Lenovo EveryScale HPC & AI Software Stack

The Lenovo EveryScale HPC & AI Software Stack combines open-source with proprietary best-of-breed Supercomputing software to provide the most consumable open-source HPC software stack embraced by all Lenovo HPC customers.

It provides a fully tested and supported, complete but customizable HPC software stack to enable the administrators and users in optimally and environmentally sustainable utilizing their Lenovo Supercomputers.

The software stack is built on the most widely adopted and maintained HPC community software for orchestration and management. It integrates third party components especially around programming environments and performance optimization to complement and enhance the capabilities, creating the organic umbrella in software and service to add value for our customers.

The software stack offers key software and support components for orchestration and management, programming environments and services and support, as shown in the following figure.

Figure 1. Lenovo EveryScale HPC & AI Software Stack

Lenovo EveryScale HPC & AI Software Stack is a modular software stack tailored to our customer's needs. Thoroughly tested, supported and periodically updated, it combines the latest open-source HPC software releases to enable organizations with an agile and scalable IT infrastructure.

The Lenovo EveryScale HPC & AI Software Stack provides the following benefits to customers.

Overcoming the Complexity of HPC Software

An HPC system software stack consists of dozens of components, that administrators must integrate and validate before an organization’s HPC applications can run on top of the stack. Ensuring stable, reliable versions of all stack components is an enormous task due to the numerous interdependencies. This task is very time consuming because of the constant release cycles and updates of individual components.

The Lenovo EveryScale HPC & AI Software Stack is fully tested, supported and periodically updated to combine the latest open-source HPC software releases, enabling organizations with an agile and scalable IT infrastructure.

Benefits of the Open-source Model

Going forward, in IDC's opinion, the development model exemplified by Linux is more workable. In this model, stack development is driven primarily by the open-source community and vendors offer supported distributions with additional capabilities for customers that require and are willing to pay for them. As the Linux initiative demonstrates, a community-based model like this has major advantages for enabling software to keep pace with requirements for HPC computing and storage hardware systems.

This model delivers new capabilities faster to users and makes HPC systems more productive and higher returning investments.

A fair number of foundational open source HPC software components already exist (e.g., Open MPI, Rocky Linux, Slurm, OpenStack, and others). Many HPC community members are already taking advantage of these.

Customers will benefit from the HPC community, as the community works to integrate a multitude of components that are commonly used in HPC systems and are freely available for open source distribution.

The key open-source components of the software stack are:

EAR is a powerful European open-source energy management suite supporting anything from monitoring over power capping to live-optimization during the application runtime. Lenovo is collaborating with Barcelona Supercomputing Centre (BSC) and EAS4DC on the continuous development and support and offers three versions with differentiating capabilities.

EAR is a powerful European open-source energy management suite supporting anything from monitoring over power capping to live-optimization during the application runtime. Lenovo is collaborating with Barcelona Supercomputing Centre (BSC) and EAS4DC on the continuous development and support and offers three versions with differentiating capabilities

• Confluent Management

  Confluent is Lenovo-developed open-source software designed to discover, provision, and manage HPC clusters and the nodes that comprise them. Confluent provides powerful tooling to deploy and update software and firmware to multiple nodes simultaneously, with simple and readable modern software syntax.

• Slurm Orchestration

  Slurm is integrated as an open source, flexible, and modern choice to manage complex workloads for faster processing and optimal utilization of the large-scale and specialized high-performance and AI resource capabilities needed per workload provided by Lenovo systems. Lenovo provides support in partnership with SchedMD.

• Energy Aware Runtime

  EAR is a powerful European open-source energy management suite supporting anything from monitoring over power capping to live-optimization during the application runtime. Lenovo is collaborating with Barcelona Supercomputing Centre (BSC) and EAS4DC on the continuous development and support and offers three versions with differentiating capabilities.

• Open OnDemand

  Open OnDemand (OOD) is an open source, web-based access portal that provides a unified Graphical User Interface (GUI) for seamless access to HPC and AI clusters. It enables users to securely access compute resources, submit and monitor jobs, and launch rich interactive applications, such as Jupyter Notebooks, VS Code, RStudio, and TensorBoard, directly from a standard web browser. Through these interactive environments, OOD enables end-users to run complete AI and data-science workflows based on popular frameworks like TensorFlow, PyTorch, RAPIDS, and more, without needing to manage the underlying cluster complexity. Supporting multiple resource managers, including Slurm, PBS Pro, LSF, and Kubernetes, Open OnDemand makes it easy to consolidate HPC and AI workloads on a shared infrastructure, significantly improving usability, accessibility, and overall productivity while eliminating the need for command-line expertise.

