Among the many areas that IBM is pushing forward with its Power architecture is in deep learning. Today IBM announced a new tool called PowerAI that aids in helping organizations learn things faster in a bid to generate advanced artificial intelligence capabilities.
“PowerAI contains the major machine, deep-learning open-source software that we have optimized and
compiled for our POWER8 with NVIDIA GPU servers,” Gupta told ServerWatch. “We are optimizing and getting the best performance out of the Power8 NVLink server and also from a cluster of these servers.”
The Power8 NVLink capabilities were first announced in September and provides improved I/O capabilities that come from NVIDA. According to IBM’s testing, the POWER8 with NVIDIA NVLink provides a 5x improvement in terms of bandwidth.
Gupta explained that IBM is aiming to make it easy for organizations to get up an running with PowerAI quickly with an easy installation approach. Gupta also noted that users can buy IBM’s reference configuration and focus on their data science / machine-learning tasks instead of trying to figure out their own configurations.
There is some overlap between what’s in PowerAI and what IBM has been doing with its Watson systems.
“PowerAI uses open-source software and creates optimized versions,” Gupta said. “IBM Watson also uses some of these deep-learning frameworks under the hood to train or teach Watson.”
While IBM’s Power architecture can run multiple operating systems, including Linux and AIX, Gupta said that PowerAI works only on Linux.
“PowerAI works on any POWER8 with NVIDIA GPU server, but is optimized for our NVLink server, the Power S822LC for High Performance Computing,” Gupta said.
Sean Michael Kerner is a senior editor at ServerWatch and InternetNews.com. Follow him on Twitter @TechJournalist