SAP Leonardo Machine Learning Portfolio is First Enterprise Offering to Use NVIDIA’s Volta AI Platform


Some of the most fabulous duos of all times like Lennon and McCartney, macaroni and cheese, or Kirk and Spock can exist without each other, but together their value is much greater than the sum of their parts.

In the business world, partnerships on an equal footing like these are part of the key to success in competitive and fast-moving environments. They bring access to expertise, more effective products and services, and greater potential for innovation and stability.

Earlier this year, SAP and NVIDIA expanded their collaboration to create business applications based on artificial intelligence. Now, as NVIDIA’s GPU Technology Conference kicks off in Munich, Germany, the partnership has gained even further substance.

SAP installed its first NVIDIA DGX-1 systems – the world’s first AI supercomputer – in Israel and Potsdam in 2016. This was followed by the implementation of NVIDIA DGX-1 systems with NVIDIA Tesla P100 graphics processing units (GPUs) in SAP’s production data center in St. Leon-Rot, Germany and in SAP’s Innovation Labs in Palo Alto, California, and Singapore in September 2017.

Each DGX-1 features eight P100 GPU accelerators – with the power of 250 two-socket servers – and provides unprecedented computing power. High performance is essential for deep learning training to power SAP’s efforts in machine learning – a subset of AI – that imitates the brain in processing unstructured data and creating patterns to support decision making.

Machine learning describes algorithms that can learn from experience, usually data, without having to be explicitly programmed. Learning in this context means optimizing the parameters for a machine learning model to solve the task. Machine learning is part of the broader field of artificial intelligence (AI).

Deep learning describes the intensive computing process involved with training and deploying layers of servers called neural networks. These networks can learn complex, non-linear problems from the input data. Deep learning networks derive their name from their “deep architectures” with several hidden layers, and have led to breakthroughs in several machine learning tasks and are currently the best bet in getting us closer to some of the goals of AI, for example making computers see and understand language.

From the outset of SAP’s machine learning efforts, NVIDIA’s computing platform has promoted the company’s training of data sets and algorithms – the core of intelligent machine learning applications in the SAP Leonardo Machine Learning portfolio.

Introducing Volta-Based Technology

Now, in an important next step, SAP upgraded its DGX-1 systems in St. Leon-Rot with the NVIDIA’s latest generation accelerators: Tesla V100 GPUs based on the company’s Volta architecture.

As SAP Chief Innovation Officer Juergen Mueller explains, “We are proud to be the first enterprise world-wide that has NVIDIA Volta technology in their productive machine learning environment.”

One NVIDIA DGX-1 system with Volta puts the power of 800 Central Processing Units (CPU) in a box. It’s also almost three times faster for training and four times faster for inference than its predecessor as per NVIDIA research on the ResNet-50 Deep Neural Network.

The new NVIDIA DGX-1 systems with Volta offer SAP a powerful engine for AI development and best-in-breed processing capabilities, making it ideal as the primary hardware platform for the SAP Leonardo Machine Learning Foundation.

This installation paves the way for SAP Leonardo Machine Learning to be the first enterprise offering using the Volta platform.

What’s in It for the Customer?

With SAP Leonardo Machine Learning, SAP brings digital intelligence to enterprise offerings and creates tremendous opportunities for customers to realize greater benefits through automated processes, targeted results-driven marketing, superior customer service, as well as increased agility and process efficiency. Some of the SAP Leonardo Machine Learning solutions trained on NVIDIA GPUs include the following.

  • SAP Brand Impact is a cloud-based machine learning solution that automatically analyzes large volumes of videos to detect brand logos in moving images and to measure their time, size, and location exposure. This helps media agencies and brands gain accurate knowledge about the return on investment of sponsorship and advertisement in film and television.
  • SAP Service Ticket Intelligence helps build an optimized customer service by automatically categorizing service tickets and providing resolution recommendations for the service center agent. By leveraging deep neural networks, the solution understands the ticket’s unstructured text message, classifies it, and recommends a solution based on best practices of previously answered tickets by the agent.
  • The solution prototype SAP Accounts Payable automates the accounts payable process by processing incoming invoices and extracting relevant information to existing financial systems. The application offers an ever-expanding suite of services like Invoice to Record that extracts information and posts the data to the general ledger, and Vendor Matching that identifies an invoice and matches it to the vendor database to find the correct vendor ID.
  • SAP Remittance Advice supports the matching of customer payments to invoices. It extracts payment information from unstructured documents like email, PDF or paper and uses it to automate the clearing process by making them available for further processing in other applications like SAP Cash Application.

Additionally, several functional services available on the SAP Leonardo Machine Learning Foundation like Image Classification, Image Feature Extraction, and Product Image Classification have been trained on NVIDIA GPUs as well.

The partnership between SAP and NVIDIA to bring DGX-1 systems with Volta to production in the SAP Data Center will give SAP customers access to machine learning services and applications from SAP’s own Data Center infrastructure. This is an additional, data-secure, managed cloud deployment option for SAP’s customers. With NVIDIA DGX-1 systems as the underlying hardware, the performance of SAP’s machine learning offerings improves.

The benefits for the customer arising from this partnership are remarkable as it accelerates enterprise intelligence by automating business processes with intelligent applications, by optimizing company spend to drive business results with AI, and by making customers’ everyday work better with the power of machine learning.

To make the vision of the intelligent enterprise a reality, SAP’s next bold step will be to incorporate SAP Leonardo Machine Learning capabilities into the new release of SAP S/4HANA, SAP’s next-generation ERP suite. This provides real-time business insights to the customer while freeing employees from repetitive tasks – a technologic evolution for which even Kirk and Spock would be proud.

For more information about deep learning, join this free openSAP course, Enterprise Deep Learning with TensorFlow.

To learn more, download the e-book, “NVIDIA and SAP: AI for the Intelligent Enterprise