Research company International Data Corp. forecasts that 75% of all enterprise apps will utilize AI by 2021.
In a move to better address the increasing market, IBM Corp. recently rolled out an artificial intelligence (AI)-based converged system dubbed Spectrum AI with Nvidia DGX.
The system joins the growing number of data center-based platforms that are custom-made for machine learning.
Back in September, Cisco Systems Inc. rolled out artificial intelligence (AI) server that features eight of Nvidia’s leading Tesla V100 graphics cards.
Earlier in 2018, both Pure Storage Inc. and NetApp Corp. rolled out their own artificial intelligence (AI)-based platforms, which are powered by Nvidia’s DGX-1 appliance.
As the name suggests, Spectrum AI with Nvidia DGX is based on a similar system.
Similar to Cisco’s server, the DGX-1 comprises of eight Tesla V100 graphics cards boasting over 45,000 processing cores between them.
Approximately 5,000 of those cores are specialized circuits dubbed Tensor Cores that are mainly tuned in a bid to run machine learning models appropriately.
According to Nvidia, this amount of silicon allows the DGX-1 to provide over a petaflop of computational potential.
IBM’s novel system integrates its Elastic Storage Server with the DGX-1.
The system is basically a storage platform that provides not only a petabyte of usable flash space but also runs IBM’s Spectrum Scale management software.
The addition of Spectrum Scale is recognizable for several reasons, aside from being utilized in Summit, which is the most powerful supercomputer in the world.
To add to its potential to accommodate more volumes of data, the software offers a broad selection of administrative capabilities.
Among these capabilities is a tool that allows information technology (IT) teams to offload some records in a given setting to the public cloud.
Another software integrated into Spectrum AI with Nvidia DGX is the RAPIDS framework, which Nvidia rolled out two months ago.
RAPIDS framework is a selection of libraries that can be installed with well-known analytics and AI tools in a bid to squeeze additional computing power out of IBM’s graphics cards.
According to IBM, the appliance has the potential to accommodate projects that vary in size.
Companies can install Spectrum AI with Nvidia DGX by utilizing and one Elastic Storage Server and some of DGX-1 appliances or deploy several racks’ worth of gear on the other end of the spectrum.
One rack featuring nine DGX-1 machines offers data throughput of a maximum of 120 gigabytes for each second, which IBM claimed exceeds all the other solutions that it has tested.