IBM has unveiled an AI-optimized converged system called Spectrum AI with Nvidia DGX. With the system, the company wants to meet the growing demand for artificial intelligence (AI) in enterprise applications. That’s what Silicon Angle reports.
The platform joins the growing number of data center platforms specifically created for machine learning. Cisco unveiled an AI server with eight Nvidia Tesla V100 GPUs in September. NetApp and Pure Storage also launched their own AI-optimised platforms with the Nvidia DGX-1 devices.
Spectrum AI with Nvidia DGX is based on the same system. As with the Cisco server, DGX-1 contains eight Tesla V100 GPUs with more than 45,000 processing cores. 5,000 of these are specialized circuits called Tensor Cores, which are made to run machine learning models properly.
This allows DGX-1 to offer more than one PETlop of computing power, Nvidia argues. A petaflop equals one quadrillion floating-point operations per second, which is a standard unit of computer performance. IBM’s new system combines the DGX-1 with its own Elastic Storage Server. It’s a storage platform that offers a petabyte of usable flash capacity and runs the Spectrum Scale management software.
The fact that Spectrum Scale is part of the solution is interesting for several reasons. This is how it is used in Summit, the most powerful supercomputer in the world. The software can not only store large amounts of data, but also offers a set of extensive administrative possibilities. One of these is a tool that allows IT teams to offload some of the records to the public cloud.
Another piece of software in Spectrum AI with Nvidia DGX is the RAPIDS framework that Nvidia launched two months ago. It is a collection of libraries that can be installed along with popular AI and analytics tools to get more computing power out of the GPUs.
The device can handle projects of different sizes, says IBM. Companies can set up a Spectrum AI with Nvidia DGX environment with a single Elastic Storage Server and a handful of DGX-1 appliances. At the other end of the spectrum it is also possible to use racks full of hardware.This news article was automatically translated from Dutch to give Techzine.eu a head start. All news articles after September 1, 2019 are written in native English and NOT translated. All our background stories are written in native English as well. For more information read our launch article.