Urika-CS
AI and Analytics

ELIMINATE AI COMPLEXITY FOR FASTER RESULTS

Contact Us

AI WORKFLOWS AT THE SPEED OF SUPERCOMPUTING

Cray's Urika®-CS AI and Analytics suite brings you supported and integrated AI applications, engineered for Cray CS series cluster supercomputers.

Pioneering organizations that have committed to artificial intelligence are under pressure to succeed, with time from idea development to model deployment becoming the visible success factor. But the infrastructure for AI — systems, frameworks and tools — is complex and often difficult to set up, administer and use. Cray Urika-XC Analytics SoftwareThe Cray Urika-CS AI suite, designed with IT administrators and data scientists in mind, eliminates complexity and allows data science teams to focus on what matters most: converting the promise of AI into a meaningful reality.

Cray’s Urika-CS AI and Analytics suite addresses four issues facing data scientists and IT teams tasked with implementing AI applications.

  • The right tools for the AI workflow: Data scientists and machine learning engineers use a variety of toolsets and frameworks to accomplish their work — tools that must be available on laptops for model development and shared AI compute systems for complete workflow processing. Additionally, different tools are used by different users to complete the workflow. Machine learning and data engineers use CPU-based systems and data analytics environments like Apache Spark or Python-based data science tools like Anaconda to gather and prepare the data necessary for AI. Data scientists use GPU-based systems and machine intelligence frameworks like TensorFlow to develop, train and refine machine and deep learning models. Cray’s Urika-CS AI suite makes the most popular AI tools and frameworks readily available on Cray CS series systems — CPU- and GPU-focused — for both user communities.
  • Reduced complexity: Machine learning and deep learning are made possible by the availability of open-source tools — which are often difficult and time-consuming to install, configure and maintain. We deliver a comprehensive set of the most popular tools and frameworks as a container, pre-tested and integrated to run on Cray CS series systems.
  • Support for open-source tools and frameworks: AI projects rely on open-source tools and frameworks developed and maintained by a variety of communities. What if there’s a problem with a tool? Is it efficient and economical to have specialized — and expensive — data science or engineering resources focused on making tools work in your environment? With Cray’s Urika-CS suite, we provide the support you need for pre-tested and integrated open-source tools and frameworks, allowing your AI team to focus on what matters most — developing and deployment.
  • Difficulty and time required to train models: Model development is a complex, multi-step process critical to the success of AI projects. Data scientists should be focused on the art and science of the model development workflow — selection and design, training, evaluation, and tuning — and not implementation details like distributed training configuration. Additionally, the time required to train a model can limit the data scientist’s choices in DNN topologies and complexity. The Urika-CS suite includes the unique Cray Distributed Training Framework, designed to simplify the data scientist’s life and reduce model development training time and the time required to achieve accuracy.

Featured Resources

Cray CS-Storm GPU-Accelerated Cluster Supercomputer Brochure

Purpose-built for the most demanding machine learning workloads, Cray CS-Storm technology provides customers with a powerful, accelerator-optimized solution for moving AI applications from pilot to production.

Cray CS500 Cluster Supercomputer Brochure

Cray CS500 cluster supercomputers are industry-standards-based, highly customizable, easy to manage and designed to handle the broadest range of simulations and data-intensive workloads.

See All Resources

AI Unleashed

Cray’s Urika®-CS AI suite provides a pre-integrated, supported set of tools and frameworks integral to the data engineering and science tasks that bring AI from concept to reality.

For data scientists tasked with implementing high-profile AI projects:

  • The Cray Urika-CS AI suite provides a complete toolset — integrated and optimized to harness the power of Cray CS series cluster supercomputers — to remove barriers to successful data preparation and model development, providing faster organizational time to ideation.

For IT staff investing in infrastructure to support AI:

  • The Cray Urika-CS AI suite saves time and resources required to deliver and maintain a complete AI environment, allowing IT resources to focus on business issues, providing lower TCO of systems and faster organizational time to ideation.

Urika-CS AI and Analytics Technology

Bringing AI to the Cray CS series

The Urika®-CS AI and Analytics software suite is a set of powerful data science tools and frameworks integrated and supported on Cray® CS-Storm™ GPU-accelerated systems and CS500™ cluster supercomputing systems. Selected to address the end-to-end data science workflows associated with AI, the Urika-CS suite allows organizations using machine learning and deep learning AI approaches to focus on data and models.

The Right Tool for the Job: Open-Source Analytics at Supercomputer Scale

The Urika-CS suite incudes the latest open-source technologies for AI workflows. Integrated with standard HPC workload managers, the Urika-CS suite brings the power of distributed AI and analytics to Cray CS series CPU and GPU systems.

Cray Urika-CS Software Suite

The Cray Distributed Training Framework

The Urika-CS AI and Analytics suite incudes the Cray Distributed Training Framework , a collection of libraries designed to simplify and speed up the distributed training of large and complex neural network models. Fully supported by Cray, the Distributed Training Framework includes:

  • The Cray PE ML plugin, leveraging supercomputer-class MPI to provide superior distributed training scalability and performance on CPU-based cluster nodes. It also eliminates time-consuming administrative tasks for you. For example, it automatically defines the nodes to use, simplifying the data scientist’s burden of configuring and infrastructure associated with distributed training.
  • The Horovod Open Source Horovod distributed training library for enhanced distributed training and scale on dense-GPU nodes.
  • The most popular frameworks for machine learning (TensorFlow, PyTorch, Keras)

Easy Deployment Using Containers

The Urika-CS AI suite leverages Singularity containers to deliver a fully-tested, integrated and supported AI environment. Cray has removed the complexity of downloading, integrating, configuring and running AI frameworks and toolkits.