The following orchestration and management software is available with Lenovo EveryScale HPC & AI Software Stack:

• Confluent (Best Recipe interoperability)

  Confluent is Lenovo-developed open source software designed to discover, provision, and manage HPC clusters and the nodes that comprise them. Confluent provides powerful tooling to deploy and update software and firmware to multiple nodes simultaneously, with simple and readable modern software syntax. Additionally, Confluent’s performance scales seamlessly from small workstation clusters to thousand-plus node supercomputers. For more information, see the Confluent documentation and the Lenovo Confluent Management Software paper.

• Slurm

  Slurm is a modern, open-source scheduler designed specifically to satisfy the demanding needs of high-performance computing (HPC), high throughput computing (HTC) and AI. Slurm is developed and maintained by SchedMD®. Slurm maximizes workload throughput, scale, reliability, and results in the fastest possible time while optimizing resource utilization and meeting organizational priorities. Slurm automates job scheduling to help admin and users manage the complexities of on-prem, hybrid, or cloud workspaces. Slurm workload manager executes faster and is more reliable ensuring increased productivity while decreasing costs. Slurm’s modern, plug-in-based architecture runs on a RESTful API supporting both large and small HPC, HTC, and AI environments. Allow your teams to focus on their work while Slurm manages their workloads.

• SchedMD Slurm Support service capabilities for Lenovo HPC and AI systems include:
  • Level 3 Support: High-performance systems must perform at high utilization and performance to meet end users and management return on the investment expectations. Customers covered by a support contract can reach out to SchedMD engineer experts to promptly resolve complex workload management issues and receive answers back to complex config questions quickly, instead of taking weeks or even months to try to resolve them in-house.
  • Remote Consulting: Valuable assistance and implementation expertise that speeds custom configuration tuning to increase throughput and utilization efficiency on complex and large-scale systems. Customers can review cluster requirements, operating environment, and organizational goals directly with a Slurm engineer to optimize the configuration and meet organizational needs.
  • Tailored Slurm Training: Tailored Slurm expert training that empowers users on harnessing Slurm capabilities to speed projects and increase technology adoption. A customer scoping call before the onsite Instruction ensures coverage of specific use cases addressing organization needs. An in-depth and comprehensive technical training is delivered in a hands-on lab workshop format to help users feel empowered on Slurm best practices in their site-specific use cases and configuration.
• NVIDIA Unified Fabric Manager (UFM) (ISV supported)

  NVIDIA Unified Fabric Manager (UFM) is InfiniBand networking management software that combines enhanced, real-time network telemetry with fabric visibility and control to support scale-out InfiniBand data centers. For more information, see the NVIDIA UFM product page.

  The two UFM offerings available from Lenovo are as follows:

  • UFM Telemetry for Real-Time Monitoring

    The UFM Telemetry platform provides network validation tools to monitor network performance and conditions, capturing and streaming rich real-time network telemetry information, application workload usage, and system configuration to an on-premises or cloud-based database for further analysis.

  • UFM Enterprise for Fabric Visibility and Control

    The UFM Enterprise platform combines the benefits of UFM Telemetry with enhanced network monitoring and management. It performs automated network discovery and provisioning, traffic monitoring, and congestion discovery. It also enables job schedule provisioning and integrates with industry-leading job schedulers and cloud and cluster managers, including Slurm and Platform Load Sharing Facility (LSF).

The following table lists all Orchestration and Management software available with Lenovo EveryScale HPC & AI Software Stack.

Programming environment

The following programming software is available with Lenovo EveryScale HPC & AI Software Stack.

• NVIDIA CUDA

  NVIDIA CUDA is a parallel computing platform and programming model for general computing on graphical processing units (GPUs). With CUDA, developers are able to dramatically speed up computing applications by harnessing the power of GPUs. When using CUDA, developers program in popular languages such as C, C++, Fortran, Python and MATLAB and express parallelism through extensions in the form of a few basic keywords. For more information, see the NVIDIA CUDA Zone.

• NVIDIA HPC Software Development Kit

  The NVIDIA HPC SDK C, C++, and Fortran compilers support GPU acceleration of HPC modeling and simulation applications with standard C++ and Fortran, OpenACC directives, and CUDA. GPU-accelerated math libraries maximize performance on common HPC algorithms, and optimized communications libraries enable standards-based multi-GPU and scalable systems programming. Performance profiling and debugging tools simplify porting and optimization of HPC applications, and containerization tools enable easy deployment on-premises or in the cloud. For more information, see the NVIDIA HPC SDK.

