IBM Puts Nvidia Tesla K80 GPU on SoftLayer Cloud

IBM announced it is adding to the high-performance computing capabilities of the IBM Cloud by offering Nvidia's Tesla K80 GPU on SoftLayer.

IBM big data

IBM is boosting the high-performance computing (HPC) capabilities of the IBM Cloud by now offering Nvidia Tesla K80 dual-graphics processing unit (GPU) accelerators on bare metal cloud servers.

The new offering brings high-speed performance to the IBM SoftLayer infrastructure and enables companies using the IBM Cloud to build supercomputing clusters without having to expand their existing technology infrastructure.

“We’re offering the Nvidia K80 on bare metal servers here at SoftLayer,” said Jerry Gutierrez, global HPC sales leader for SoftLayer, in an interview with eWEEK. “It’s been through beta testing with some of our customers over the last several months. And they have all gone into production. What we’re addressing here is a need by the HPC industry as a whole to move to the cloud. We see a huge demand from those customers. They’re the last segment of customers that are able to move.”

Gutierrez said IBM Cloud is the only cloud infrastructure-as-a-service (IaaS) provider to offer GPU-accelerated computing on bare metal servers directly to customers. And it is now the only IaaS provider to offer its customers Tesla K80 GPU accelerators in the cloud, he said. By providing such capabilities in the cloud, IBM gives companies of all sizes easier and more affordable access to supercomputing resources.

“What has happened over the last couple of years is some HPC users have attempted to move to some of our competitors and been met with very lackluster performance, Gutierrez said. “By offering the K80 on bare metal, we’re able to offer a full solution for them to be able to deploy clusters for anything from machine learning, to analytics and be used across all industries. We’re seeing it in oil and gas, we’re seeing it in media and entertainment; we’re also seeing it in medical uses as well.”

In addition to bringing SoftLayer supercomputing capabilities to the enterprise, this new cloud GPU capability is particularly important for startups and research facilities, which typically start small, using only a few machines and GPU accelerators for testing and development workloads, Gutierrez said. With the Tesla K80, IBM Cloud provides a scalable supercomputing option that supports discovery and insight for customers in a variety of industries, including genomics, data analysis, machine learning and deep learning.

“This opens the door for a lot of these startups that come out of universities that have been using supercomputers or HPC at their universities to come out and hit the ground running with starting their business,” he said. “We’re seeing some really unique things that customers are doing – everything from image recognition to machine learning. This allows them to get started on a small scale and then grow into the business.”

Some of the early users of the Nvidia K80 on SoftLayer include New York University, MapD and Artomatix. NYU recently used Tesla K80 accelerators on IBM Cloud to support a deep-learning course.

“Students in our course were assigned to teams that used the K80 nodes on the SoftLayer infrastructure to complete deep-learning course work, assignments, and projects,” said Christian Puhrsch, former section leader at NYU, in a statement.

Startup MapD has been using Tesla K80 accelerators on IBM Cloud for data and analytics. The solution enables multiple users to query and visualize multi-billion row data sets with latencies measured in milliseconds, achieving orders-of-magnitude increases in speed over other solutions.

“While we offer on-premises appliance solutions, having a cloud GPU offering on SoftLayer is a perfect way to deploy to customers who would rather use MapD on a subscription basis or ‘try before they buy,’” said Todd Mostak, founder and CEO of MapD, in a statement. “The fact that IBM Cloud offers SoftLayer bare metal GPU servers dovetails perfectly with our focus on overwhelming performance, as virtualization can increase query latencies.”