SAP at GTC: Advancing GPU-Accelerated Machine Learning

by Renee Yao

NVIDIA made many announcements extending its GPU computing platform for AI innovations last week at GTC 2018. We’re working with SAP to extend GPU-accelerated machine learning to enterprises.

TensorRT 4 and Kubernetes on GPUs

NVIDIA announced TensorRT 4 and TensorFlow integration as well as Kubernetes support on multi-cloud GPU clusters. SAP leverages both products to put its Leonardo Machine Learning offerings into production.

TensorRT is our deep learning inference optimizer. Markus Noga, head of Machine Learning at SAP, stated that the software maker is experiencing a 45x increase in inferencing speed and throughput with it, compared to running on a CPU-based platform, and believes that it can dramatically improve the productivity of SAP’s enterprise customers.

TensorRT 4, the latest version, now provides capabilities to accelerate speech recognition, neural machine translation and recommender systems.

To facilitate inference deployment in the multi-cloud environment, SAP shared that it’s using Kubernetes for scheduling deep learning workloads.

“SAP Leonardo Machine Learning has been running Kubernetes with NVIDIA GPUs in SAP’s biggest data center since October 2017,” said Dr. Sebastian Wieczorek, head of Leonardo Machine Learning foundation. “We schedule GPU-accelerated TensorFlow training jobs using Kubernetes on NVIDIA DGX-1 deep learning supercomputers.”

2x Memory Increase for Tesla V100

NVIDIA also announced it doubled the memory for Tesla V100 GPUs to 32GB. Trained on NVIDIA DGX-1, which has eight V100 GPUs, the SAP Brand Impact application automatically analyzes brand exposure in videos by leveraging advanced computer vision techniques and proprietary algorithms. Given the nature of high-definition video footage and the need for real-time results, the increase in memory helped SAP with higher accuracy.

“We evaluated DGX-1 with the new Tesla V100 32GB for our SAP Brand Impact application, which automatically analyzes brand exposure in videos in near real time,” said Michael Kemelmakher, vice president, SAP Innovation Center, Israel. “The additional memory improved our ability to handle higher definition images on a larger ResNet-152 model, reducing error rates by 40 percent on average. This results in accurate, timely and auditable services at scale.”

Leonid Bobovich, lead architect and dev manager on the SAP Brand Impact team, continues to train more jobs and plans to share additional improved performance at SAP SAPPHIRE in June.

CXO Summit: AI Roadmap and Journey Workshop

Last Wednesday, NVIDIA hosted an exclusive AI for Business CXO Summit, which brought together 60+ senior executives to discuss cross-industry challenges and insights for building AI enterprises and practices. We were honored to have Juergen Mueller, chief innovation officer at SAP, as one of the co-hosts.

Several key insights were highlighted during the full-day workshop:

  • Importance of data quality for accurate model training
  • Distributed data system for security, privacy and transparency
  • Open source model of staff management for employees’ trust and productivity
  • Formulate problems into components for effective solutions

As SAP and NVIDIA innovate together and bring more GPU-accelerated solutions to the market, improving customer and employee productivity and minimizing waste will lay the foundation for building an AI enterprise.

Making Business Application Intelligent Using SAP Leonardo Machine Learning

At GTC, Frank Wu, head of SAP Machine Learning Business Network, and Nazanin Zaker, SAP lead data scientist, delivered an informative talk on building a machine learning application, called Catalog Normalization, which processes catalogs received from suppliers, extracts attributes from free-text descriptions and normalizes attribute names and values.

The application is also trained on NVIDIA DGX systems for faster deployment.

SAP Leonardo Machine Learning provides capabilities, micro-services, applications and technology that enable the integration and adoption of machine learning in the enterprise. Wu and Zaker leveraged this application to show how deep learning models can be used to solve this problem for enterprises.

Lots happened at GTC, but don’t worry if you missed it. We’ll be coming to you around the world. Next stop: SAP SAPPHIRE in Orlando, Fla. In the meantime, tune in here for the latest updates.

Similar Stories