• Intel oneAPI
  The Intel oneAPI Base & HPC Toolkit is a comprehensive software development suite designed to empower developers in creating HPC & AI solutions that exploit the full potential of modern hardware architectures. This toolkit encompasses an array of advanced tools, libraries, and compilers, enabling programmers to efficiently design, optimize, and deploy parallel applications across diverse computing platforms, including CPUs, GPUs, and FPGAs. With a focus on fostering code portability and performance scalability, the Intel oneAPI Base & HPC Toolkit equips developers with the means to enhance productivity, streamline software development, and achieve exceptional performance outcomes in the realm of high-performance computing.
  • For more information, see Intel® oneAPI Base and HPC Toolkit
  • System-based pricing (introduced in mid-2023) can range from a small system (64-256 nodes), medium system (257-512 nodes), or large system (512+ nodes)
  • Developer-based pricing is for systems with fewer than 64 nodes and offers support for a limited number of users.
  • Part numbers are available for Commercial or Academic customers.
  • Support Renewals are available.
  • Commercial parts have different part numbers if they are quoted with or without Intel hardware.

The following tables list the relevant ordering part numbers.

The following software suite is available with Lenovo EveryScale HPC & AI Software Stack.

• NVIDIA AI Enterprise
  NVAIE is an end-to-end, cloud-native suite of AI and data analytics software, optimized, certified, and supported by NVIDIA to run on VMware vSphere and bare-metal with NVIDIA-Certified Systems™.  It includes key enabling technologies from NVIDIA for rapid deployment, management, and scaling of AI workloads in the modern hybrid cloud. NVAIE is licensed on a per-GPU basis and can be purchased as either a perpetual license with support services, or as an annual or multi-year subscription.
  • The perpetual license provides the right to use the NVIDIA AI Enterprise software indefinitely, with no expiration. NVIDIA AI Enterprise with perpetual licenses must be purchased in conjunction with one-year, three-year, or five-year support services. A one-year support service is also available for renewals.
  • The subscription offerings are an affordable option to allow IT departments to better manage the flexibility of license volumes. NVIDIA AI Enterprise software products with subscription includes support services for the duration of the software’s subscription license.

The features of NVAIE Software are listed in the following table.

• NVIDIA Run: ai

NVIDIA Run: ai is an enterprise platform for AI workloads and GPU orchestration, accelerating AI operations with dynamic orchestration across the AI life cycle, maximizing GPU efficiency, and integrating seamlessly into hybrid AI infrastructure. The platform provides features such as AI-native workload orchestration, unified AI infrastructure management, flexible AI deployment, and open architecture, supporting public clouds, private clouds, hybrid environments, and on-premises data centers. For more information, see the NVIDIA Run:ai product page.

For more information, see the NVIDIA Run:ai product page:
https://www.nvidia.com/en-us/software/run-ai/

• NVIDIA Mission Control

NVIDIA Mission Control is a software platform that powers AI factory operations, bringing instant agility to inference and training workloads, and full-stack intelligence for world-class infrastructure resiliency. It simplifies AI operations, from cluster deployment to workload orchestration, providing benefits such as instant agility, hyperscale-grade efficiency, and gold-standard infrastructure resiliency. The platform includes features like seamless workload orchestration, energy-optimized power profiles, and autonomous job recovery.

For more information, see the NVIDIA Mission Control product page:
https://www.nvidia.com/en-us/data-center/mission-control/

• ClearML: Innovating AI at Scale with Lenovo

As a Lenovo innovator in the AI software stack, ClearML delivers an enterprise-grade AI and GenAI platform that accelerates the entire AI lifecycle. Designed for flexibility and performance, ClearML provides a cloud-like experience with the efficiency and security of Lenovo’s infrastructure solutions.

ClearML’s architecture is built on three layers – Infrastructure Control Plane, AI Development Center, and GenAI App Engine – each enabling Lenovo customers to unlock the full potential of AI:

