Oracle brings Nvidia's Volta-based Tesla GPUs to Cloud Infrastructure

In addition to the new bare metal infrastructure offering, Oracle is introducing more deep learning and HPC tools, as well as new design and engineering apps.
Written by Stephanie Condon, Senior Writer

Video: Cryptocurrency mining raises GPU prices

About six months after releasing its first bare-metal GPU offering based on Nvidia's Pascal architecture, Oracle on Tuesday is announcing the general availability of bare metal Tesla GPUs based on the Volta architecture.

In addition to providing a new bare metal offering -- the first that has eight Volta-based GPUs -- Oracle Cloud Infrastructure is introducing more deep learning and HPC tools that exploit the Volta architecture, as well as new design and engineering apps.

Oracle expects these products will bring customers to the Oracle Cloud because of its competitive pricing but also because of Oracle's primary focus on the enterprise customer, Leo Leung, senior director of products & strategy for Oracle Cloud, told ZDNet.

"Everything we've built makes it very, very easy for enterprises to move existing applications as well as build new ones," he said. "That focus on being able to move existing things is something very different from other clouds and is very competitive to on-premise ways of operating."

The bare metal GPU offering has eight Tesla V100 GPUs, 768GB of memory, and up to 512TB of block storage at $2.25 an hour. It's now available in Oracle's US Ashburn Region, with a global expansion planned for the near future. This comes on top of previously released compute instances, available in both US and Europe regions, that provide up to two Pascal-based Tesla GPUs.

Leung pointed to one customer, Yellow Dog, to illustrate how bare metal GPUs can open up new business models for companies and various industries. As a broker for animation houses and entertainment studios, Yellow Dog effectively offers rendering-as-a-service. Its work is very project-oriented, compelling them to spin up, for example, 100 machines to finish one job. Previously, an animation company would have had to invest in its own servers to finish a cartoon -- a capital expenditure that wouldn't make sense for a business like Yellow Dog.

"We think this is a huge opportunity for customers that are used to running compute in the cloud as well as enterprises starting to do more and more in the cloud, whether for projects or entire applications," Leung said.

In addition to the bare metal offering, Oracle on Tuesday is announcing limited availability for the Nvidia GPU Cloud (NGC), with both Pascal and Volta-based compute offerings. NGC gives researchers and data scientists access to containerized deep learning frameworks and a range of GPU-optimized software tools for deep learning and HPC.

Lastly, Oracle is also launching limited availability of NVIDIA GRID for GPU-accelerated graphics on Oracle Cloud Infrastructure. GRID enables customers to use it as a powerful desktop to run different sets of applications. Oracle is partnering with Citrix so customers can run Citrix's XenApp and XenDesktop, as well as with Teradici to make Teradici Cloud Access Software available to try on Oracle's GPU instances.

Editorial standards