GMO Internet Launches GPU Cloud byGMO, a GPU Server For Machine Learning and Deep Learning: Accelerating the improvement in GPU environment, and GPU resource utilization in R&D and services within the GMO Internet Group

Tokyo, Japan – March 28, 2019 – GMO Internet (https://www.gmo.jp/en/ ) will today launch its new cloud service “GPU Cloud byGMO” (https://gpu.cloud.gmo.jp ) equipped with NVIDIA’s latest graphics processing unit (GPU) Tesla® V100, which is capable of fast arithmetic processing that is needed in machine learning, deep learning, and artificial intelligence (AI).

As part of these efforts to develop leading‐edge technology using an AI and a machine learning, GMO Internet will build infrastructure base equipped with GPU, and promote technological research and technology utilization in businesses related to AI and machine learning, including GMO Internet Group’s Internet Securities, Ecommerce and Internet Media businesses.

GPU resources are always available to GMO Internet Group’s engineers through Internal Engineer System, promoting effective use of resources by owning the GPU environment. It is hoped that we will be able to speed up GPU resource utilization in research and services as well as build a foundation to create new ideas by utilizing not only AI and machine learning but also the GPU environment by creating an internal environment in which engineers can use GPU resources whenever they want.

 

 

GPU Cloud byGMO

GPU is a processor that has been developed for the rendering (the processing of an image such as 3D graphics and video) that needs sophisticated parallel arithmetic processing. Recently, GPU demand has been increasing in the areas that need sophisticated computation, such as machine learning, deep learning, and AI.

GMO Internet’s new service “GPU Cloud byGMO” provides fast arithmetic processing resources combining fast interconnect InfiniBand EDR with the Lustre - a system that distributes files - as a bare metal server (physical server) through the cloud, while employing NVIDIA’s latest GPU Tesla® V100. Customers do not need to make an investment in equipment on their own and can procure GPU resources whenever and as much as they need.

GPU Cloud byGMO employs the Tesla® V100, the performance of which is up to 100 CPUs per unit. It helps engineers and researchers work on the issues that need data processing, which has been difficult until now.

 

■ The major characteristics of GPU Cloud byGMO

（1）Customers can use GPU environment whenever and as much as they need

Customers can start using GPU Cloud byGMO on the day they have placed the order – at the earliest – as the introduction of the cloud service is not time-consuming. Customers can procure flexibly the GPU resources if more computing resources are needed or when spot usage offsets the shortage of computing resources that the customers own. The company will provide a monthly plan when the service is launched but it plans to offer an hourly plan in the future.

 

（2）The development environment is already set up, and the GPU environment can be used immediately

GMO Internet offers a programming environment for GPU, setting up a software development environment, such as OpenCL and NVIDIA’s CUDA®. It supports massive learning by multi-nodes through the adoption of container engine Singularity, can import Docker^*1 image, and minimizes the set up supporting the environment.

^{*1 Docker is an open source software that provides the container-like virtual environment, which can allow the file to be independent, though it is one OS environment. Docker image is a file system needed to make the container.}

 

（3）Realize broadband/low latency with fast interconnect InfiniBand EDR

We should achieve fast communication with broadband/low latency^*2 in communication between file storage and GPU server in order to utilize the features of the GPU. We enable superfast (100 Gb/s) forwarding in communication between file storage and GPU server, or between GPU servers, by adopting a fast interconnect InfiniBand EDR that is used by parallel computing and high-performance computing (HPC) in the field of science and technologies.

^{*2 Latency is a delay in communication.}

 

（4）Employing Lustre, a file distribution system

Intensive computation requires a high-performance file server storing original data. The fields of HPC and genome that carry out massive analysis and simulation employ a file distribution system^*3, which is a solution to the above issues, and the number of customers using Lustre is growing. We realize a loading and writing of data on the GPU server and data sharing between up to tens of thousands of GPU servers by introducing Lustre. This enables data processing on GPU server.

^{*3 File distribution system is a file system that stores data across multiple computers, but the data can be treated as one data.}

 

（5）The NVLINK connection between GPUs

GMO Internet has expanded the bandwidth of PCIe^*4 through NVIDIA® NVLink™ technology, which enables superfast internode connection, and prepared a plan, the system of which is comprised of up to 4 units of interconnected Tesla ®V100 per server. This realizes a speed that is about 5 times the speed of bandwidth when only normal PCIe is interconnected.

