replaces advanced simulations are also conceivable. One of these is a method called surrogate modeling, which uses AI to substitute for complex physical simulations such as fluid dynamics. For example, in the aerodynamic design of an automobile, first, a surrogate model trained on a GPU is built based on simulation results. In the phase where rough analysis results are obtained, inference using the surrogate model, which can provide results quickly even on a CPU, is utilized. In the phase where results more faithful to physical phenomena are required, high-precision simulations are conducted using the CPU. By utilizing surrogate models according to the phase in this way, we can aim to improve the efficiency of the entire business process.

*2 AI-specific instructions: Refers to low-precision floating-point arithmetic instructions such as FP16 (16-bit floating-point number) and FP8 (8-bit floating-point number). While high computational precision is required for general applications, speed in performing large amounts of computation at once is prioritized in AI processing. By reducing precision, a large amount of computation can be executed at once, enabling high-speed AI processing.

*3 Generative AI engine: Refers to software related to generative AI designed to demonstrate high performance on 'FUJITSU-MONAKA.' We are currently working on making software that mainly runs on GPUs executable on 'FUJITSU-MONAKA,' aiming to make generative AI reasonably usable with CPUs alone.

Through this collaboration, what kind of AI computing is Fujitsu aiming for?

Fujitsu is aiming to build more open platforms for AI and HPC through this collaboration with AMD. We will strengthen our cooperation with open source software (OSS) communities and organizations, and promote the development of OSS-based software for AI. By doing so, we will establish an environment where many developers can easily develop AI applications using various hardware, promote the opening of AI, and support the social implementation of AI. Through our collaboration with AMD, Fujitsu will realize more open and accessible AI computing that contributes to the realization of a sustainable society.

Recommended Articles