GMO GPU Cloud: Managed HPC Cluster Service Now Supports Container Runtime 'Enroot' and Slurm Plugin 'Pyxis' [GMO Internet]

GMO Internet, Inc. (Representative Director, President and Executive Officer: Masashi Ito; hereinafter, GMO Internet), a member of GMO Internet Group, announced that its high-performance GPU cloud service, 'GMO GPU Cloud' (https://gpucloud.gmo/), will begin supporting the container runtime 'Enroot' (*1) and the Slurm plugin 'Pyxis' (*2) for its managed HPC cluster service starting May 13, 2026 (Wednesday).

With this new support, users of the 'GMO GPU Cloud' managed HPC cluster service will be able to utilize NVIDIA's container runtime 'Enroot' and the 'Pyxis' plugin, which directly calls containers from Slurm, in addition to the previously provided container runtime 'Singularity'. This allows for more flexible operations, as users can select the optimal container runtime and job submission method according to their workload and usage scenarios. Furthermore, this marks the first commercial HPC-type GPU cloud service in Japan to standardly support both 'Singularity' and 'Pyxis/Enroot' (*3).

(*1) Enroot: A lightweight, rootless container runtime developed and provided by NVIDIA. It supports OCI-compatible container images and is characterized by its design strongly focused on HPC environments.

(*2) Pyxis: A plugin for Slurm provided by NVIDIA. It facilitates container execution on HPC clusters by directly calling 'Enroot' containers from Slurm's job submission commands.

(*3) As of April 16, 2026, according to our company's research.

[Background of Service Provision]

In recent years, with the rapid advancement of research and development in generative AI and large language models (LLMs), ensuring the portability and reproducibility of container-based workloads has become indispensable in the HPC domain. The 'GMO GPU Cloud' managed HPC cluster service has previously offered 'Singularity', a container runtime widely used in HPC environments.

However, there has been increasing demand for features such as 'natively launching containers directly from Slurm's job submission commands' and 'running workloads compliant with NVIDIA's software stack as is'. To meet these demands, we have added support for 'Enroot', NVIDIA's lightweight container runtime for HPC, and 'Pyxis', a SPANK (*4) plugin that seamlessly calls 'Enroot' from Slurm.

(*4) SPANK (Slurm Plug-in Architecture for Node and job Kontrol): A plugin architecture for extending Slurm's job execution platform.

[Value and Main Functions of This Support]

* Users can select the optimal container runtime for their workload from 'Singularity' and 'Pyxis/Enroot'. * Docker/OCI (Open Container Initiative) compatible container images can be directly submitted to Slurm jobs via 'Pyxis'. * Improved reproducibility and portability of research and development environments, enabling smooth migration from local environments to cluster environments. * Flexible container operations are possible while leveraging existing job script assets.

Through this support, the 'GMO GPU Cloud' managed HPC cluster service becomes a flexible platform that allows users to select container runtimes according to the characteristics of their workloads and existing development environments. By improving environment reproducibility and portability, it contributes to reducing the cost of migrating from local to production clusters and accelerating the development and research cycle.

[Future Developments]

In an era where AI forms the foundation of society and industry, the domestic development of high-performance GPU infrastructure is a critical issue that will determine Japan's AI competitiveness. GMO Internet will continuously evolve the 'GMO GPU Cloud' managed HPC cluster service, providing development and research infrastructure for AI/machine learning and HPC domains, and contributing to the creation of social and industrial innovation as an indispensable domestic infrastructure for Japan's AI industry.

Past Reference Releases (Excerpt)

* June 11, 2024: GMO Internet Group Becomes First Domestic Cloud Provider to Adopt NVIDIA Spectrum-X for Generative AI GPU Cloud Service https://www.gmo.jp/news/article/9005/ * November 22, 2024: 'GMO GPU Cloud', Ranked in TOP500 Supercomputer List, Begins Service https://inter