【 ‘GMO GPU Cloud’ measurement results】
In this year's measurement in the global supercomputer rankings, the GMO GPU Cloud recorded a LINPACK performance (*3) of 38.06 PFLOPS (petaflops) and an execution efficiency of 73.0% on a 96-node (768 GPU) configuration (*2) utilising the latest H200 GPUs The LINPACK performance (*3) was 38.06 PFLOPS (petaflops) and the execution efficiency 73.0%. In recognition of these, it was ranked 37th in the world and 6th in Japan.
As of November 2024, El Capitan (USA) was ranked No. 1 in the world on the TOP500 list, with a measured result of 1,742.00 PFLOPS.
GMO Internet Group has maximised the performance of the H200 GPU by adopting the H200 GPU, which has approximately 1.7 times the memory and 1.4 times the memory bandwidth of the H100 GPU, as well as a cluster configuration that delivers high performance even when using multiple nodes. This cluster makes the service the fastest commercial GPU cloud in Japan.
The GMO GPU Cloud is also the first Japanese cloud provider to adopt NVIDIA Spectrum-X, which is designed specifically for Ethernet networking for AI. This cloud environment is optimised for the development of generative AI, machine learning and large-scale language models (LLMs). Thanks to the NVIDIA Spectrum-X Ethernet networking platform, which enables system scalability and fast GPU-to-GPU communication, the ‘GMO GPU Cloud’ enables computing processes at maximum efficiency.
TOP500 ranking: https://www.top500.org/lists/top500/list/2024/11/
(*2) A 96-node configuration refers to a system configuration in which 96 servers (768 GPUs) are interconnected to perform parallel processing.
(*3) LINPACK performance is a measure of the processing power of supercomputers and high-performance computing systems.
【Masatoshi Kumagai, Group Representative Director and President of GMO Internet Group, commented. 】
GMO Internet Group is honoured to have achieved the world's highest level of computing performance with the introduction of the high-performance H200 GPU and NVIDIA Spectrum-X, the first cloud service provider in Japan to achieve this. NVIDIA Spectrum-X, the first cloud service provider in Japan, has achieved the fastest computing environment in Japan. This infrastructure is designed as a computing environment for next-generation AI workloads, from the development of large-scale language models to inference.
We are confident that this ranking will make a significant contribution to improving the competitiveness of Japan's AI industry, and we are proud to be able to provide a world-class computing environment in Japan. Through the GMO GPU Cloud, we will continue to contribute to the development of Japan's AI industry by providing high-performance computing environments to research institutions and companies.
【About the ‘TOP500’ ranking】（URL：https://www.top500.org/）
The TOP500 is a biannual ranking of the world's supercomputer performance published since 1993 and is known as one of the most authoritative rankings in the field of HPC, with rankings determined based on the results of the LINPACK performance benchmark, which is widely used in the field of scientific and technical computing. The ranking is based on the results of the LINPACK performance benchmark, which is widely used in scientific and technical computing.
【About the GMO GPU Cloud】（URL：https://gpucloud.gmo/）
GMO GPU Cloud is the fastest GPU cloud service in Japan. It uses the high-performance ‘H200 GPU’, which, compared to conventional GPUs, significantly reduces the learning time of large-scale language models and dramatically improves the efficiency of AI development.
In addition, we are the first domestic cloud provider to adopt NVIDIA Spectrum-X. The combination of the H200 GPU and NVIDIA Spectrum-X has created a high-level GPU cloud environment optimised for generative AI development and machine learning.
Through this service, GMO Internet Group will provide companies and research institutions working in the fields of generative AI and HPC with a high-standard computing environment that requires no infrastructure tuning, thereby contributing to shorter development times and lower costs for customers and promoting the development of the domestic AI industry.
・Launch date: Scheduled for release on 22 November 2024 (Friday).
■Features of the GMO GPU Cloud
▲NVIDIA H200 Tensor core GPU
1. ‘NVIDIA H200 Tensor Core GPU’ on-board
The H200 GPU is optimised for developers and researchers of large-scale language models with significantly increased GPU memory capacity and memory bus bandwidth - approximately 1.7 times the capacity and 1.4 times the memory bandwidth of the NVIDIA H100 Tensor Core GPU.
▲NVIDIA Spectrum-X
2. first domestic cloud provider to adopt NVIDIA Spectrum-X
First in Japan to adopt NVIDIA Spectrum―Ⅹ, the world's first Ethernet network designed specifically for AI workloads; Spectrum-X enhances network performance and enables faster processing, analysis and execution of AI workloads.
3. cloud network acceleration with NVIDIA BlueField-3 DPUs
NVIDIA BlueField-3 data processing units accelerate GPU access to data, streamline the delivery of AI applications and enhance cloud infrastructure security posture.
4. ultra-fast storage from DDN
DDN's high-speed storage is adopted for optimum performance in combination with the NVIDIA platform. Provides a one-stop AI development platform with powerful performance.
5．rapid environment construction and management with NVIDIA AI Enterprise
NVIDIA AI Enterprise is an end-to-end, cloud-native software platform that accelerates data science pipelines and streamlines the development and deployment of production-grade copilots and other generative AI applications.
6. industry-standard job scheduler Slurm.
The job scheduler is the industry standard for cluster systems. It provides resource allocation, job control and monitoring functions.
【About GMO Internet Group Inc.】
Founded in December 1995, GMO Internet Group, Inc. has focused its management resources on the provision of an Internet platform, under the corporate slogan: “Internet for Everyone.” The Group’s mission is to make the Internet more enriching and more convenient for all.
The company operates across four core domains:
・Internet Infrastructure
・Internet Advertising and Media
・Internet Finance
・Crypto Assets
As of the end of September 2024, the Group serves 15.18 million customers, with a network of 111 group companies, including 10 publicly listed entities, and approximately 7,500 group partners.
With the vision of becoming the No.1 corporate group that creates the future with AI, the Group is actively implementing generative AI across all companies to:
1. Save time and reduce costs
2. Improve the quality of existing services
3. Develop new services in the AI industry