^{*4 PCIe is an abbreviation of PCI Express and is one of the expansion slot standard specifications to boost computer performance.}

 

GMO Internet supports a solution to the important issues in the areas that need massive computation, such as machine learning, deep learning, and AI, by launching GPU Cloud byGMO.

 

Main use cases

GPU Cloud byGMO will work with any request to procure advanced computing resources. Specifically, it is perfect for a series of simulations, science and technologies, and computation in the fields of machine learning, deep learning, and AI, and a molecular dynamics simulation and application simulation in the field of drug development. It is hoped that the GPU Cloud byGMO will be used in various industries, leading to enhanced work efficiency, the detection of abnormality, match, and speculations especially in the fields of AI, so GPU server demand is increasing, with the increase in demand for the AI. It is also possible to process images, such as real-time, high-resolution volume rendering of the viewer for the analysis of 3D graphics for medical use.

 

 

■ Major specs and plans of GPU Cloud byGMO

Customers may choose from 3 plans, depending on the number of GPUs.

Plan		1 GPU	2 GPU	4 GPU
Initial cost (JPY)		882,000	1,560,000	2,880,000
(Installment payment; JPY)		(73,500)	(130,000)	(240,000)
Monthly payments (JPY)		97,000	142,000	230,000
GPU type		NVIDIA® Tesla® V100 SXM2
GPU memory		16 GB	32 GB	64 GB
NVIDIA Tensor cores		640	1,280	2,560
NVIDIA CUDA® cores		5,120	10,240	20,480
CPU		Intel® Xeon® Gold 5122 4 cores 3.60GHz ×2
OS		Ubuntu 16.04
Memory		192 GB
Storage		SSD 480GB ＋ Lustre file storage
Installation location		Japan
Interconnect		InfiniBand EDR 100Gbps
Operational performance	Single-precision floating-point operation	Approx. 15.7 TFLOPS	Approx. 31.4 TFLOPS	Approx. 62.8 TFLOPS
	Double-precision floating-point operation	Approx. 7.8 TFLOPS	Approx. 15.6 TFLOPS	Approx. 31.2 TFLOPS

*Please note that the above may be subject to change.

 

Background of launching GPU Cloud byGMO

A server equipped with GPU is drawing attention in various industries, including machine learning, deep learning, and AI.

GMO Internet has been driving the spread and utilization of the Internet in a lot of industries since the dawn of the Internet industry. As of today, businesses in Internet Infrastructure segment each hold top share in their respective markets in Japan, and the number of contracts has come to approximately 10 million, which is a foundation that supports the development of Japan’s Internet industry.

Amid such circumstance, GMO Internet believes services using a GPU will become a foundation that supports the essential technology for the development of each industry in the future, such as AI, robotics, medicine, autonomous driving, finance, and ray tracing.

Accordingly, GMO Internet decided to provide the GPU Cloud byGMO with an aim to provide solutions to the issues in these fields that need massive computation and to make GPU Cloud byGMO the foundation that supports Japan’s industries.

The GMO Internet Group company GMO Click Global Markets is already working towards financial research using GPU Cloud.

GMO Internet will introduce the expertise and knowledge, gained through in-house GPU resource utilization in research and services, to GPU Cloud byGMO, offering an environment in which customers can easily use the GPU for their services or research.

 

GPU infrastructure base in GMO Internet Group

GMO Internet Group offers advertisement distribution services utilizing ad tech, FX trading services that have No. 1 transaction volume worldwide, and Japan's leading ecommerce website solutions. GMO Internet Group is driving the adoption of GPU Cloud in these fields, putting effort into researching machine learning.

GMO Internet Group is preparing an environment in which GPU Cloud is available to its engineers, to actively promote the GPU Cloud, develop data scientists, and pursue new ideas and new businesses by being exposed to new technologies.

 

 

GMO Internet Group

GMO Internet Group is an Internet service industry leader, developing and operating Japan’s most widely used domain, hosting & cloud, ecommerce, security, and payment solutions. The Group also includes the world’s largest online FX trading platform, as well as online advertising, Internet media, and cryptocurrency related services. GMO Internet, Inc. (TSE: 9449) is headquartered in Tokyo, Japan. For more information, please visit https://www.gmo.jp/en/