IBM, Nvidia target deep learning, AI workloads

IBM and Nvidia said they are collaborating on a deep learning tool that trains computers faster and scales better.

Specifically, IBM launched a software kit dubbed PowerAI that will run on its OpenPOWER LC servers for high performance computing.
The Power S822LC for high performance computing (HPC) uses Nvidia NVLink technology to work with IBM's architecture. IBM and Nvidia said that the S822LC system can run AlexNet with Caffe faster than x86 servers with multiple configurations.
Caffe is a deep learning framework created by Berkeley Vision and Learning Center (BVLC). Caffe is widely used, but IBM's PowerAI tookit supports another four deep learning software packages.
IBM and Nvidia see HPC as a growth market as more enterprises adopt deep learning for big data and analytics workloads. Both companies are members of the OpenPOWER Foundation, which aims to spread the POWER architecture in the data center.
Big Blue's POWER8 processor is integrated with Nvidia's Tesla P100 Pascal GPU accelerators in the S822LC systems. IBM said the new servers are being used to study the human brain, crunch data and supercomputing.

http://www.zdnet.com/article/ibm-nvidia-target-deep-learning-ai-workloads/