• Infrastructure Control Plane – ClearML enables GPU optimization and management across heterogeneous environments with dynamic resource allocation for NVIDIA and AMD GPUs, granular quota management with per-tenant billing, and isolated multi-tenancy. The platform supports silicon-agnostic workload scheduling across any environment, enables hybrid multi-cluster orchestration with cloud bursting, and deploys flexibly across on-premise or air-gapped infrastructure.
  • Dynamic Fractional GPUs – ClearML enables highly efficient use of GPU resources by allowing granular allocation and sharing of GPU fractions. This dynamic approach optimizes performance for a wide range of AI workloads, supporting both NVIDIA MIG-capable and non-MIG devices and AMD MI300X and newer GPUs. It allows organizations to maximize their hardware investment and adapt resource allocation in real time as workload demands change.
  • Multi-tenant, Multi-cluster – The platform supports secure, dynamic resource sharing across multiple users and clusters. Its secure multi-tenancy model includes robust network isolation, ensuring privacy and compliance for different teams or projects. Resources are allocated flexibly based on real-time demand, and new tenants can be onboarded quickly without the need for dedicated clusters, maximizing overall resource utilization.
• AI Development Center – The platform offers a comprehensive workbench for AI/ML development, integrating tools for data management, experiment tracking, and pipeline orchestration. Users can launch familiar development environments like Jupyter and VSCode with a single-click, manage datasets, and automate complex AI workflows—all from a unified interface. This abstraction allows data scientists and engineers to focus on their code and experiments, minimizing the need to interact directly with Kubernetes or underlying infrastructure.
• GenAI App Engine – ClearML simplifies the testing, deployment, and management of large language models (LLMs). With one-click access, users can deploy custom or off-the-shelf (Hugging Face) models, NVIDIA NIM containers, or custom applications with pre-configured networking and RBAC authentication. An endpoint dashboard provides teams with full visibility on usage and performance, and static routing provides horizontal scalability for peak periods. The GenAI App Engine supports secure deployment of both open-source and custom models.
• NVIDIA AI Enterprise integration – ClearML integrates with NVIDIA AI Enterprise (NVAIE), including dynamic license allocation that allows organizations to optimize costs and scale efficiently. The platform orchestrates NVIDIA-optimized containers (such as NIM and Triton), accelerating AI development and providing observability for inference workloads.

Lenovo and ClearML together deliver a secure, scalable, and innovative AI platform – empowering organizations to accelerate innovation and turn vision into production-grade AI solutions.

For more information: https://clear.ml/

The following tables list the ordering part numbers.

Note: SKUs marked ‘INC’ are part of NVIDIA’s Inception Program, a global initiative supporting AI startups.
Learn more at: https://www.nvidia.com/en-us/startups/

Note: SKUs marked ‘INC’ are part of NVIDIA’s Inception Program, a global initiative supporting AI startups.
Learn more at: https://www.nvidia.com/en-us/startups/

The following Energy monitoring and optimization software is available with Lenovo EveryScale HPC & AI Software Stack.

• Energy Aware Runtime (EAR)
  EAR is licensed under EPL 2.0 and developed by BSC and Energy Aware Solutions (EAS). While the EAR core version remains open-source, EAR also features extensions developed by the EAS team and provided under a proprietary license. EAR packs are the different solutions provided by EAS, which include the EAR core, EAR extensions, and EAS installation and support services. EAR packs can be purchased from Lenovo under the EveryScale HPC & AI Software Stack CTO and are delivered through EAS.
• EAS offers three EAR packs:

• EAR Detective Pro – Includes Data Center monitoring, Analysis and Accounting modules.
• EAR Optimizer – Includes all features of EAR Detective Pro, with the addition of the energy optimization module for both CPU and GPU.
• EAR Optimizer Pro – Build upon EAR Optimizer by introducing the smart power capping module.

EAR is compatible with Slurm, PBS Pro and Kubernetes clusters.

The following table lists the ordering part numbers for EAR Licensing, Support and Services

Note: These part numbers are applicable only to clusters that meet all of the following criteria:

• Contain a single CPU type and a single GPU type
• Have a combined total of up to 1,000 sockets (including both CPU and GPU sockets)
• Do not require smart power capping (i.e. do not use EAR Optimizer Pro)

Special BID Requirement: A Special BID from EAS is required in the following scenarios:

• The cluster exceeds 1,000 sockets
• The cluster includes multiple CPU or GPU types
• The deployment requires smart power capping, available only through EAR Optimizer Pro

For more information, see these resources:

• Implementing Lenovo Confluent Management Software (Planning / Implementation):
  https://lenovopress.lenovo.com/lp2266-implementing-lenovo-confluent-management-software
• Lenovo DSCS configurator:
  https://dcsc.lenovo.com
• Optimizing Power and Energy in HPC data centers with Energy Aware Runtime
  https://lenovopress.lenovo.com/lp1646
• Energy Aware Runtime software and documentation:
  http://www.eas4dc.com
• Lenovo Confluent documentation:
  https://hpc.lenovo.com/users/documentation/
• Lenovo Compute Orchestration in HPC Data Centers with Slurm - Solution Brief:
  https://lenovopress.lenovo.com/lp1701-lenovo-compute-orchestration-in-hpc-data-centers-with-slurm