IBM Launches Converged System Tailored to Machine Learning Applications

Big Blue’s latest product launch will look to NVIDIA GPUs to solve big data challenges.

IBM’s latest AI solution leans heavily on NVIDIA’s DGX-1 hardware. (Image courtesy of

IBM’s latest AI solution leans heavily on NVIDIA’s DGX-1 hardware. (Image courtesy of

Citing research that suggests a large majority of all enterprise applications will take advantage of artificial intelligence by 2021, IBM has announced the launch of a new converged system optimized for AI. Dubbed “Spectrum AI,” the new platform is designed for the data center, with machine learning at the core of its functionality. IBM’s latest offering comes amid a string of AI-enhanced systems powered by NVIDIA’s DGX-1 suite to hit the market this year, with Cisco, NetApp and Pure Storage Inc. also debuting platforms.  

NVIDIA brings massive computing power

Powered by NVIDIA’s DGX-1, Spectrum AI will in many ways resemble the UCS C480 ML M5 server released by Cisco back in September that featured eight of NVIDIA’s Tesla V100 graphics cards. The eight cards represent over 45,000 processing cores, 5,000 of which are “Tensor Cores” which are optimized for AI. The upshot of those statistics is that IBM’s new DGX-1-based system will bring over a petaflop of processing power. In computing terms, this means the platform will be able to perform over a quadrillion floating point operations per second.

IBM delivers users more flexibility and capacity than ever before

IBM plans to tie an enterprise-friendly bow around this package by integrating the system with its Elastic Storage Server, which brings a petabyte of flash storage and can run its Spectrum Scale management software. Spectrum Scale is tailored to handle massive data volumes while simultaneously bringing useful administrative functions, like a built-in tool that allows organizations to migrate records to the public cloud at intervals. Notably, IBM’s Summit—the current title-holder for “world’s most powerful supercomputer”—is equipped with Spectrum Scale.

While that should put to rest any doubts as to the robustness of IBM’s new rollout, there’s also this tidbit—Spectrum AI also comes equipped with NVIDIA’s recently-introduced RAPIDS framework. Designed to accelerate the performance of a server’s GPUs, the collection of popular machine learning libraries optimizes deep learning applications by enabling even more iterative modeling. 

A range of use possibilities 

A graphical layout of Spectrum AI’s value model. (Image courtesy of IBM.)

A graphical layout of Spectrum AI’s value model. (Image courtesy of IBM.)

IBM bills Spectrum AI as a solution to projects of all sizes. Organizations can customize the platform to their needs by manipulating the volume of storage hardware and DGX-1 devices based on the size of data sets. As AI and machine learning increasingly drive business decisions at companies of all sizes, the ability to customize a solution’s computing power is a valuable tool in the arsenal of modern executives.