Deep learning framework training time improved by approximately 4x

IBM announced today next-generation Power System servers equipped with the newly designed POWER9 processor.

The new Power9 system, specialized for artificial intelligence (AI) workloads requiring large-scale computation, can improve deep learning framework training times by up to approximately four times compared to existing x86 servers, enabling enterprises to rapidly develop more accurate AI applications.

The new Power9-based AC922 Power System is the industry's first to feature built-in PCI-Express 4.0, next-generation NVIDIA NVLink 2.0, and OpenCAPI, which transmits data 9.5 times faster than x86 system-based PCI-E 3.0.

This system is designed to provide enhanced performance in accelerated databases such as Kinetica, as well as popular AI frameworks such as Chainer, TensorFlow, and Caffe. Therefore, data scientists can develop applications such as deep learning in scientific research, real-time fraud detection, and credit risk analysis more quickly.

Power9 is installed on the U.S. Department of Energy’s Summit and Sierra, which are expected to become the world’s most powerful supercomputers, playing a role in supporting the world-leading performance of these two computers, and is already being utilized by numerous organizations such as Google.

Bart Sano, Google’s Vice President of Platform, stated, “Google is paying attention to the achievements IBM has made in developing the latest Power technology,” adding, “The Power 9 processor, OpenCAPI bus, and high-capacity memory will provide opportunities for innovation in Google’s data centers.”

Bob Picciano, Senior Vice President of IBM Cognitive Systems, stated, “IBM has built a game-changing, innovative system for AI and cognitive workloads. The IBM Power9 system not only powers the world’s highest-performing supercomputers but also enables companies worldwide to gain superior insights, generating scientific discoveries and innovative business outcomes across all industries.”